rustyfi_lang/lib.rs
1//! Abstract syntax tree, elaboration, evaluator, and primitives — the
2//! language core of the SATySFi port.
3
4pub mod ast;
5pub(crate) mod compile;
6pub mod crossref;
7pub mod elaborate;
8pub mod eval;
9pub mod exhaustive;
10pub mod hyphenation;
11pub mod prim_types;
12pub mod primitives;
13pub mod quoted;
14pub mod regexp;
15pub mod symbol;
16pub mod typecheck;
17pub mod types;
18pub mod unify;
19pub mod v1;
20pub mod value;
21pub mod visit;
22
23use crossref::{CrossRefs, Verdict};
24use rustyfi_backend::{
25 place_block_at, placed_line_extent, shift_graphics, DecoId, FontMetrics, GraphicsElem, Length,
26 PureHorzBox, VertBox,
27};
28use std::cell::RefCell;
29use std::collections::{BTreeMap, BTreeSet};
30use std::rc::Rc;
31use value::{DocumentValue, Value};
32
33#[derive(Debug, thiserror::Error)]
34pub enum CompileError {
35 #[error(transparent)]
36 Parse(#[from] rustyfi_syntax::ParseFileError),
37 #[error(transparent)]
38 Elaborate(#[from] elaborate::ElabError),
39 #[error(transparent)]
40 Type(#[from] typecheck::TypeError),
41 #[error(transparent)]
42 Eval(#[from] eval::EvalError),
43 #[error("the file's expression evaluated to {0}, not a document")]
44 NotADocument(&'static str),
45 #[error(transparent)]
46 Lower(#[from] v1::lower::LowerError),
47 /// A `V0_0` dependency spliced into a `V0_1` program referenced `name`, a
48 /// builtin primitive/type that is version-forked (bound, or shaped,
49 /// differently between `V0_0` and `V0_1` — see
50 /// `typecheck::forked_type_names`). The
51 /// merged program's single `base_env_with_version(V0_1)` can only bind
52 /// ONE closure per name, so accepting this would silently mis-resolve
53 /// `name` to the WRONG version's primitive.
54 ///
55 /// The trailing `— {}` is `v1::xver_adapt::forked_note`, keyed on
56 /// `name`: WHY this particular name cannot cross — a missing bridge
57 /// feature (a wrapper could be written), or a REPRESENTATION fork
58 /// (`page`, `font`) where the generations disagree about what the
59 /// runtime value IS and no amount of bridge work helps.
60 #[error(
61 "cross-version import ({slice}): dependency {dep} references `{name}`, a \
62 version-forked builtin — {}",
63 v1::xver_adapt::forked_note(.name)
64 )]
65 CrossVersionUnsupportedName {
66 name: String,
67 dep: String,
68 slice: &'static str,
69 },
70}
71
72/// Compile a `.saty` source string down to a typeset document:
73/// lex → parse → elaborate → evaluate.
74pub fn compile_document(
75 src: &str,
76 metrics: &dyn FontMetrics,
77) -> Result<std::rc::Rc<DocumentValue>, CompileError> {
78 let file = rustyfi_syntax::parse_file(src)?;
79 compile_document_cst(&file, metrics)
80}
81
82/// Compile an already-parsed (possibly loader-merged) file. The multi-file
83/// loader concatenates library preludes into one synthetic `cst::File` and
84/// enters here.
85///
86/// Thin wrapper over [`compile_document_cst_with_trials`] that drops the
87/// trial count — the stable entry point for the CLI (`main.rs`).
88pub fn compile_document_cst(
89 file: &rustyfi_syntax::cst::File,
90 metrics: &dyn FontMetrics,
91) -> Result<std::rc::Rc<DocumentValue>, CompileError> {
92 compile_document_cst_with_trials(file, metrics).map(|(doc, _trials)| doc)
93}
94
95/// Same as [`compile_document_cst`], but also returns how many fixpoint
96/// trials it took (& the fixpoint) — exposed for tests that must confirm the
97/// fixpoint actually iterated, not just that it produced the right answer on
98/// a lucky first pass.
99pub fn compile_document_cst_with_trials(
100 file: &rustyfi_syntax::cst::File,
101 metrics: &dyn FontMetrics,
102) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
103 compile_document_cst_with_aux(file, metrics, &mut crossref::AuxTable::new())
104}
105
106/// [`compile_document_cst_with_trials`] threading an AUXILIARY cross-reference table: `aux` seeds the
107/// fixpoint from a previous run and is overwritten with the final table.
108/// Seeding only affects how fast the fixpoint converges — see
109/// [`crossref::CrossRefs::seeded`] and [`crossref::CrossRefs::seed_unvalidated`],
110/// which together guarantee the output is the same as a cold run's.
111pub fn compile_document_cst_with_aux(
112 file: &rustyfi_syntax::cst::File,
113 metrics: &dyn FontMetrics,
114 aux: &mut crossref::AuxTable,
115) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
116 compile_document_cst_with_stages(file, metrics, aux, &std::collections::HashMap::new())
117}
118
119/// The stage a file's `@stage:` header declares, if any.
120///
121/// The loader merges every library's prelude into one file and drops the
122/// headers, so each caller that merges has to read this off first and record
123/// which entries it covers -- see [`compile_document_cst_with_stages`].
124pub fn declared_stage(file: &rustyfi_syntax::cst::File) -> Option<types::Stage> {
125 use rustyfi_syntax::token::Token;
126 file.headers.iter().find_map(|h| match h {
127 rustyfi_syntax::cst::Header::Stage(st) => match st.tok {
128 Token::HeaderPersistent0 => Some(types::Stage::Persistent0),
129 Token::HeaderStage0 => Some(types::Stage::Stage0),
130 Token::HeaderStage1 => Some(types::Stage::Stage1),
131 _ => None,
132 },
133 _ => None,
134 })
135}
136
137/// Record `file`'s declared stage against the prelude slots `start..end` its
138/// bindings just landed in, when that stage is not the default.
139fn note_stage(
140 stages: &mut std::collections::HashMap<usize, types::Stage>,
141 file: &rustyfi_syntax::cst::File,
142 start: usize,
143 end: usize,
144) {
145 if let Some(stage) = declared_stage(file).filter(|s| *s != types::Stage::default()) {
146 stages.extend((start..end).map(|i| (i, stage)));
147 }
148}
149
150/// Splice compiler-generated cross-version glue at the END of `prelude` and
151/// tag it with `stage`, the DECLARED stage of the dependency whose bindings
152/// the glue names.
153///
154/// The stage is load-bearing, not bookkeeping. `Stage::can_reference` is not
155/// symmetric: a `@stage: persistent` binding may not read a default-stage one.
156/// Every generated wrapper/shadow here re-applies a dependency's own export by
157/// name, so splicing it at the default stage silently makes it unreadable from
158/// the very consumers the forward deco/paren wrapper and its view-scheduling
159/// exist to serve — the failure surfaces as
160/// `invalid occurrence of variable .. as to stage`, which reads like a user
161/// error in a document that mentions neither binding. A FIXED stage would be
162/// as wrong for a `@stage: 0`/default dependency as the default is for a
163/// persistent one (cf. `unite_helper_prelude`'s explicit `Persistent0`).
164fn splice_staged(
165 prelude: &mut Vec<rustyfi_syntax::cst::TopBinding>,
166 stages: &mut std::collections::HashMap<usize, types::Stage>,
167 stage: Option<types::Stage>,
168 bindings: Vec<rustyfi_syntax::cst::TopBinding>,
169) {
170 let start = prelude.len();
171 prelude.extend(bindings);
172 if let Some(st) = stage {
173 stages.extend((start..prelude.len()).map(|i| (i, st)));
174 }
175}
176
177/// One [`v1::xver_adapt::deco_upgrade_prelude`] call per DISTINCT declared
178/// stage among `exports`, each spliced at that stage.
179///
180/// Grouping rather than one flat call is what keeps [`splice_staged`]'s
181/// argument true when a program crosses exports from dependencies that
182/// declared DIFFERENT `@stage:` headers: the glue for each is emitted in its
183/// own contiguous, correctly-tagged run. With one stage (every real program so
184/// far) it is exactly one call, in `exports` order.
185fn splice_upgrade_glue(
186 prelude: &mut Vec<rustyfi_syntax::cst::TopBinding>,
187 stages: &mut std::collections::HashMap<usize, types::Stage>,
188 exports: &[(v1::xver_adapt::DecoExport, Option<types::Stage>)],
189 step: v1::xver_adapt::UpgradeStep,
190) {
191 let mut order: Vec<Option<types::Stage>> = Vec::new();
192 for (_, st) in exports {
193 if !order.contains(st) {
194 order.push(*st);
195 }
196 }
197 for st in order {
198 let group: Vec<v1::xver_adapt::DecoExport> = exports
199 .iter()
200 .filter(|(_, s)| *s == st)
201 .map(|(e, _)| e.clone())
202 .collect();
203 splice_staged(
204 prelude,
205 stages,
206 st,
207 v1::xver_adapt::deco_upgrade_prelude(&group, step),
208 );
209 }
210}
211
212/// A phase stopwatch that only exists when `$RUSTYFI_TIMING` asked for one.
213///
214/// The point is the `None` case. `Instant::now()` has no implementation on
215/// `wasm32-unknown-unknown` — there is no clock to read, and the call PANICS —
216/// so taking one unconditionally meant the browser build could not compile a
217/// document at all, however little anyone wanted the timing. An untimed run
218/// has no use for the value either way, so not taking it costs nothing.
219#[derive(Clone, Copy)]
220struct Phase(Option<std::time::Instant>);
221
222impl Phase {
223 fn start(timing: bool) -> Self {
224 Phase(timing.then(std::time::Instant::now))
225 }
226
227 /// Milliseconds since [`Self::start`]. `0.0` when untimed — every caller
228 /// is inside an `if timing` that will not print it.
229 fn ms(self) -> f64 {
230 self.0
231 .map(|t| t.elapsed().as_secs_f64() * 1e3)
232 .unwrap_or(0.0)
233 }
234}
235
236/// [`compile_document_cst_with_aux`] told which merged prelude entries came
237/// from a file that declared a non-default `@stage:`.
238///
239/// The loader concatenates every library's prelude into one file, which loses
240/// the per-file header; this hands that back, so a `@stage: 0` library is
241/// typechecked at stage 0 (where `&e` is legal) while the document around it
242/// stays at stage 1 (where it is not).
243pub fn compile_document_cst_with_stages(
244 file: &rustyfi_syntax::cst::File,
245 metrics: &dyn FontMetrics,
246 aux: &mut crossref::AuxTable,
247 stages: &std::collections::HashMap<usize, types::Stage>,
248) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
249 let timing = std::env::var_os("RUSTYFI_TIMING").is_some();
250 let t = Phase::start(timing);
251 let env0 = primitives::base_env();
252 // The BRANDED front half lives in its own scope: the `SymbolStore`, the
253 // elaborated `Ast<Symbol>` and the typechecker's tables are all dead by
254 // the time the fixpoint trials run below.
255 //
256 // The DE-BRANDED `body` it yields, however, must stay alive until after
257 // `eval_document_trials` returns — `Interp::eval_arg` memoizes compiled
258 // command arguments by `&Ast` ADDRESS (`eval.rs`'s `arg_cache`), which is
259 // sound only while every node it can reach is pinned. Binding it to a
260 // local (rather than passing `&debrand(..)` as a temporary) is what pins
261 // it.
262 let body = {
263 let store = symbol::SymbolStore::new();
264 let scope = elaborate::Scope::new(&store, env0.names());
265 let program = elaborate::elaborate_program_with_stages(file, &scope, stages)?;
266 if timing {
267 eprintln!("TIMING elaborate {:>8.1}ms", t.ms());
268 }
269 let t = Phase::start(timing);
270 typecheck::typecheck(&program)?;
271 if timing {
272 eprintln!("TIMING typecheck {:>8.1}ms", t.ms());
273 }
274 // The compile membrane: resolve every `Symbol` back to its text, so
275 // nothing downstream (the `CompiledExpr`, the per-trial `Env`s,
276 // `Value`) carries the store's borrow. See `ast::debrand`.
277 ast::debrand(&program.body, &store)
278 };
279 // Compile the elaborated body into a closure tree ONCE. Each trial below
280 // re-runs this same `compiled` against a fresh env + a fresh (except
281 // `crossrefs`) `Interp` — safe because `CompiledExpr::run` takes `&self`
282 // and re-executes the whole tree from scratch, reproducing upstream's
283 // `eval_main i env_freezed ast` per trial (`main.ml:337-397`).
284 let t = Phase::start(timing);
285 let compiled = compile::compile_program(&body, &env0);
286 if timing {
287 eprintln!("TIMING compile-tree {:>8.1}ms", t.ms());
288 }
289 eval_document_trials(
290 &compiled,
291 metrics,
292 rustyfi_syntax::RustyfiVersion::V0_0,
293 aux,
294 )
295}
296
297/// [`compile_document_cst_with_stages`]'s front half and nothing else:
298/// elaborate and typecheck, then STOP — no closure tree, no font store, no
299/// evaluation. The 0.0.6 sibling of [`check_document_v1`]; see that function
300/// for what the distinction is for.
301pub fn check_document_cst_with_stages(
302 file: &rustyfi_syntax::cst::File,
303 stages: &std::collections::HashMap<usize, types::Stage>,
304) -> Result<(), CompileError> {
305 let env0 = primitives::base_env();
306 let store = symbol::SymbolStore::new();
307 let scope = elaborate::Scope::new(&store, env0.names());
308 let program = elaborate::elaborate_program_with_stages(file, &scope, stages)?;
309 typecheck::typecheck(&program)?;
310 Ok(())
311}
312
313/// Concatenate a loader-resolved 0.0.6 program's dependency-ordered library
314/// preludes ahead of the entry document's own, producing the one synthetic
315/// file [`compile_document_cst_with_stages`] and
316/// [`check_document_cst_with_stages`] elaborate — plus the per-slot `@stage:`
317/// table the concatenation would otherwise lose.
318///
319/// The 0.0.6 counterpart of `assemble_v1`, and the CLI's `merge_program`:
320/// both the compiler and the language server need it, so it lives here rather
321/// than in either of them.
322///
323/// Panics if handed a `LoadedCst::V0_1` file — a mixed-generation 0.0.6-rooted
324/// load belongs on [`compile_document_v006_xver_with_aux`]'s (or
325/// [`check_document_v006_xver`]'s) path, which the caller selects by testing
326/// for a `V0_1` file before calling this.
327pub fn merge_v006_program(
328 program: rustyfi_loader::LoadedProgram,
329) -> (
330 rustyfi_syntax::cst::File,
331 std::collections::HashMap<usize, types::Stage>,
332) {
333 fn as_v006(cst: rustyfi_loader::LoadedCst) -> rustyfi_syntax::cst::File {
334 match cst {
335 rustyfi_loader::LoadedCst::V0_0(f) => f,
336 rustyfi_loader::LoadedCst::V0_1(_) => unreachable!(
337 "merge_v006_program is the V0_0-only path; a V0_1 file belongs \
338 on compile_document_v1's or compile_document_v006_xver's"
339 ),
340 }
341 }
342
343 let mut files = program.files;
344 let entry = files.pop().expect("loader always yields the entry last");
345 let entry_cst = as_v006(entry.cst);
346 let mut prelude = Vec::new();
347 // Concatenation drops each file's headers, so `@stage:` — a property of
348 // its BINDINGS, not of the file as a document — is recorded here
349 // against the slots they land in. The entry document is stage 1 by
350 // definition and contributes nothing.
351 let mut stages = std::collections::HashMap::new();
352 for lib in files {
353 let mut cst = as_v006(lib.cst);
354 let start = prelude.len();
355 prelude.extend(std::mem::take(&mut cst.prelude));
356 note_stage(&mut stages, &cst, start, prelude.len());
357 }
358 prelude.extend(entry_cst.prelude);
359 (
360 rustyfi_syntax::cst::File {
361 headers: Vec::new(),
362 prelude,
363 in_kw: entry_cst.in_kw,
364 body: entry_cst.body,
365 eoi: entry_cst.eoi,
366 },
367 stages,
368 )
369}
370
371/// Typecheck a whole loader-resolved program, whichever of the three shapes it
372/// is, without evaluating it.
373///
374/// The dispatch mirrors the CLI's own (`rustyfi`'s `cmd_compile`) exactly, and
375/// for the same reasons: a `V0_1` load is a module program; a `V0_0` load that
376/// carries a foreign `V0_1` dependency is a cross-version splice; anything
377/// else is the flat 0.0.6 prelude merge. `version` is the load's own
378/// [`rustyfi_loader::LoadOptions::version`] — Axis A, the entry's generation —
379/// not any individual file's.
380pub fn check_document_program(
381 program: rustyfi_loader::LoadedProgram,
382 version: rustyfi_syntax::RustyfiVersion,
383) -> Result<(), CompileError> {
384 match version {
385 rustyfi_syntax::RustyfiVersion::V0_1 => check_document_v1(&program.files),
386 _ => {
387 let has_v01_dep = program
388 .files
389 .iter()
390 .any(|f| matches!(f.cst, rustyfi_loader::LoadedCst::V0_1(_)));
391 if has_v01_dep {
392 check_document_v006_xver(&program.files)
393 } else {
394 let (merged, stages) = merge_v006_program(program);
395 check_document_cst_with_stages(&merged, &stages)
396 }
397 }
398 }
399}
400
401/// The compile-once + fixpoint-trial tail shared by the `V0_0` and `V0_1`
402/// entry points (`compile_document_cst_with_trials` above and
403/// `compile_document_v1_with_trials` below). The only version-sensitive step
404/// is the fresh per-trial env (`primitives::base_env_with_version(version)`);
405/// everything else (crossrefs persistence, `fire_hooks`, `DocExtras` attach)
406/// is identical regardless of which SATySFi generation produced `compiled`.
407fn eval_document_trials(
408 compiled: &compile::CompiledExpr,
409 metrics: &dyn FontMetrics,
410 version: rustyfi_syntax::RustyfiVersion,
411 aux: &mut crossref::AuxTable,
412) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
413 // Seed the fixpoint from the previous run's auxiliary table, if any; if
414 // the final trial read a seeded value it never re-derived, redo cold
415 // instead (see `CrossRefs::seed_unvalidated`) — this is what keeps a warm
416 // build byte-identical to a cold one.
417 if !aux.is_empty() {
418 let (doc, trials, table, unvalidated) =
419 eval_trials_seeded(compiled, metrics, version, aux.clone())?;
420 if !unvalidated {
421 *aux = table;
422 return Ok((doc, trials));
423 }
424 }
425 let (doc, trials, table, _) =
426 eval_trials_seeded(compiled, metrics, version, crossref::AuxTable::new())?;
427 *aux = table;
428 Ok((doc, trials))
429}
430
431/// One complete fixpoint run against `seed`. Returns the final cross-reference
432/// table alongside the document, plus whether the seed turned out to be
433/// load-bearing but unverified ([`CrossRefs::seed_unvalidated`]).
434fn eval_trials_seeded(
435 compiled: &compile::CompiledExpr,
436 metrics: &dyn FontMetrics,
437 version: rustyfi_syntax::RustyfiVersion,
438 seed: crossref::AuxTable,
439) -> Result<(std::rc::Rc<DocumentValue>, u32, crossref::AuxTable, bool), CompileError> {
440 let timing = std::env::var_os("RUSTYFI_TIMING").is_some();
441 let crossrefs = Rc::new(RefCell::new(CrossRefs::seeded(seed)));
442 let mut trials = 0u32;
443 loop {
444 trials += 1;
445 let t_trial = Phase::start(timing);
446 // Fresh per trial: `let-mutable` store state resets (== upstream's
447 // `env_freezed` re-eval), and a fresh `Interp` resets `hooks`/
448 // `images` too — only `crossrefs` is threaded through.
449 //
450 // The runtime environment is just an empty root frame: the base
451 // environment is a COMPILE-time table already folded into the
452 // compiled tree, and top-level bindings live in the compiler's slot
453 // table, which the spine rewrites as it re-executes each trial.
454 // Nothing resolves a name here.
455 let env = value::Env::root();
456 let mut interp = eval::Interp::new(metrics);
457 interp.crossrefs = crossrefs.clone();
458 // Threads `version` onto the `Interp` so `read_inline`'s `EmbedMath`
459 // fallback arm (no installed math command — unit-test contexts
460 // only) can dispatch between `reflect_math_elem`/
461 // `reflect_math_elem_v01`.
462 interp.version = version;
463 let doc = match compiled.run(&env, &mut interp)? {
464 Value::Document(doc) => doc,
465 other => return Err(CompileError::NotADocument(other.type_name())),
466 };
467 let t_hooks = Phase::start(timing);
468 let run_ms = t_trial.ms();
469 // Fire every placed page-break hook now that `break_pages` has given
470 // every one of them its final page number/point; hooks mutate
471 // `crossrefs` (the only place that seam is legally crossed — see
472 // `fire_hooks`'s doc comment).
473 fire_hooks(&mut interp, &doc)?;
474 if timing {
475 eprintln!(
476 "TIMING trial {trials}: run(eval+layout) {:>8.1}ms fire_hooks {:>6.1}ms",
477 run_ms,
478 t_hooks.ms()
479 );
480 }
481 let verdict = crossrefs.borrow_mut().verdict();
482 match verdict {
483 Verdict::NeedsAnotherTrial => continue,
484 Verdict::CanTerminate(_) | Verdict::CountMax => {
485 // Attach the final trial's accumulated extras. `doc` is
486 // usually uniquely held here; if the program's env still
487 // holds a clone, fall back to a one-time deep clone.
488 let mut final_doc = Rc::try_unwrap(doc).unwrap_or_else(|rc| (*rc).clone());
489 final_doc.extras = rustyfi_backend::DocExtras {
490 annotations: std::mem::take(&mut interp.annotations),
491 destinations: std::mem::take(&mut interp.destinations),
492 outline: std::mem::take(&mut interp.outline),
493 page_graphics: std::mem::take(&mut interp.page_graphics),
494 doc_info: interp.doc_info.take(),
495 };
496 // The DecoId-keyed link/destination side-channel, same
497 // timing as `extras` above (only known once `fire_hooks`
498 // has run).
499 final_doc.reflow_links = std::mem::take(&mut interp.link_decos);
500 final_doc.reflow_dests = std::mem::take(&mut interp.dest_decos);
501 final_doc.reflow_frame_decos = std::mem::take(&mut interp.frame_decos);
502 let refs = crossrefs.borrow();
503 return Ok((
504 Rc::new(final_doc),
505 trials,
506 refs.export(),
507 refs.seed_unvalidated(),
508 ));
509 }
510 }
511 }
512}
513
514/// Compile a loader-resolved SATySFi 0.1 program (`LoadOptions { version:
515/// V0_1, .. }`): dependency libraries (`files[..n-1]`, loader
516/// dependency-first order) are each lowered to one `TopBinding::Module`
517/// (qualified exports — see `v1/lower.rs`'s module doc) via
518/// [`v1::lower::lower_file_v1`], the entry (`files[n-1]`, always last —
519/// `LoadedProgram::files`'s contract) via [`v1::lower::lower_document_v1`],
520/// assembled into ONE synthetic `cst::File` — the same shape the CLI's
521/// `merge_program` builds for 0.0.6 — and pushed through the SHARED
522/// elaborate -> typecheck(V0_1) -> compile -> fixpoint-eval pipeline.
523/// Signature ascriptions (`:>`) are enforced per binding by
524/// `v1::module_check::check_program`.
525pub fn compile_document_v1(
526 files: &[rustyfi_loader::LoadedFile],
527 metrics: &dyn FontMetrics,
528) -> Result<std::rc::Rc<DocumentValue>, CompileError> {
529 compile_document_v1_with_trials(files, metrics).map(|(doc, _trials)| doc)
530}
531
532/// Trial-count-reporting sibling, mirroring
533/// `compile_document_cst_with_trials` (same rationale: fixture tests that
534/// must see the fixpoint iterate).
535pub fn compile_document_v1_with_trials(
536 files: &[rustyfi_loader::LoadedFile],
537 metrics: &dyn FontMetrics,
538) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
539 compile_document_v1_with_aux(files, metrics, &mut crossref::AuxTable::new())
540}
541
542/// [`compile_document_v1_with_trials`] threading an AUXILIARY cross-reference
543/// table — see [`compile_document_cst_with_aux`]'s doc comment for what
544/// seeding `aux` does and why it can't change the output.
545pub fn compile_document_v1_with_aux(
546 files: &[rustyfi_loader::LoadedFile],
547 metrics: &dyn FontMetrics,
548 aux: &mut crossref::AuxTable,
549) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
550 use rustyfi_syntax::RustyfiVersion;
551
552 let asm = assemble_v1(files)?;
553 let env0 = primitives::base_env_with_version(RustyfiVersion::V0_1);
554 // Branded front half scoped so the store, the `Ast<Symbol>` tree and the
555 // module checker's tables are dead before the fixpoint trials run; the
556 // de-branded `body` stays pinned for the trials' sake — see
557 // `compile_document_cst_with_trials` for both halves of that contract.
558 let body = {
559 let store = symbol::SymbolStore::new();
560 let program = check_v1(&asm, &store, &env0)?;
561 ast::debrand(&program.body, &store)
562 };
563 let compiled = if asm.v006_indices.is_empty() {
564 compile::compile_program(&body, &env0)
565 } else {
566 let env0_v006 = primitives::base_env_with_version(RustyfiVersion::V0_0);
567 compile::compile_program_xver(&body, &env0, &env0_v006)
568 };
569 eval_document_trials(&compiled, metrics, RustyfiVersion::V0_1, aux)
570}
571
572/// [`compile_document_v1_with_aux`]'s front half and nothing else: assemble,
573/// elaborate, typecheck, enforce every `:>` seal — then STOP, without
574/// compiling a closure tree, without a font store, and without evaluating
575/// anything.
576///
577/// What it is for is a language server. Elaboration, typechecking and sealing
578/// are what take a document from "parses" to "would compile", and they are
579/// also the phases that need the whole resolved program rather than one
580/// buffer. What follows them — `compile::compile_program` and the fixpoint
581/// trials — produces pages, which an editor has no use for and which cost far
582/// more than the answer is worth on every keystroke.
583pub fn check_document_v1(files: &[rustyfi_loader::LoadedFile]) -> Result<(), CompileError> {
584 let asm = assemble_v1(files)?;
585 let env0 = primitives::base_env_with_version(rustyfi_syntax::RustyfiVersion::V0_1);
586 let store = symbol::SymbolStore::new();
587 check_v1(&asm, &store, &env0).map(|_program| ())
588}
589
590/// The synthetic single-file program a loader-resolved 0.1 load is compiled
591/// through, plus the three side tables the elaborator and the module checker
592/// need to read it correctly. Produced by [`assemble_v1`], consumed by
593/// [`check_v1`].
594struct AssembledV1<'a> {
595 /// The merged `cst::File` — every dependency's bindings, dependency-first,
596 /// then the entry's own body.
597 file: rustyfi_syntax::cst::File,
598 /// The 0.1 dependencies' own `cst_v1` trees, for `:>` seal enforcement.
599 dep_csts: Vec<&'a rustyfi_syntax::cst_v1::FileV1>,
600 /// Which top-level `file.prelude` slots a spliced 0.0.6 dependency
601 /// contributed (`Ast::VersionScope(V0_0, _)`).
602 v006_indices: std::collections::HashSet<usize>,
603 /// Which slots came from a file that declared a non-default `@stage:`.
604 stages: std::collections::HashMap<usize, types::Stage>,
605}
606
607/// Elaborate + typecheck + `:>`-check an [`AssembledV1`], returning the
608/// elaborated program (which the compile path then de-brands, and the check
609/// path drops).
610fn check_v1<'s>(
611 asm: &AssembledV1<'_>,
612 store: &'s symbol::SymbolStore,
613 env0: &value::BaseEnv,
614) -> Result<elaborate::Program<'s>, CompileError> {
615 use rustyfi_syntax::RustyfiVersion;
616
617 // A spliced `V0_0` dependency may name a `V0_0`-ONLY primitive
618 // (`text-in-math`, `get-axis-height`, `math-color`, …). Elaboration
619 // resolves names against ONE flat set built from the ambient version,
620 // and it runs BEFORE `Ast::VersionScope` can mean anything — the scope
621 // wraps an already-elaborated RHS — so such a name was simply
622 // "unbound variable" at elaborate time, no matter how correctly the
623 // later phases were version-scoped.
624 //
625 // So when (and ONLY when) a `V0_0` dependency was actually spliced,
626 // widen the elaboration name set to the UNION of both versions'
627 // primitives. This set answers one question — "is this a known global
628 // rather than a free variable?" — and version-correct resolution still
629 // happens downstream: `compile.rs`'s fold picks the `V0_0` `PrimDef`
630 // inside a `VersionScope(V0_0, _)`, and `typecheck.rs` picks that
631 // version's scheme. A pure `V0_1` program takes the other branch and
632 // keeps its "unbound variable" diagnostics for `V0_0`-only names.
633 let scope_names: Vec<String> = if asm.v006_indices.is_empty() {
634 env0.names()
635 } else {
636 let mut n = env0.names();
637 n.extend(primitives::base_env_with_version(RustyfiVersion::V0_0).names());
638 n.sort();
639 n.dedup();
640 n
641 };
642 let scope = elaborate::Scope::new_with_version(store, scope_names, RustyfiVersion::V0_1);
643 // `v006_indices` is empty whenever no `V0_0` dependency was
644 // spliced above, and `elaborate_program_with_versions` then emits no
645 // `Ast::VersionScope` node at all — so a `V0_1`-only load's
646 // `program`/`compiled` are structurally identical to a plain
647 // `elaborate_program`/`compile_program` pair's.
648 let program = elaborate::elaborate_program_with_versions(
649 &asm.file,
650 &scope,
651 &asm.v006_indices,
652 &asm.stages,
653 None,
654 )?;
655 v1::module_check::check_program(&asm.dep_csts, &program)?;
656 Ok(program)
657}
658
659/// Assemble a loader-resolved 0.1 program into the one synthetic `cst::File`
660/// the shared pipeline runs on — `merge_program`'s V0_1 analogue, plus the
661/// whole cross-version splice (X1/X3/X3b/X3c).
662fn assemble_v1<'a>(
663 files: &'a [rustyfi_loader::LoadedFile],
664) -> Result<AssembledV1<'a>, CompileError> {
665 use rustyfi_syntax::RustyfiVersion;
666
667 // -- assemble the synthetic cst::File (merge_program's V0_1 analogue) --
668 let (entry, deps) = files
669 .split_last()
670 .expect("loader always yields at least the entry file");
671 // Only ever called on the entry: under `compile_document_v1`, the entry
672 // is ALWAYS `V0_1` (the loader's own contract — `load_legacy`'s per-file
673 // version-detection rule
674 // only ever downgrades a DEPENDENCY to `V0_0`, never the entry; see
675 // `LoadedFile::version`'s doc comment). A `V0_0` dependency is instead
676 // routed through the cross-version splice arm below — it never
677 // reaches this helper.
678 fn as_v01(f: &rustyfi_loader::LoadedFile) -> &rustyfi_syntax::cst_v1::FileV1 {
679 match &f.cst {
680 rustyfi_loader::LoadedCst::V0_1(cst) => cst,
681 rustyfi_loader::LoadedCst::V0_0(_) => unreachable!(
682 "as_v01 called on a V0_0-parsed file — the entry is always \
683 V0_1 under compile_document_v1, and every V0_0 dependency \
684 is routed through the X1 cross-version splice arm instead"
685 ),
686 }
687 }
688 // One `SurfaceEnv` threaded across every V0_1 dependency
689 // in load order, so a module alias/named-signature reference in a
690 // LATER-loaded library can resolve an EARLIER one (`module M =
691 // OtherLib`, `:> OtherLib.S`). `build_file_surface` runs (pure `cst_v1`
692 // walk, no lowering needed) BEFORE each dep is lowered, so a dep's own
693 // internal aliases/named signatures resolve too.
694 //
695 // `deps` is a MIXED-version list (`LoadedFile::version`). A `V0_1`
696 // dep is lowered as usual; a `V0_0` dep contributes its `cst::File.prelude`
697 // bindings DIRECTLY (they are already `cst::TopBinding`s — no syntactic
698 // bridge needed), positioned dependency-first (loader order).
699 // `v006_indices` records which TOP-LEVEL `prelude` slots a V0_0 dep
700 // contributed, so `elaborate::elaborate_program_with_versions` (below) can
701 // wrap those bindings' RHS in `Ast::VersionScope(V0_0, _)` — the mechanism
702 // that makes a version-forked primitive referenced INSIDE such a dependency
703 // (`page-break`, `math-*`, …) resolve against `V0_0`'s
704 // `PrimDef`/type/runtime-version instead of the merged program's ambient
705 // `V0_1`. `dep_csts` collects the V0_1 subset only — `check_program`
706 // (below) has no `cst_v1` vocabulary for a `V0_0` file.
707 let mut surfaces = v1::surface::SurfaceEnv::default();
708 let mut prelude = Vec::new();
709 let mut dep_csts: Vec<&rustyfi_syntax::cst_v1::FileV1> = Vec::new();
710 let mut v006_indices: std::collections::HashSet<usize> = std::collections::HashSet::new();
711 // A spliced 0.0.6 dependency brings its `@stage:` with it, exactly as it
712 // would on the 0.0-rooted path -- a `@stage: 0` library must be readable
713 // from a 0.1 document too, or the same library compiles from one
714 // generation and not the other.
715 let mut stages: std::collections::HashMap<usize, types::Stage> =
716 std::collections::HashMap::new();
717 // Placement state for the forward view-scheduling mechanism: every
718 // `deco`/`deco-set`/`paren` export the forward deco/paren wrapper has
719 // adapted so far, and whether the 0.0.6-shaped (UNWRAPPED) view of them is
720 // the one currently installed at this point in the merged prelude.
721 //
722 // The forward wrapper installs the 0.1-shaped view by shadowing the
723 // export's own name,
724 // and a shadow is permanent — the prelude is one flat `Ast::LetIn`
725 // chain, and `Ast::VersionScope(V0_0, _)` wraps a binding's RHS, not the
726 // continuation after it. So the view is SCHEDULED (as the reverse deco
727 // coercion schedules the
728 // reverse one): captured once under a private name while the wrapped
729 // view is in force, the ORIGINAL restored on entering a 0.0.6-authored
730 // block, the wrapped view re-installed on entering a 0.1-authored one
731 // (the entry always is, and is always last). Without this, a LATER
732 // 0.0.6-authored dependency reads the export at 0.1's shape:
733 // `math.satyh`'s `val paren-right : paren` against `latexcmds`'
734 // five-argument call, and the `graphics list`/`graphics` mismatch for
735 // `deco`.
736 //
737 // Both transitions are lazy, so a program whose 0.0.6 dependencies
738 // never consume each other's crossed exports emits NOTHING extra.
739 // `deco_view_captured` counts how many of `deco_exports` already have a
740 // private `Capture` of their WRAPPED view in the prelude — an `Install`
741 // may only name a view that has been captured.
742 //
743 // Each export is carried with the DECLARED STAGE of the dependency that
744 // exported it, because the glue NAMES that dependency's own binding and
745 // `Stage::can_reference` is not symmetric: a `@stage: persistent`
746 // dependency may not read a default-stage binding, so a `Restore` of
747 // `M.frame` spliced at the default stage would make the very consumer
748 // this mechanism exists for fail with a STAGE error instead of a type one.
749 let mut deco_exports: Vec<(v1::xver_adapt::DecoExport, Option<types::Stage>)> = Vec::new();
750 let mut v006_view_installed = false;
751 let mut deco_view_captured: usize = 0;
752 for dep in deps {
753 match &dep.cst {
754 rustyfi_loader::LoadedCst::V0_1(cst) => {
755 // Transition back INTO 0.1-authored code: this dependency
756 // reads any crossed export at the adapted 0.1 shape, which is
757 // what the forward deco/paren wrapper is for.
758 if v006_view_installed {
759 splice_upgrade_glue(
760 &mut prelude,
761 &mut stages,
762 &deco_exports[..deco_view_captured],
763 v1::xver_adapt::UpgradeStep::Install,
764 );
765 v006_view_installed = false;
766 }
767 v1::surface::build_file_surface(cst, &mut surfaces);
768 prelude.extend(v1::lower::lower_file_v1_with_surfaces(cst, &surfaces)?);
769 dep_csts.push(cst);
770 }
771 rustyfi_loader::LoadedCst::V0_0(cst) => {
772 // Transition INTO 0.0.6-authored code: this
773 // dependency means 0.0.6's shape by every name it writes, so
774 // any export the forward wrapper has already adapted must
775 // read as its
776 // UNWRAPPED original here. Deliberately emitted BEFORE `start`
777 // is taken below, so this 0.1-authored glue never lands in
778 // `v006_indices`/`stages`.
779 //
780 // The `Capture` rides along with the FIRST such transition
781 // rather than being emitted per-dependency: this is the last
782 // position at which naming the export's own key still yields
783 // the forward wrapper's wrapped view, and emitting it lazily is what keeps a
784 // program with no 0.0.6-to-0.0.6 consumption byte-identical to
785 // the splice as it stood before this view-scheduling was added.
786 if !v006_view_installed || deco_view_captured < deco_exports.len() {
787 if deco_view_captured < deco_exports.len() {
788 splice_upgrade_glue(
789 &mut prelude,
790 &mut stages,
791 &deco_exports[deco_view_captured..],
792 v1::xver_adapt::UpgradeStep::Capture,
793 );
794 deco_view_captured = deco_exports.len();
795 }
796 splice_upgrade_glue(
797 &mut prelude,
798 &mut stages,
799 &deco_exports,
800 v1::xver_adapt::UpgradeStep::Restore,
801 );
802 v006_view_installed = !deco_exports.is_empty();
803 }
804 // `collect_free_globals` below only checks `free.types`
805 // against `forked_type_names` (see the "guard-narrowing"
806 // banner comment for which surface sites count and why).
807 // Residual gap (no per-name PolyType table): an UNANNOTATED
808 // top-level binding whose inferred type carries a forked
809 // shape has no syntactic site to catch. Genuine misuse still
810 // fails whole-program HM unification at the use site.
811 //
812 // `math` is representationally IDENTICAL to `V0_1`'s
813 // `math-text` (both `Base(MathText)`, `types.rs`; the same
814 // shared `Value::MathText`/`Value::Math` runtime rep,
815 // value.rs:39-56), so it RELABELS with zero value-level
816 // coercion. `reject_type_names()` is `forked_type_names()`
817 // PLUS `page`, whose bare name lowers identically under both
818 // versions (never appearing in the automatic diff) but whose
819 // runtime rep forks (9-ctor ADT vs a tuple), so it is
820 // rejected explicitly.
821 //
822 // `deco`/`deco-set` cross too
823 // (`classify_deco_exports`/`deco_coercion_prelude`). Their
824 // bare NAME already means the right thing
825 // (`typecheck::name_to_mono("deco", V0_1)` is
826 // `t_deco(V0_1)` unconditionally), so a textual mention with
827 // no attached VALUE (a `type .. = deco` synonym) is already
828 // safe. The VALUE needs adapting: a `V0_0` `deco` closure
829 // returns `graphics list`
830 // (`prim_types::t_graphics_output`/`coerce_graphics_result`)
831 // but every `V0_1` call site applying a `deco`
832 // (`primitives::apply_deco`) expects a SINGLE `graphics`.
833 // For a bare top-level `let-rec name : deco | patbot* = ..`/
834 // `: deco-set` export, splice a SECOND, un-scoped binding of
835 // the SAME name shadowing the original: it re-applies the
836 // still-unshadowed original positionally and unites its
837 // `graphics list` into one `graphics` via the real `V0_1`
838 // `unite-graphics` (`primitives::prim_unite_graphics`).
839 // HM-checked, so a wrapper that doesn't fit fails to
840 // typecheck rather than mis-rendering. Every OTHER forked
841 // name stays rejected.
842 //
843 // `reject_type_names_from_v006`, not the shared
844 // `reject_type_names`: this dependency's text is
845 // 0.0.6-AUTHORED, and `code` forks only in that reading
846 // (0.0.6 has no `code` spelling, so `τ code` is an opaque
847 // nominal there, while the merged program's hard-coded
848 // `V0_1` `Checker` reads the same text as the real staged
849 // type). The reverse arm keeps the shared set — a foreign
850 // 0.1 dependency's `code` is already in the ambient
851 // vocabulary.
852 let free = collect_free_globals(&cst.prelude);
853 let reject_t = v1::xver_adapt::reject_type_names_from_v006();
854 let touched: std::collections::BTreeSet<String> =
855 free.types.intersection(&reject_t).cloned().collect();
856 // Anything outside the combined whitelist above (`math`,
857 // `deco`, `deco-set`) rejects the WHOLE dependency — no
858 // partial acceptance.
859 if let Some(name) = touched
860 .iter()
861 .find(|n| !matches!(n.as_str(), "math" | "deco" | "deco-set" | "paren"))
862 {
863 return Err(CompileError::CrossVersionUnsupportedName {
864 name: name.clone(),
865 dep: dep.path.display().to_string(),
866 slice: "X3",
867 });
868 }
869 // A module-scoped deco wrapper lives INSIDE the spliced
870 // dependency, hence inside its `VersionScope(V0_0, _)`,
871 // where `unite-graphics` does not exist. Bind the V0_1
872 // primitive to a plain name FIRST, outside the range
873 // `v006_indices` is about to cover, so the scoped wrapper can
874 // reach it as an ordinary variable.
875 if touched.contains("deco")
876 || touched.contains("deco-set")
877 || touched.contains("paren")
878 {
879 let probe = v1::xver_adapt::classify_deco_exports(
880 &cst.prelude,
881 RustyfiVersion::V0_0,
882 RustyfiVersion::V0_1,
883 );
884 if probe
885 .as_ref()
886 .map(|e| v1::xver_adapt::needs_unite_helper(e))
887 == Ok(true)
888 {
889 let helper_start = prelude.len();
890 prelude.extend(v1::xver_adapt::unite_helper_prelude());
891 // Persistent, so the wrapper that calls it can name it
892 // from whatever stage the DEPENDENCY declared: these
893 // helpers are compiler-generated machinery spliced
894 // outside the dependency's own `@stage:` range, and a
895 // `@stage: persistent` dependency may not name a
896 // stage-1 binding (`Stage::can_reference`). Persistent
897 // is the one stage every other stage may reach, which
898 // is exactly the property a generated helper needs.
899 stages.extend(
900 (helper_start..prelude.len())
901 .map(|i| (i, types::Stage::Persistent0)),
902 );
903 }
904 }
905 let start = prelude.len();
906 if touched.is_empty() {
907 // No forked-type-name text anywhere in this dep —
908 // splice verbatim (the GOLDEN/non-regression fast path).
909 prelude.extend(cst.prelude.iter().cloned());
910 } else if touched.contains("math") {
911 // Relabel every `math` leaf inside a `type` declaration's
912 // body to `math-text` (the note above) and splice the
913 // adapted prelude. `deco`/`deco-set`, if also touched,
914 // need no textual relabel — so this one call covers the
915 // whole prelude regardless of which combination of the
916 // two is touched.
917 let adapted = v1::xver_adapt::relabel_type_decls(
918 &cst.prelude,
919 RustyfiVersion::V0_0,
920 RustyfiVersion::V0_1,
921 )
922 .map_err(|be| {
923 CompileError::CrossVersionUnsupportedName {
924 name: match &be {
925 v1::xver_adapt::BoundaryError::ForkedTypeExport {
926 ty_name, ..
927 } => ty_name.clone(),
928 },
929 dep: dep.path.display().to_string(),
930 slice: "X3",
931 }
932 })?;
933 prelude.extend(adapted);
934 } else {
935 // Only `deco`/`deco-set` (no `math`) is touched — no
936 // textual relabel needed, splice verbatim (the value-
937 // level coercion, if any, is appended separately below).
938 prelude.extend(cst.prelude.iter().cloned());
939 }
940 v006_indices.extend(start..prelude.len());
941 note_stage(&mut stages, cst, start, prelude.len());
942
943 if touched.contains("deco")
944 || touched.contains("deco-set")
945 || touched.contains("paren")
946 {
947 let exports = v1::xver_adapt::classify_deco_exports(
948 &cst.prelude,
949 RustyfiVersion::V0_0,
950 RustyfiVersion::V0_1,
951 )
952 .map_err(|be| {
953 CompileError::CrossVersionUnsupportedName {
954 name: match &be {
955 v1::xver_adapt::BoundaryError::ForkedTypeExport {
956 ty_name, ..
957 } => ty_name.clone(),
958 },
959 dep: dep.path.display().to_string(),
960 slice: "X3b",
961 }
962 })?;
963 // Deliberately NOT added to `v006_indices` (structural
964 // honesty, not a soundness requirement): this synthetic
965 // code is genuinely `V0_1`-authored (it calls
966 // `unite-graphics`, a `V0_1`-only primitive) — no `V0_0`
967 // `PrimDef` shares this name, so even inside a
968 // `VersionScope(V0_0, _)` the fold cursor would miss and
969 // `compile.rs` would fall back to the eval-time
970 // `env.lookup` against the ambient `V0_1` runtime env.
971 // Two halves: a TOP-LEVEL export is shadowed by a new
972 // top-level binding appended after the dependency
973 // (`deco_coercion_prelude`); a MODULE-scoped one cannot
974 // be (`let Deco.simple-frame` is not syntax), so its
975 // wrapper is appended inside that module's own `decls`
976 // (`inject_module_deco_wrappers`), one scope deeper.
977 v1::xver_adapt::inject_module_deco_wrappers(&mut prelude[start..], &exports);
978 // At the DEPENDENCY's own stage, not the default one: this
979 // top-level wrapper re-applies the dependency's export by
980 // name, and a `@stage: persistent` dependency's binding is
981 // unreadable from a default-stage one (`splice_staged`).
982 // The in-module wrappers above need no such care — they
983 // are spliced INSIDE `prelude[start..]`, already covered by
984 // this dependency's own `note_stage` range.
985 let dep_stage =
986 declared_stage(cst).filter(|s| *s != types::Stage::default());
987 splice_staged(
988 &mut prelude,
989 &mut stages,
990 dep_stage,
991 v1::xver_adapt::deco_coercion_prelude(&exports),
992 );
993 // From here on this export has TWO views in the
994 // program — the wrapper just spliced, and the unwrapped
995 // original the two injectors above kept reachable under
996 // `xver-fwd-orig-`. Record it so the transitions can pick
997 // the right one for whatever block comes next.
998 deco_exports.extend(exports.into_iter().map(|e| (e, dep_stage)));
999 }
1000 }
1001 }
1002 }
1003 // The last (and, for every single-generation dependency set, the ONLY)
1004 // transition back into 0.1-authored code: the entry itself, which is
1005 // always `V0_1` here and reads every crossed export at the forward
1006 // wrapper's adapted
1007 // shape. Emitted only if some intervening 0.0.6 dependency restored the
1008 // originals; with no such dependency the whole schedule stays silent.
1009 if v006_view_installed {
1010 splice_upgrade_glue(
1011 &mut prelude,
1012 &mut stages,
1013 &deco_exports[..deco_view_captured],
1014 v1::xver_adapt::UpgradeStep::Install,
1015 );
1016 }
1017 let entry_cst = as_v01(entry);
1018 let body = v1::lower::lower_document_v1(entry_cst)?;
1019 let eoi = match entry_cst {
1020 rustyfi_syntax::cst_v1::FileV1::Document { eoi, .. } => eoi.clone(),
1021 _ => unreachable!("lower_document_v1 already rejected a Library entry"),
1022 };
1023 let file = rustyfi_syntax::cst::File {
1024 headers: Vec::new(),
1025 prelude,
1026 in_kw: Some(rustyfi_syntax::leaf::KwIn(rustyfi_syntax::Span::default())),
1027 body: Some(body),
1028 eoi,
1029 };
1030
1031 Ok(AssembledV1 {
1032 file,
1033 dep_csts,
1034 v006_indices,
1035 stages,
1036 })
1037}
1038
1039/// Compile a loader-resolved SATySFi 0.0.6 program (`LoadOptions { version:
1040/// V0_0, .. }`) whose entry (or one of its native 0.0.6 co-dependencies)
1041/// `@require:`s at least one **foreign 0.1** package.
1042///
1043/// This is the REVERSE of [`compile_document_v1_with_trials`]'s direction, but
1044/// reuses its exact polarity rather than flipping it: the AMBIENT
1045/// elaborate/typecheck/compile tag stays `V0_1` (0.1's grammar is a strict
1046/// syntactic superset of 0.0.6's, so elaborating genuinely 0.0.6-authored code
1047/// under an ambient `V0_1` scope never rejects it), and it is the
1048/// 0.0.6-authored code — the ENTRY's own top-level bindings and document tail,
1049/// plus any native 0.0.6 co-dependency's bindings — that gets wrapped in
1050/// [`ast::Ast::VersionScope`]`(V0_0, _)`. A foreign 0.1 dependency splices in
1051/// UNWRAPPED, exactly like a native 0.1 dependency does in
1052/// `compile_document_v1_with_trials`; its own `:>`-sealed exports are enforced
1053/// by `v1::module_check::check_program` exactly as for a pure-0.1 consumer.
1054///
1055/// A pure-0.0.6 load (no 0.1 dependency) never reaches this function — the
1056/// CLI/loader only route here once a `V0_0`-rooted load's dependency graph
1057/// actually contains a `LoadedCst::V0_1` node.
1058pub fn compile_document_v006_xver(
1059 files: &[rustyfi_loader::LoadedFile],
1060 metrics: &dyn FontMetrics,
1061) -> Result<std::rc::Rc<DocumentValue>, CompileError> {
1062 compile_document_v006_xver_with_trials(files, metrics).map(|(doc, _trials)| doc)
1063}
1064
1065/// Trial-count-reporting sibling, mirroring `compile_document_v1_with_trials`.
1066pub fn compile_document_v006_xver_with_trials(
1067 files: &[rustyfi_loader::LoadedFile],
1068 metrics: &dyn FontMetrics,
1069) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
1070 compile_document_v006_xver_with_aux(files, metrics, &mut crossref::AuxTable::new())
1071}
1072
1073/// [`compile_document_v006_xver_with_trials`] threading an AUXILIARY
1074/// cross-reference table — see [`compile_document_cst_with_aux`]'s doc
1075/// comment for what seeding `aux` does and why it can't change the output.
1076pub fn compile_document_v006_xver_with_aux(
1077 files: &[rustyfi_loader::LoadedFile],
1078 metrics: &dyn FontMetrics,
1079 aux: &mut crossref::AuxTable,
1080) -> Result<(std::rc::Rc<DocumentValue>, u32), CompileError> {
1081 use rustyfi_syntax::RustyfiVersion;
1082
1083 let asm = assemble_v006_xver(files)?;
1084 let env0 = primitives::base_env_with_version(RustyfiVersion::V0_1);
1085 let store = symbol::SymbolStore::new();
1086 let program = check_v006_xver(&asm, &store, &env0)?;
1087 let env0_v006 = primitives::base_env_with_version(RustyfiVersion::V0_0);
1088 // `v006_indices` is NEVER empty here (the entry's own bindings are
1089 // always indexed into it above), so this always takes the `_xver` fold
1090 // path — matching `compile_document_v1_with_trials`'s own `if v006_
1091 // indices.is_empty() { .. } else { compile_program_xver }` branch,
1092 // specialized since the `else` arm is the only reachable one.
1093 // Bound to a local, not passed as a temporary: `Interp::eval_arg`
1094 // memoizes by `&Ast` address, so the de-branded tree must outlive the
1095 // trials (see `compile_document_cst_with_trials`).
1096 let body = ast::debrand(&program.body, &store);
1097 let compiled = compile::compile_program_xver(&body, &env0, &env0_v006);
1098 eval_document_trials(&compiled, metrics, RustyfiVersion::V0_0, aux)
1099}
1100
1101/// [`compile_document_v006_xver_with_aux`]'s front half and nothing else —
1102/// the reverse-direction sibling of [`check_document_v1`], with the same
1103/// rationale.
1104pub fn check_document_v006_xver(files: &[rustyfi_loader::LoadedFile]) -> Result<(), CompileError> {
1105 let asm = assemble_v006_xver(files)?;
1106 let env0 = primitives::base_env_with_version(rustyfi_syntax::RustyfiVersion::V0_1);
1107 let store = symbol::SymbolStore::new();
1108 check_v006_xver(&asm, &store, &env0).map(|_program| ())
1109}
1110
1111/// [`AssembledV1`]'s reverse-direction counterpart: additionally carries the
1112/// qualified member keys the reverse deco coercion rebound, which the `:>`
1113/// seal check must exempt from a second conformance test.
1114struct AssembledXver<'a> {
1115 file: rustyfi_syntax::cst::File,
1116 dep_csts: Vec<&'a rustyfi_syntax::cst_v1::FileV1>,
1117 v006_indices: std::collections::HashSet<usize>,
1118 stages: std::collections::HashMap<usize, types::Stage>,
1119 xver_shadows: std::collections::HashSet<String>,
1120}
1121
1122/// Elaborate + typecheck + `:>`-check an [`AssembledXver`] — [`check_v1`]'s
1123/// reverse-direction counterpart.
1124fn check_v006_xver<'s>(
1125 asm: &AssembledXver<'_>,
1126 store: &'s symbol::SymbolStore,
1127 env0: &value::BaseEnv,
1128) -> Result<elaborate::Program<'s>, CompileError> {
1129 use rustyfi_syntax::RustyfiVersion;
1130
1131 let scope = elaborate::Scope::new_with_version(store, env0.names(), RustyfiVersion::V0_1);
1132 // `wrap_body_version = Some(V0_0)`: the ENTRY's own document tail
1133 // (`file.body`, always 0.0.6-authored here) is wrapped in
1134 // `Ast::VersionScope(V0_0, _)` too — the one new elaborate.rs
1135 // capability this reverse direction adds beyond wrapping dependency bindings.
1136 let program = elaborate::elaborate_program_with_versions(
1137 &asm.file,
1138 &scope,
1139 &asm.v006_indices,
1140 &asm.stages,
1141 Some(RustyfiVersion::V0_0),
1142 )?;
1143 // `dep_csts` here is the foreign 0.1 dependencies' OWN `cst_v1` trees, so
1144 // a `:>`-sealed export (e.g. `V01Sealed.t`) is enforced against the WHOLE
1145 // merged spine exactly as it would be for a pure-0.1 consumer.
1146 //
1147 // `xver_shadows` is the ONE thing this arm asks the checker
1148 // to treat differently, and only for names it has itself just rebound:
1149 // the exporting module's own alias is still conformance-checked, and
1150 // only the coercion shadow that FOLLOWS it is exempted from a second
1151 // check against a signature it deliberately does not match. Empty
1152 // whenever no 0.1 `deco` export crossed.
1153 v1::module_check::check_program_with_xver_shadows(
1154 &asm.dep_csts,
1155 &program,
1156 &asm.xver_shadows,
1157 )?;
1158 Ok(program)
1159}
1160
1161/// Assemble a loader-resolved 0.0.6-rooted program that carries at least one
1162/// foreign 0.1 dependency — [`assemble_v1`]'s reverse-direction counterpart.
1163fn assemble_v006_xver<'a>(
1164 files: &'a [rustyfi_loader::LoadedFile],
1165) -> Result<AssembledXver<'a>, CompileError> {
1166 // The entry is whichever file is a document (`LoadedCst::is_document`) —
1167 // NOT necessarily `files.last()` (that assumption is specific to
1168 // `compile_document_v1_with_trials`'s pure-V0_1-entry contract); scan
1169 // defensively.
1170 let (entry_idx, entry) = files
1171 .iter()
1172 .enumerate()
1173 .find(|(_, f)| f.cst.is_document())
1174 .expect("loader validated exactly one document (the entry)");
1175 let entry_cst = match &entry.cst {
1176 rustyfi_loader::LoadedCst::V0_0(f) => f,
1177 rustyfi_loader::LoadedCst::V0_1(_) => unreachable!(
1178 "compile_document_v006_xver is the V0_0-entry sibling of \
1179 compile_document_v1 — a V0_1 entry belongs there instead"
1180 ),
1181 };
1182
1183 let mut surfaces = v1::surface::SurfaceEnv::default();
1184 let mut prelude = Vec::new();
1185 let mut dep_csts: Vec<&rustyfi_syntax::cst_v1::FileV1> = Vec::new();
1186 let mut v006_indices: std::collections::HashSet<usize> = std::collections::HashSet::new();
1187 // A spliced 0.0.6 dependency brings its `@stage:` with it, exactly as it
1188 // would on the 0.0-rooted path -- a `@stage: 0` library must be readable
1189 // from a 0.1 document too, or the same library compiles from one
1190 // generation and not the other.
1191 let mut stages: std::collections::HashMap<usize, types::Stage> =
1192 std::collections::HashMap::new();
1193 // The qualified member keys (`"M.frame"`) this arm rebinds to
1194 // a version-adapted view, exempted from a SECOND `:>` seal check below.
1195 let mut xver_shadows: std::collections::HashSet<String> = std::collections::HashSet::new();
1196 // Placement state for the reverse deco coercion: every `deco`/`deco-set` export crossed so far, and
1197 // whether the 0.0.6-shaped VIEW of them is the one currently installed at
1198 // this point in the merged prelude.
1199 //
1200 // The prelude is one flat `Ast::LetIn` chain and `Ast::VersionScope(V0_0,
1201 // _)` wraps a binding's RHS, not the continuation after it, so a
1202 // rebinding of `M.frame` is visible to EVERYTHING that follows
1203 // regardless of which generation authored it. A position-indexed view
1204 // is sufficient because each block spliced below is homogeneous — a
1205 // `V0_0` dependency's whole `prelude` goes into `v006_indices`, a `V0_1`
1206 // dependency's whole `lowered` stays out of it, the entry (always
1207 // 0.0.6-authored) is last — and the loader orders dependencies
1208 // topologically, so a consumer's block always follows what it
1209 // `@require:`s. So the coerced view installs lazily on entering a
1210 // 0.0.6-authored block and is put back on entering a 0.1-authored one.
1211 //
1212 // Both transitions are lazy, so the common case (every 0.1 dependency,
1213 // then the 0.0.6 entry — every bundled package) emits exactly one install
1214 // and no restore at all.
1215 let mut deco_exports: Vec<v1::xver_adapt::DecoExport> = Vec::new();
1216 let mut v006_view_installed = false;
1217
1218 for (i, dep) in files.iter().enumerate() {
1219 if i == entry_idx {
1220 continue;
1221 }
1222 match &dep.cst {
1223 // Native 0.0.6 co-dependency (e.g. the entry ALSO `@require:`s
1224 // an ordinary 0.0.6 package `list.satyg`-style): splice + wrap.
1225 // Its VALUE half is unrestricted — every binding here is
1226 // `Ast::VersionScope(V0_0, _)`-wrapped, so a forked primitive
1227 // resolves against 0.0.6's own `PrimDef`. Its TYPE-DECLARATION
1228 // half is NOT unrestricted: the reverse arm's 0.0.6 type-text guard
1229 // (`guard_v006_type_text`,
1230 // and the banner above it) refuses/relabels 0.0.6-authored
1231 // `type` text that a merged program's hard-coded-`V0_1` `Checker`
1232 // would otherwise re-read with the wrong vocabulary.
1233 rustyfi_loader::LoadedCst::V0_0(cst) => {
1234 let adapted = guard_v006_type_text(&cst.prelude, &dep.path)?;
1235 // Transition INTO 0.0.6-authored code: this dependency reads
1236 // any crossed `deco` export at 0.0.6's `graphics list` shape.
1237 // Deliberately BEFORE `start` is taken, so the generated glue
1238 // (0.1-authored) never lands in `v006_indices`/`stages`.
1239 if !v006_view_installed && !deco_exports.is_empty() {
1240 prelude.extend(v1::xver_adapt::deco_downgrade_prelude(
1241 &deco_exports,
1242 v1::xver_adapt::DowngradeStep::Install,
1243 ));
1244 v006_view_installed = true;
1245 }
1246 let start = prelude.len();
1247 prelude.extend(adapted);
1248 v006_indices.extend(start..prelude.len());
1249 note_stage(&mut stages, cst, start, prelude.len());
1250 }
1251 // Foreign 0.1 dependency: lower (exactly like a native V0_1 dep
1252 // in `compile_document_v1_with_trials`) and splice UNWRAPPED
1253 // (ambient V0_1), PLUS the reverse import guard on what it EXPORTS
1254 // (the forward arm's forked-name guard — narrowed to export
1255 // position, with the whitelist adaptation — reversed).
1256 //
1257 // `Checker.version` for TYPE DECLARATIONS
1258 // (`v1::module_check::check_program_inner`'s
1259 // `ck.declare_synonym`/`declare_variant`, and the
1260 // `base_type_env_with_version` seeding the phase-D spine walk;
1261 // `module_check.rs:238-239,271`) is HARD-CODED to
1262 // `RustyfiVersion::V0_1` on BOTH arms — every type declaration in
1263 // the merged program is read under V0_1 vocabulary
1264 // unconditionally. That is why the FORWARD arm's
1265 // `relabel_type_decls(dep.prelude, V0_0, V0_1)` (above) is
1266 // necessary: a 0.0.6 dependency's own "math" spelling must
1267 // become "math-text" before it reaches `program.type_decls`, or
1268 // the V0_1 lookup resolves it to an unrelated unbound nominal.
1269 //
1270 // The REVERSE consequence is NOT the naive mirror: a foreign 0.1
1271 // dependency's own "math-text"/"math-boxes" spelling is ALREADY
1272 // the ambient vocabulary, so no relabeling is needed or wanted —
1273 // renaming it to 0.0.6's "math" would corrupt text the
1274 // hard-coded-V0_1 `Checker` must read natively, turning a
1275 // working type into an unbound-nominal mismatch. So this arm
1276 // calls `collect_free_globals` purely as a WHITELIST GUARD: any
1277 // export-boundary forked type name outside `{"math-text",
1278 // "math-boxes"}` — a proven-identical-representation set
1279 // (shared `Value::MathText`/`Value::Math` runtime rep; 0.0.6
1280 // code has no syntax that could observe the lost distinction) —
1281 // rejects the WHOLE dependency. False-reject is safe,
1282 // false-accept is not. `page`/`graphics`/`deco`/`pre-path`/
1283 // `path`/`image`/`font`/`paren` all still reject in THIS
1284 // direction too. Past the whitelist, the dependency splices
1285 // VERBATIM.
1286 //
1287 // `deco`/`deco-set` CROSS in this direction too — the
1288 // reverse mirror of the forward wrapper's `unite-graphics` wrap, coercing the
1289 // OPPOSITE way. A crossing `V0_1` deco returns a single
1290 // `graphics`; every `V0_0`-authored consumer call site (and
1291 // every `V0_0`-scoped `inline-frame-outer`/`inline-frame-
1292 // breakable` TYPE) expects a `graphics list`, so the wrap is a
1293 // SINGLETON LIST, `[name p w h d]`. Three steps:
1294 //
1295 // 1. `classify_deco_exports_v01_sig` reads the dependency's
1296 // PRE-lowering `cst_v1` sig (the ONE textual site 0.1's
1297 // grammar can name a `deco` export's type at all — lowering
1298 // DROPS `sig_annot` entirely, so this scan must happen here
1299 // and not off `lowered`). It descends through nested
1300 // `module`/`include` decls and dereferences named signature
1301 // references against `surfaces`, which is exactly why
1302 // `build_file_surface` above must run FIRST. Anything it
1303 // still cannot express — a `paren`, a `deco` buried in a
1304 // compound, an OPEN optional row, or a `deco` behind a
1305 // functor signature member (whose members have no member
1306 // path until some later file APPLIES it) — REJECTS.
1307 // 2. `deco_downgrade_prelude` generates the coercion glue: a
1308 // private `Capture` of the 0.1 original immediately after
1309 // the dependency, then an `Install` — a top-level rebinding
1310 // of each export's own qualified key (`M.frame`) that
1311 // re-applies the captured original positionally and wraps
1312 // the result in a singleton list — deferred to the next
1313 // 0.0.6-authored block, and a `Restore` on the way back
1314 // into a 0.1-authored one (see `deco_exports`/
1315 // `v006_view_installed` above). None of it is added to
1316 // `v006_indices` — this is `V0_1`-authored glue, exactly
1317 // like the forward arm's `deco_coercion_prelude`.
1318 // 3. those qualified keys are collected into `xver_shadows`
1319 // and handed to `check_program_with_xver_shadows` below,
1320 // which exempts the SECOND `Ast::LetIn` of each from the
1321 // `:>` seal re-check (the module's own alias is still
1322 // checked; see that function's doc comment for why that
1323 // exemption cannot hide a real violation).
1324 //
1325 // A BARE `type foo = deco` synonym (no value attached, safe with
1326 // zero coercion — same reasoning as the forward direction's
1327 // `type xver-deco-alias = deco`) is UNAFFECTED: it is not a sig
1328 // `val` item, so this scan never sees it and it splices verbatim.
1329 rustyfi_loader::LoadedCst::V0_1(cst) => {
1330 // Transition back INTO 0.1-authored code: this dependency
1331 // reads any crossed `deco` export at 0.1's own single-
1332 // `graphics` shape, which is the whole point of the schedule.
1333 if v006_view_installed {
1334 prelude.extend(v1::xver_adapt::deco_downgrade_prelude(
1335 &deco_exports,
1336 v1::xver_adapt::DowngradeStep::Restore,
1337 ));
1338 v006_view_installed = false;
1339 }
1340 v1::surface::build_file_surface(cst, &mut surfaces);
1341 let lowered = v1::lower::lower_file_v1_with_surfaces(cst, &surfaces)?;
1342
1343 let free = collect_free_globals(&lowered);
1344 let reject_t = v1::xver_adapt::reject_type_names();
1345 let touched: BTreeSet<String> =
1346 free.types.intersection(&reject_t).cloned().collect();
1347 if let Some(name) = touched.iter().find(|n| {
1348 !matches!(n.as_str(), "math-text" | "math-boxes" | "deco" | "deco-set")
1349 }) {
1350 return Err(CompileError::CrossVersionUnsupportedName {
1351 name: name.clone(),
1352 dep: dep.path.display().to_string(),
1353 slice: "X4a",
1354 });
1355 }
1356 // `touched.contains("deco"/"deco-set")` here can only mean
1357 // the SAFE, no-coercion-needed case (a bare `type foo =
1358 // deco` synonym — no value attached; this arm's own doc
1359 // comment above): a REAL sig-declared VALUE export is
1360 // invisible to this POST-lowering scan (sig is dropped) and
1361 // is instead classified by the PRE-lowering scan just below,
1362 // independently of `touched`.
1363 let dep_deco_exports = v1::xver_adapt::classify_deco_exports_v01_sig(
1364 cst, &surfaces,
1365 )
1366 .map_err(|be| CompileError::CrossVersionUnsupportedName {
1367 name: match &be {
1368 v1::xver_adapt::BoundaryError::ForkedTypeExport { ty_name, .. } => {
1369 ty_name.clone()
1370 }
1371 },
1372 dep: dep.path.display().to_string(),
1373 slice: "X4b",
1374 })?;
1375
1376 prelude.extend(lowered);
1377 // The private capture goes here and only here: `M.frame` is
1378 // bound by `lowered` just above, and the 0.1 view is in force
1379 // at this point (the `Restore` above guarantees it), so this
1380 // is the one position where naming `M.frame` yields the
1381 // uncoerced original every later `Install` has to wrap.
1382 prelude.extend(v1::xver_adapt::deco_downgrade_prelude(
1383 &dep_deco_exports,
1384 v1::xver_adapt::DowngradeStep::Capture,
1385 ));
1386 for exp in &dep_deco_exports {
1387 xver_shadows.insert(v1::xver_adapt::deco_export_qualified_name(exp));
1388 }
1389 deco_exports.extend(dep_deco_exports);
1390 dep_csts.push(cst);
1391 }
1392 }
1393 }
1394
1395 // The last (and, for every bundled package, the ONLY) transition into
1396 // 0.0.6-authored code: the entry's own prelude AND its document tail are
1397 // both wrapped in `Ast::VersionScope(V0_0, _)` below, so both read a
1398 // crossed `deco` export at 0.0.6's `graphics list` shape.
1399 if !v006_view_installed && !deco_exports.is_empty() {
1400 prelude.extend(v1::xver_adapt::deco_downgrade_prelude(
1401 &deco_exports,
1402 v1::xver_adapt::DowngradeStep::Install,
1403 ));
1404 }
1405
1406 // Entry's OWN top-level lets: splice + wrap, same as a native 0.0.6 dep
1407 // (a new source of V0_0-tagged items beyond dependency splicing: not just
1408 // dependencies, but the entry itself) — and through the same
1409 // 0.0.6-authored type-text guard, for the same reason: the entry's
1410 // `type` declarations are hoisted into `Program::type_decls` alongside
1411 // everyone else's and read under the one hard-coded `V0_1` `Checker`,
1412 // with no `Ast::VersionScope` in reach.
1413 let entry_adapted = guard_v006_type_text(&entry_cst.prelude, &entry.path)?;
1414 let entry_start = prelude.len();
1415 prelude.extend(entry_adapted);
1416 v006_indices.extend(entry_start..prelude.len());
1417
1418 let file = rustyfi_syntax::cst::File {
1419 headers: Vec::new(),
1420 prelude,
1421 in_kw: entry_cst.in_kw.clone(),
1422 body: entry_cst.body.clone(),
1423 eoi: entry_cst.eoi.clone(),
1424 };
1425
1426 Ok(AssembledXver {
1427 file,
1428 dep_csts,
1429 v006_indices,
1430 stages,
1431 xver_shadows,
1432 })
1433}
1434
1435// ============================================================================
1436// Forked-name guard: before splicing a V0_0 dependency's `prelude` into a
1437// V0_1 program (above), walk it for the free (unqualified, unshadowed)
1438// primitive/type names it references and hard-reject any that is
1439// version-forked. This is what keeps the splice sound rather than silently wrong —
1440// see `compile_document_v1_with_trials`'s dep loop for the actual check.
1441//
1442// There is no generic CST visitor in this crate (the closest precedent,
1443// `typecheck.rs`'s `walk_atom`/`walk_expr` quartet, walks only
1444// `ast::TypeExpr` for tyvars); this is modeled on it but covers the FULL
1445// `cst::TopBinding`/`ast::Expr`/`ast::Pattern`/`ast::TypeExpr` grammar.
1446//
1447// Guard-narrowing: `free.values` is checked against nothing. For
1448// `free.types` the walk collects EXPORT-POSITION surface sites only — the
1449// ones a `V0_1` consumer of this dependency can actually observe:
1450// - a TOP-LEVEL `TopBinding::LetRec`'s (or `and` sibling's) own `: ty`
1451// ascription (`walk_top_binding`'s `LetRec` arm, `boundary = true`);
1452// - a `TopBinding::Module`'s `sig` items (`walk_sig_annot` — a `module ..`
1453// is only ever a top-level/struct-decl construct, never nested inside an
1454// expression, so every site it is walked from is already boundary);
1455// - a `TopBinding::Type` declaration's body (`walk_type_decl`, kept
1456// UNCONDITIONALLY boundary: a `type` declaration's ctor payload/synonym
1457// body is registered ONCE under the merged program's single ambient
1458// `V0_1` `Checker`, never inside an `Ast::VersionScope`, and a flat
1459// splice makes the declared name visible to the consumer — so "unused
1460// within this dependency" is not provable-safe here).
1461// An INTERNAL `Expr::LetRecIn` ascription is SKIPPED (`boundary = false`,
1462// `walk_expr`'s `LetRecIn` arm); it is the one place a forked type name can
1463// appear buried in an expression body in this port's 0.0.6 grammar, which has
1464// no local-lambda-parameter or local-`type` ascription syntax at all. See
1465// `walk_rec_binding_body`'s doc comment for the mechanism and the residual
1466// risk.
1467// ============================================================================
1468
1469/// The free, unqualified global names a spliced V0_0 dependency's
1470/// `prelude` references, split by namespace (values/commands vs. types)
1471/// because they are checked against DIFFERENT forked-name sets. See
1472/// `collect_free_globals`'s doc comment for the walk itself.
1473#[derive(Default, Debug)]
1474struct FreeGlobals {
1475 /// Value-position occurrences that could resolve to `base_env`:
1476 /// `Atomic::Var`/`Ctor`/`OpRef`/`Command`, the `Plain` arm of an
1477 /// `AnyHorz`/`Vert`/`MathCmdTok` reference, and an unqualified
1478 /// `…Elem::Embed`/`MathBot::Embed`. (No longer checked against
1479 /// anything — collected for completeness/tests only, see this module's
1480 /// banner comment above `walk_top_binding`.)
1481 values: BTreeSet<String>,
1482 /// EXPORT-BOUNDARY type-position occurrences only (see this
1483 /// module's "Guard-narrowing" banner comment above): `TypeAtom::Name`
1484 /// and the `ctor` of `TypeApp::Applied`, collected ONLY from a top-level
1485 /// binding's own ascription, a module's `sig`, or a `type` declaration's
1486 /// body — never from a purely-internal/local ascription.
1487 types: BTreeSet<String>,
1488}
1489
1490/// The binder scope threaded through `collect_free_globals`'s walk: two
1491/// independent namespaces (values/commands vs. types), each a plain stack of
1492/// names — pushing a name shadows an outer/global name of the SAME
1493/// namespace for the extent of whatever construct introduced it (`mark`/
1494/// `truncate_to` bound that extent, mirroring a lexical block's entry/exit).
1495///
1496/// **Soundness note.** This is a rejection GUARD, so over-approximation is
1497/// the safe direction: failing to push a genuine local binder just makes a
1498/// local look "free" (over-reporting — at worst an over-*rejection*, never
1499/// silently accepting something unsound). The one thing the walk must never
1500/// do is drop a binder scope *too early* / push something that ISN'T really
1501/// bound at that point, which would hide a genuine reference to a
1502/// version-forked global (see `Expr::OpenIn`'s arm below, which deliberately
1503/// binds NOTHING for an `open Mod in …` rather than guess at Mod's members).
1504#[derive(Default)]
1505struct XverScope {
1506 values: Vec<String>,
1507 types: Vec<String>,
1508}
1509
1510impl XverScope {
1511 fn mark(&self) -> (usize, usize) {
1512 (self.values.len(), self.types.len())
1513 }
1514
1515 fn truncate_to(&mut self, mark: (usize, usize)) {
1516 self.values.truncate(mark.0);
1517 self.types.truncate(mark.1);
1518 }
1519
1520 fn push_value(&mut self, name: &str) {
1521 self.values.push(name.to_string());
1522 }
1523
1524 fn push_type(&mut self, name: &str) {
1525 self.types.push(name.to_string());
1526 }
1527
1528 fn has_value(&self, name: &str) -> bool {
1529 self.values.iter().any(|v| v == name)
1530 }
1531
1532 fn has_type(&self, name: &str) -> bool {
1533 self.types.iter().any(|v| v == name)
1534 }
1535}
1536
1537fn emit_value(scope: &XverScope, out: &mut FreeGlobals, name: &str) {
1538 if !scope.has_value(name) {
1539 out.values.insert(name.to_string());
1540 }
1541}
1542
1543fn emit_type(scope: &XverScope, out: &mut FreeGlobals, name: &str) {
1544 if !scope.has_type(name) {
1545 out.types.insert(name.to_string());
1546 }
1547}
1548
1549/// Enumerate the *free, unqualified* global names a spliced V0_0
1550/// dependency references — `TopBinding`/`ast::Expr`/`ast::Pattern`/
1551/// `ast::TypeExpr`, each threading a binder scope stack so a locally-bound
1552/// name shadows a primitive of the same name (per `XverScope`'s doc
1553/// comment). A module-qualified reference (`Atomic::VarWithMod`,
1554/// `\Mod.cmd`/`+Mod.cmd`/`#Mod.var`) is deliberately SKIPPED: a primitive or
1555/// builtin type is only ever reachable by a BARE name, so a qualified
1556/// reference resolves inside a module and can never collide with a forked
1557/// primitive (0.0.6 has no qualified *type*-name form at all, so every type
1558/// reference is in scope for this check).
1559fn collect_free_globals(prelude: &[rustyfi_syntax::cst::TopBinding]) -> FreeGlobals {
1560 let mut out = FreeGlobals::default();
1561 let mut scope = XverScope::default();
1562 for tb in prelude {
1563 walk_top_binding(tb, &mut scope, &mut out);
1564 }
1565 out
1566}
1567
1568fn walk_top_binding(
1569 tb: &rustyfi_syntax::cst::TopBinding,
1570 scope: &mut XverScope,
1571 out: &mut FreeGlobals,
1572) {
1573 use rustyfi_syntax::cst::TopBinding;
1574 match tb {
1575 // Recursive: every clause's own name is bound BEFORE any clause body
1576 // is walked (and stays bound for every sibling `and` clause too).
1577 TopBinding::LetRec { first, ands, .. } => {
1578 scope.push_value(&first.name.name);
1579 for and in ands {
1580 scope.push_value(&and.binding.name.name);
1581 }
1582 // TOP-LEVEL — a consumer-observable export; `boundary = true`
1583 // (this binding's own `: ty` ascription IS export-position
1584 // text).
1585 walk_rec_binding_body(first, true, scope, out);
1586 for and in ands {
1587 walk_rec_binding_body(&and.binding, true, scope, out);
1588 }
1589 }
1590 TopBinding::Let(tl) => {
1591 // TOP-LEVEL, so this binding's own `: ty` ascription is
1592 // export-position text, exactly as `LetRec`'s is. Skipping it
1593 // would let `let x : page = ...` cross silently while `type
1594 // alias = page` was rejected — the same forked name, caught or
1595 // not depending on which way the package spelled it.
1596 if let Some(asc) = &tl.ascription {
1597 walk_type_expr(&asc.ty, scope, out);
1598 }
1599 let mark = scope.mark();
1600 for p in &tl.params {
1601 walk_param_binder(p, scope, out);
1602 }
1603 walk_expr(&tl.value, scope, out);
1604 scope.truncate_to(mark);
1605 scope.push_value(&tl.name.name);
1606 }
1607 TopBinding::LetPattern { value, .. } => {
1608 // Destructuring `let pat = value`: only the scrutinee references
1609 // free globals. The pattern-bound names become new bindings; not
1610 // pushing them here is sound (this walk over-approximates the free
1611 // set — see the module banner).
1612 walk_expr(value, scope, out);
1613 }
1614 TopBinding::LetInline {
1615 ctx,
1616 cmd,
1617 params,
1618 value,
1619 ..
1620 } => {
1621 let mark = scope.mark();
1622 if let Some(c) = ctx {
1623 scope.push_value(&c.name);
1624 }
1625 for p in params {
1626 walk_param_binder(p, scope, out);
1627 }
1628 walk_expr(value, scope, out);
1629 scope.truncate_to(mark);
1630 scope.push_value(&cmd.name);
1631 }
1632 TopBinding::LetBlock {
1633 ctx,
1634 cmd,
1635 params,
1636 value,
1637 ..
1638 } => {
1639 let mark = scope.mark();
1640 if let Some(c) = ctx {
1641 scope.push_value(&c.name);
1642 }
1643 for p in params {
1644 walk_param_binder(p, scope, out);
1645 }
1646 walk_expr(value, scope, out);
1647 scope.truncate_to(mark);
1648 scope.push_value(&cmd.name);
1649 }
1650 TopBinding::LetMath {
1651 cmd, params, value, ..
1652 } => {
1653 let mark = scope.mark();
1654 for p in params {
1655 walk_param_binder(p, scope, out);
1656 }
1657 walk_expr(value, scope, out);
1658 scope.truncate_to(mark);
1659 scope.push_value(&cmd.name);
1660 }
1661 TopBinding::Type(td) => {
1662 walk_type_decl(td, scope, out);
1663 scope.push_type(&td.name.name);
1664 }
1665 TopBinding::LetMutable { name, value, .. } => {
1666 walk_expr(value, scope, out);
1667 scope.push_value(&name.name);
1668 }
1669 TopBinding::Module { sig, decls, .. } => {
1670 if let Some(sig) = sig {
1671 walk_sig_annot(sig, scope, out);
1672 }
1673 // A nested module's own decls get a scope extent of their own —
1674 // its LOCAL bindings must not leak to a sibling top binding
1675 // outside the module.
1676 let mark = scope.mark();
1677 for d in decls {
1678 walk_top_binding(&d.0, scope, out);
1679 }
1680 scope.truncate_to(mark);
1681 // The module's own NAME (a `CtorTok`, uppercase-initial) is a
1682 // third namespace this guard doesn't track — it can never
1683 // collide with a lowercase primitive/type name.
1684 }
1685 // `open Mod` unqualified-imports Mod's members — unknowable
1686 // statically here (no elaboration has run yet), so this
1687 // conservatively binds NOTHING new: see `XverScope`'s doc comment
1688 // for why that is the safe direction (may over-reject, never hides
1689 // a real forked-name reference).
1690 TopBinding::Open { .. } => {}
1691 }
1692}
1693
1694/// Walk one `RecBinding`'s own params/value/`extra` clauses (shared by
1695/// `TopBinding::LetRec` and `Expr::LetRecIn`) — every clause's parameters are
1696/// scoped to that clause alone.
1697///
1698/// `boundary`: whether THIS `RecBinding` is a TOP-LEVEL,
1699/// consumer-observable export (`TopBinding::LetRec`/its `and` siblings —
1700/// `true`) or a purely LOCAL binding nested inside another binding's
1701/// expression body (`Expr::LetRecIn` — `false`). Only when `boundary` is
1702/// true does the binding's OWN `: ty` ascription get walked into
1703/// `out.types` — see the "Guard-narrowing" banner above
1704/// `collect_free_globals`.
1705fn walk_rec_binding_body(
1706 rb: &rustyfi_syntax::cst::ast::RecBinding,
1707 boundary: bool,
1708 scope: &mut XverScope,
1709 out: &mut FreeGlobals,
1710) {
1711 if boundary {
1712 if let Some(asc) = &rb.ascription {
1713 walk_type_expr(&asc.ty, scope, out);
1714 }
1715 }
1716 let mark = scope.mark();
1717 for p in &rb.params {
1718 walk_patbot_binder(p, scope, out);
1719 }
1720 walk_expr(&rb.value.0, scope, out);
1721 scope.truncate_to(mark);
1722 for clause in &rb.extra {
1723 let mark = scope.mark();
1724 for p in &clause.params {
1725 walk_patbot_binder(p, scope, out);
1726 }
1727 walk_expr(&clause.value.0, scope, out);
1728 scope.truncate_to(mark);
1729 }
1730}
1731
1732fn walk_param_binder(
1733 p: &rustyfi_syntax::cst::ast::Param,
1734 scope: &mut XverScope,
1735 out: &mut FreeGlobals,
1736) {
1737 use rustyfi_syntax::cst::ast::Param;
1738 match p {
1739 Param::Optional { name, .. } => scope.push_value(&name.name),
1740 Param::Pat(pb) => walk_patbot_binder(pb, scope, out),
1741 Param::Bundled { opts, body } => {
1742 for e in &opts.entries {
1743 scope.push_value(&e.var.name);
1744 }
1745 walk_patbot_binder(body, scope, out);
1746 }
1747 }
1748}
1749
1750/// Walk a full `patas` (a pattern plus its optional `as name` binding) in
1751/// BINDER mode: every `Var`/`AsClause.name` is pushed (never emitted); every
1752/// `Ctor`/`CtorApplied.ctor` is a REFERENCE — emitted for completeness (the
1753/// corpus's constructors are neutral, but this keeps the walk total).
1754fn walk_pattern_binder(
1755 pat: &rustyfi_syntax::cst::ast::Pattern,
1756 scope: &mut XverScope,
1757 out: &mut FreeGlobals,
1758) {
1759 walk_patcons_binder(&pat.head, scope, out);
1760 if let Some(ac) = &pat.as_clause {
1761 scope.push_value(&ac.name.name);
1762 }
1763}
1764
1765fn walk_patcons_binder(
1766 pc: &rustyfi_syntax::cst::ast::PatCons,
1767 scope: &mut XverScope,
1768 out: &mut FreeGlobals,
1769) {
1770 walk_patbot_binder(&pc.head, scope, out);
1771 for seg in &pc.tail {
1772 walk_patbot_binder(&seg.tail, scope, out);
1773 }
1774}
1775
1776fn walk_patbot_binder(
1777 pb: &rustyfi_syntax::cst::ast::PatBot,
1778 scope: &mut XverScope,
1779 out: &mut FreeGlobals,
1780) {
1781 use rustyfi_syntax::cst::ast::PatBot;
1782 match pb {
1783 PatBot::CtorApplied { ctor, arg } => {
1784 emit_value(scope, out, &ctor.name);
1785 walk_patbot_binder(arg, scope, out);
1786 }
1787 PatBot::Ctor(ctor) => emit_value(scope, out, &ctor.name),
1788 PatBot::Int(_) | PatBot::True(_) | PatBot::False(_) | PatBot::Str(_) | PatBot::Wild(_) => {}
1789 PatBot::Var(v) => scope.push_value(&v.name),
1790 PatBot::Unit { .. } => {}
1791 PatBot::Paren { inner, .. } => {
1792 walk_pattern_binder(&inner.first.0, scope, out);
1793 for cp in &inner.rest {
1794 walk_pattern_binder(&cp.value.0, scope, out);
1795 }
1796 }
1797 PatBot::List { items, .. } => {
1798 for it in items {
1799 walk_pattern_binder(&it.value.0, scope, out);
1800 }
1801 }
1802 }
1803}
1804
1805// ============================================================================
1806// The REVERSE arm's guard on **0.0.6-authored** type text
1807// (`compile_document_v006_xver_with_aux`'s `LoadedCst::V0_0` branch and the
1808// entry's own prelude).
1809//
1810// **The misreading**, reached from the other side of the forward arm's: a
1811// merged cross-version program has exactly one `Checker`, hard-coded to
1812// `V0_1` (`v1::module_check::check_program_inner`'s `ck.set_version`) on
1813// BOTH arms, because `elaborate` hoists every `type` declaration out of the
1814// `Ast` spine into `Program::type_decls`/`synonym_decls` — never inside an
1815// `Ast::VersionScope`. Forward, the 0.0.6 text re-read under 0.1's
1816// vocabulary is a spliced dependency's; reverse, it is the ENTRY's own
1817// prelude plus every native 0.0.6 co-dependency — potentially the whole
1818// 0.0.6 corpus.
1819//
1820// So `math` takes the **same** relabel here as forward, `math` ->
1821// `math-text` (`xver_adapt::relabel_type_decls(_, V0_0, V0_1)`), NOT the
1822// mirror `math-text` -> `math`: the target vocabulary is `V0_1` either way.
1823// (`relabel_or_reject_name`'s mirror arm is deliberately not wired to the
1824// reverse arm's `LoadedCst::V0_1` branch — a foreign 0.1 dependency's text
1825// is already in the ambient vocabulary.)
1826//
1827// **Why this scan is narrower than `collect_free_globals`.** The forward
1828// arm over-approximates on purpose, also collecting from a `let-rec`'s
1829// `: ty` ascription and a `module .. : sig .. end`'s `val` items — both
1830// parsed and then ignored by `elaborate.rs`, so over-rejecting on them only
1831// costs a 0.1 document a 0.0.6 package it could have had. Reversed, the
1832// same over-approximation would be WRONG, not conservative: the bundled
1833// 0.0.6 corpus writes forked names in exactly those decorative positions
1834// all the time (`vdecoset.satyh`'s `val paper : deco-set`, `math.satyh`'s
1835// `direct \frac : [math; math] math-cmd`), so rejecting on them would
1836// refuse ordinary 0.0.6 documents for text no phase reads. This walk
1837// instead collects from `TopBinding::Type` bodies alone (recursing through
1838// `TopBinding::Module`'s nested `decls`) — exactly the site set
1839// `xver_adapt::relabel_type_decls` rewrites, and the text that reaches
1840// `declare_variant`/`declare_synonym`.
1841//
1842// **What is refused.** `reject_type_names_from_v006()` — the same
1843// producer-keyed set the forward arm's `V0_0` branch uses, so `code`
1844// refuses here too (a foreign 0.1 dependency's `code`, the reverse arm's
1845// OTHER branch, keeps the shared `reject_type_names()` and does not). The
1846// whitelist is `{"math"}` alone: this branch has no
1847// `classify_deco_exports`/`deco_coercion_prelude` pairing to make a
1848// `deco`/`deco-set`/`paren` mention safe, and the 0.1 reading of those names
1849// is wrong for a 0.0.6-authored consumer anyway (0.0.6's `deco` returns
1850// `graphics list`; `name_to_mono("deco", V0_1)` types it as a single
1851// `graphics`). `page` is the sharp one: its bare name lowers to the
1852// same nominal `Variant("page",[])` under both versions, so a mismatch is
1853// not a type error at all — a 9-ctor `Value::Ctor` meeting a `length *
1854// length` `Value::Product`.
1855// ============================================================================
1856
1857/// The free type names a 0.0.6-authored `prelude`'s `type` DECLARATIONS
1858/// mention — the whole of that prelude's text a merged cross-version
1859/// program's single hard-coded-`V0_1` `Checker` actually reads (see this
1860/// module's banner above for why the decorative
1861/// ascription/`sig` sites are deliberately NOT collected here, though
1862/// `collect_free_globals` does collect them for the forward arm).
1863fn collect_type_decl_globals(
1864 prelude: &[rustyfi_syntax::cst::TopBinding],
1865) -> std::collections::BTreeSet<String> {
1866 let mut out = FreeGlobals::default();
1867 let mut scope = XverScope::default();
1868 for tb in prelude {
1869 walk_type_decls_only(tb, &mut scope, &mut out);
1870 }
1871 out.types
1872}
1873
1874fn walk_type_decls_only(
1875 tb: &rustyfi_syntax::cst::TopBinding,
1876 scope: &mut XverScope,
1877 out: &mut FreeGlobals,
1878) {
1879 use rustyfi_syntax::cst::TopBinding;
1880 match tb {
1881 TopBinding::Type(td) => {
1882 walk_type_decl(td, scope, out);
1883 scope.push_type(&td.name.name);
1884 }
1885 // A nested `type` declaration is hoisted into the SAME
1886 // `Program::type_decls` as a top-level one (`elaborate::
1887 // walk_bindings` threads one `type_decls` sink through every level),
1888 // so it is read under the same hard-coded `V0_1` `Checker` and must
1889 // be scanned too. Its locally-declared names stay local, matching
1890 // `walk_top_binding`'s own `Module` arm.
1891 TopBinding::Module { decls, .. } => {
1892 let mark = scope.mark();
1893 for d in decls {
1894 walk_type_decls_only(&d.0, scope, out);
1895 }
1896 scope.truncate_to(mark);
1897 }
1898 _ => {}
1899 }
1900}
1901
1902/// Check one 0.0.6-authored `prelude` on the REVERSE arm and
1903/// return the bindings to splice — relabeled (`math` -> `math-text`) when
1904/// that is all it touches, cloned verbatim when it touches nothing, and a
1905/// `CompileError::CrossVersionUnsupportedName` naming the offending type
1906/// otherwise. `path` is the file the text was authored in (the 0.0.6 entry,
1907/// or a native 0.0.6 co-dependency); the resulting error records which
1908/// DIRECTION refused, since the forward arm's guard checks the same
1909/// producer-keyed set under its own tag.
1910fn guard_v006_type_text(
1911 prelude: &[rustyfi_syntax::cst::TopBinding],
1912 path: &std::path::Path,
1913) -> Result<Vec<rustyfi_syntax::cst::TopBinding>, CompileError> {
1914 use rustyfi_syntax::RustyfiVersion;
1915 let reject_t = v1::xver_adapt::reject_type_names_from_v006();
1916 let touched: BTreeSet<String> = collect_type_decl_globals(prelude)
1917 .intersection(&reject_t)
1918 .cloned()
1919 .collect();
1920 // `math` is the whole whitelist here — see the banner above.
1921 if let Some(name) = touched.iter().find(|n| n.as_str() != "math") {
1922 return Err(CompileError::CrossVersionUnsupportedName {
1923 name: name.clone(),
1924 dep: path.display().to_string(),
1925 slice: "X4c",
1926 });
1927 }
1928 if touched.is_empty() {
1929 // Byte-identical to the `prelude.extend(cst.prelude.iter()
1930 // .cloned())` fast path every non-`math` 0.0.6 file takes.
1931 return Ok(prelude.to_vec());
1932 }
1933 v1::xver_adapt::relabel_type_decls(prelude, RustyfiVersion::V0_0, RustyfiVersion::V0_1).map_err(
1934 |be| CompileError::CrossVersionUnsupportedName {
1935 name: match &be {
1936 v1::xver_adapt::BoundaryError::ForkedTypeExport { ty_name, .. } => ty_name.clone(),
1937 },
1938 dep: path.display().to_string(),
1939 slice: "X4c",
1940 },
1941 )
1942}
1943
1944fn walk_type_decl(
1945 td: &rustyfi_syntax::cst::TypeDecl,
1946 scope: &mut XverScope,
1947 out: &mut FreeGlobals,
1948) {
1949 walk_type_decl_body(&td.body, scope, out);
1950 for a in &td.ands {
1951 walk_type_decl_body(&a.body, scope, out);
1952 }
1953}
1954
1955fn walk_type_decl_body(
1956 body: &rustyfi_syntax::cst::TypeDeclBody,
1957 scope: &mut XverScope,
1958 out: &mut FreeGlobals,
1959) {
1960 use rustyfi_syntax::cst::TypeDeclBody;
1961 match body {
1962 TypeDeclBody::Variant { first, rest, .. } => {
1963 walk_variant_def(first, scope, out);
1964 for bv in rest {
1965 walk_variant_def(&bv.def, scope, out);
1966 }
1967 }
1968 TypeDeclBody::Synonym(ty) => walk_type_expr(ty, scope, out),
1969 }
1970}
1971
1972fn walk_variant_def(
1973 vd: &rustyfi_syntax::cst::VariantDef,
1974 scope: &mut XverScope,
1975 out: &mut FreeGlobals,
1976) {
1977 // `vd.ctor` DECLARES a new constructor — not a reference, nothing to
1978 // emit for it.
1979 if let Some(of_ty) = &vd.of_ty {
1980 walk_type_expr(&of_ty.ty, scope, out);
1981 }
1982}
1983
1984fn walk_sig_annot(
1985 sig: &rustyfi_syntax::cst::SigAnnot,
1986 scope: &mut XverScope,
1987 out: &mut FreeGlobals,
1988) {
1989 use rustyfi_syntax::cst::SigItem;
1990 for item in &sig.items {
1991 match item {
1992 SigItem::ValHorzCmd { ty, .. }
1993 | SigItem::ValVertCmd { ty, .. }
1994 | SigItem::Val { ty, .. }
1995 | SigItem::DirectHorzCmd { ty, .. }
1996 | SigItem::DirectVertCmd { ty, .. } => walk_type_expr(ty, scope, out),
1997 SigItem::Type { .. } => {}
1998 }
1999 }
2000}
2001
2002fn walk_expr(e: &rustyfi_syntax::cst::ast::Expr, scope: &mut XverScope, out: &mut FreeGlobals) {
2003 use rustyfi_syntax::cst::ast::Expr;
2004 match e {
2005 Expr::LetRecIn {
2006 first, ands, body, ..
2007 } => {
2008 let mark = scope.mark();
2009 scope.push_value(&first.name.name);
2010 for and in ands {
2011 scope.push_value(&and.binding.name.name);
2012 }
2013 // INTERNAL — a local binding nested inside some enclosing
2014 // binding's own body; `boundary = false` (this `let rec`'s
2015 // OWN `: ty` ascription is not, by itself, any export's
2016 // observable signature text — see `walk_rec_binding_body`'s doc
2017 // comment).
2018 walk_rec_binding_body(first, false, scope, out);
2019 for and in ands {
2020 walk_rec_binding_body(&and.binding, false, scope, out);
2021 }
2022 walk_expr(body, scope, out);
2023 scope.truncate_to(mark);
2024 }
2025 Expr::LetIn {
2026 name,
2027 params,
2028 value,
2029 body,
2030 ..
2031 } => {
2032 let mark = scope.mark();
2033 for p in params {
2034 walk_param_binder(p, scope, out);
2035 }
2036 walk_expr(value, scope, out);
2037 scope.truncate_to(mark);
2038 let mark = scope.mark();
2039 scope.push_value(&name.name);
2040 walk_expr(body, scope, out);
2041 scope.truncate_to(mark);
2042 }
2043 Expr::LetPatternIn {
2044 pat, value, body, ..
2045 } => {
2046 walk_expr(value, scope, out);
2047 let mark = scope.mark();
2048 walk_pattern_binder(&pat.0, scope, out);
2049 walk_expr(body, scope, out);
2050 scope.truncate_to(mark);
2051 }
2052 Expr::If {
2053 cond,
2054 then_branch,
2055 else_branch,
2056 ..
2057 } => {
2058 walk_expr(cond, scope, out);
2059 walk_expr(then_branch, scope, out);
2060 walk_expr(else_branch, scope, out);
2061 }
2062 Expr::Fun { params, body, .. } => {
2063 let mark = scope.mark();
2064 for p in params {
2065 walk_patbot_binder(p, scope, out);
2066 }
2067 walk_expr(body, scope, out);
2068 scope.truncate_to(mark);
2069 }
2070 Expr::FunRows {
2071 opts, param, body, ..
2072 } => {
2073 let mark = scope.mark();
2074 for e in &opts.entries {
2075 scope.push_value(&e.var.name);
2076 }
2077 walk_patbot_binder(param, scope, out);
2078 walk_expr(body, scope, out);
2079 scope.truncate_to(mark);
2080 }
2081 Expr::Match {
2082 scrutinee,
2083 first,
2084 rest,
2085 ..
2086 } => {
2087 walk_expr(scrutinee, scope, out);
2088 walk_match_arm(first, scope, out);
2089 for ba in rest {
2090 walk_match_arm(&ba.arm, scope, out);
2091 }
2092 }
2093 Expr::LetMutableIn {
2094 name, init, body, ..
2095 } => {
2096 walk_expr(init, scope, out);
2097 let mark = scope.mark();
2098 scope.push_value(&name.name);
2099 walk_expr(body, scope, out);
2100 scope.truncate_to(mark);
2101 }
2102 Expr::LetMathIn {
2103 cmd,
2104 params,
2105 value,
2106 body,
2107 ..
2108 } => {
2109 let mark = scope.mark();
2110 for p in params {
2111 walk_param_binder(p, scope, out);
2112 }
2113 walk_expr(value, scope, out);
2114 scope.truncate_to(mark);
2115 let mark = scope.mark();
2116 scope.push_value(&cmd.name);
2117 walk_expr(body, scope, out);
2118 scope.truncate_to(mark);
2119 }
2120 // `open Mod in body` — see `TopBinding::Open`'s arm for why this
2121 // binds nothing new.
2122 Expr::OpenIn { body, .. } => walk_expr(body, scope, out),
2123 Expr::WhileDo { cond, body, .. } => {
2124 walk_expr(cond, scope, out);
2125 walk_expr(body, scope, out);
2126 }
2127 Expr::Overwrite { name, value, .. } => {
2128 emit_value(scope, out, &name.name);
2129 walk_expr(&value.0, scope, out);
2130 }
2131 Expr::Ops(chain) => walk_opchain(chain, scope, out),
2132 }
2133}
2134
2135fn walk_match_arm(
2136 arm: &rustyfi_syntax::cst::ast::MatchArm,
2137 scope: &mut XverScope,
2138 out: &mut FreeGlobals,
2139) {
2140 let mark = scope.mark();
2141 walk_pattern_binder(&arm.pat.0, scope, out);
2142 if let Some(g) = &arm.guard {
2143 walk_expr(&g.cond.0, scope, out);
2144 }
2145 walk_expr(&arm.body.0, scope, out);
2146 scope.truncate_to(mark);
2147}
2148
2149fn walk_opchain(
2150 oc: &rustyfi_syntax::cst::ast::OpChain,
2151 scope: &mut XverScope,
2152 out: &mut FreeGlobals,
2153) {
2154 walk_appexpr(&oc.head, scope, out);
2155 for r in &oc.tail {
2156 walk_appexpr(&r.rhs, scope, out);
2157 }
2158 if let Some(bt) = &oc.before {
2159 walk_expr(&bt.body.0, scope, out);
2160 }
2161}
2162
2163fn walk_appexpr(
2164 ae: &rustyfi_syntax::cst::ast::AppExpr,
2165 scope: &mut XverScope,
2166 out: &mut FreeGlobals,
2167) {
2168 walk_atomic(&ae.head, scope, out);
2169 // `head_accesses`: `#label` record-field accesses — field labels, not
2170 // globals, skip.
2171 for arg in &ae.args {
2172 walk_apparg(arg, scope, out);
2173 }
2174}
2175
2176fn walk_apparg(a: &rustyfi_syntax::cst::ast::AppArg, scope: &mut XverScope, out: &mut FreeGlobals) {
2177 use rustyfi_syntax::cst::ast::AppArg;
2178 match a {
2179 AppArg::Optional { value, .. } => walk_atomic(value, scope, out),
2180 AppArg::Omission(_) => {}
2181 AppArg::Atom { atom, .. } => walk_atomic(atom, scope, out),
2182 AppArg::Ctor(c) => emit_value(scope, out, &c.name),
2183 AppArg::Bundled { opts, atom, .. } => {
2184 for e in &opts.entries {
2185 walk_expr(&e.value.0, scope, out);
2186 }
2187 walk_atomic(atom, scope, out);
2188 }
2189 AppArg::BundledCtor { opts, ctor } => {
2190 for e in &opts.entries {
2191 walk_expr(&e.value.0, scope, out);
2192 }
2193 emit_value(scope, out, &ctor.name);
2194 }
2195 }
2196}
2197
2198fn walk_atomic(a: &rustyfi_syntax::cst::ast::Atomic, scope: &mut XverScope, out: &mut FreeGlobals) {
2199 use rustyfi_syntax::cst::ast::Atomic;
2200 match a {
2201 Atomic::Length(_)
2202 | Atomic::Float(_)
2203 | Atomic::Int(_)
2204 | Atomic::Literal(_)
2205 | Atomic::True(_)
2206 | Atomic::False(_) => {}
2207 Atomic::Ctor(c) => emit_value(scope, out, &c.name),
2208 Atomic::Var(v) => emit_value(scope, out, &v.name),
2209 // Qualified — resolves inside the module, never against `base_env`.
2210 Atomic::VarWithMod(_) => {}
2211 Atomic::OpRef(op) => emit_value(scope, out, &op.name),
2212 Atomic::Command { name, .. } => walk_any_horz_cmd_ref(name, scope, out),
2213 Atomic::Unit { .. } => {}
2214 Atomic::Paren { inner, .. } => walk_paren_body(inner, scope, out),
2215 Atomic::OpenModule { body, .. } => walk_paren_body(body, scope, out),
2216 Atomic::Record { body, .. } => walk_record_body(body, scope, out),
2217 Atomic::List { items, .. } => {
2218 for it in items {
2219 walk_expr(&it.value.0, scope, out);
2220 }
2221 }
2222 Atomic::InlineText { elems, .. } => {
2223 for el in elems {
2224 walk_inline_elem(el, scope, out);
2225 }
2226 }
2227 Atomic::BlockText { elems, .. } => {
2228 for el in elems {
2229 walk_block_elem(el, scope, out);
2230 }
2231 }
2232 Atomic::MathText { elems, .. } => {
2233 for el in elems {
2234 walk_math_elem(&el.0, scope, out);
2235 }
2236 }
2237 }
2238}
2239
2240fn walk_any_horz_cmd_ref(
2241 n: &rustyfi_syntax::leaf::AnyHorzCmdTok,
2242 scope: &XverScope,
2243 out: &mut FreeGlobals,
2244) {
2245 use rustyfi_syntax::leaf::AnyHorzCmdTok;
2246 match n {
2247 AnyHorzCmdTok::Plain(t) => emit_value(scope, out, &t.name),
2248 AnyHorzCmdTok::Mod(_) => {} // qualified — skip
2249 }
2250}
2251
2252fn walk_any_vert_cmd_ref(
2253 n: &rustyfi_syntax::leaf::AnyVertCmdTok,
2254 scope: &XverScope,
2255 out: &mut FreeGlobals,
2256) {
2257 use rustyfi_syntax::leaf::AnyVertCmdTok;
2258 match n {
2259 AnyVertCmdTok::Plain(t) => emit_value(scope, out, &t.name),
2260 AnyVertCmdTok::Mod(_) => {} // qualified — skip
2261 }
2262}
2263
2264fn walk_any_math_cmd_ref(
2265 n: &rustyfi_syntax::leaf::AnyMathCmdTok,
2266 scope: &XverScope,
2267 out: &mut FreeGlobals,
2268) {
2269 use rustyfi_syntax::leaf::AnyMathCmdTok;
2270 match n {
2271 AnyMathCmdTok::Plain(t) => emit_value(scope, out, &t.name),
2272 AnyMathCmdTok::Mod(_) => {} // qualified — skip
2273 }
2274}
2275
2276fn walk_paren_body(
2277 pb: &rustyfi_syntax::cst::ast::ParenBody,
2278 scope: &mut XverScope,
2279 out: &mut FreeGlobals,
2280) {
2281 walk_expr(&pb.first.0, scope, out);
2282 for ce in &pb.rest {
2283 walk_expr(&ce.value.0, scope, out);
2284 }
2285}
2286
2287fn walk_record_body(
2288 rb: &rustyfi_syntax::cst::ast::RecordBody,
2289 scope: &mut XverScope,
2290 out: &mut FreeGlobals,
2291) {
2292 use rustyfi_syntax::cst::ast::RecordBody;
2293 match rb {
2294 RecordBody::Update { base, fields, .. } => {
2295 walk_expr(&base.0, scope, out);
2296 for f in fields {
2297 walk_expr(&f.value.0, scope, out);
2298 }
2299 }
2300 RecordBody::Fields(fields) => {
2301 for f in fields {
2302 walk_expr(&f.value.0, scope, out);
2303 }
2304 }
2305 }
2306}
2307
2308fn walk_inline_elem(
2309 el: &rustyfi_syntax::cst::ast::InlineElem,
2310 scope: &mut XverScope,
2311 out: &mut FreeGlobals,
2312) {
2313 use rustyfi_syntax::cst::ast::InlineElem;
2314 match el {
2315 InlineElem::Char(_)
2316 | InlineElem::CodeText(_)
2317 | InlineElem::Space(_)
2318 | InlineElem::Break(_) => {}
2319 InlineElem::Embed { var, .. } => {
2320 if var.mods.is_empty() {
2321 emit_value(scope, out, &var.name);
2322 }
2323 }
2324 InlineElem::EmbedMath { elems, .. } => {
2325 for m in elems {
2326 walk_math_elem(&m.0, scope, out);
2327 }
2328 }
2329 InlineElem::Cmd { name, tail } => {
2330 walk_any_horz_cmd_ref(name, scope, out);
2331 walk_cmd_tail(tail, scope, out);
2332 }
2333 InlineElem::ItemBullet(_) | InlineElem::Sep(_) => {}
2334 }
2335}
2336
2337fn walk_block_elem(
2338 el: &rustyfi_syntax::cst::ast::BlockElem,
2339 scope: &mut XverScope,
2340 out: &mut FreeGlobals,
2341) {
2342 use rustyfi_syntax::cst::ast::BlockElem;
2343 match el {
2344 BlockElem::Embed { var, .. } => {
2345 if var.mods.is_empty() {
2346 emit_value(scope, out, &var.name);
2347 }
2348 }
2349 BlockElem::Cmd { name, tail } => {
2350 walk_any_vert_cmd_ref(name, scope, out);
2351 walk_cmd_tail(tail, scope, out);
2352 }
2353 }
2354}
2355
2356fn walk_cmd_tail(
2357 t: &rustyfi_syntax::cst::ast::CmdTail,
2358 scope: &mut XverScope,
2359 out: &mut FreeGlobals,
2360) {
2361 use rustyfi_syntax::cst::ast::CmdTail;
2362 match t {
2363 CmdTail::Semi(_) => {}
2364 CmdTail::Args { first, rest, .. } => {
2365 walk_apparg(&first.0, scope, out);
2366 for a in rest {
2367 walk_apparg(&a.0, scope, out);
2368 }
2369 }
2370 }
2371}
2372
2373fn walk_math_elem(
2374 m: &rustyfi_syntax::cst::ast::MathElemCst,
2375 scope: &mut XverScope,
2376 out: &mut FreeGlobals,
2377) {
2378 walk_math_bot(&m.base, scope, out);
2379 for s in &m.scripts {
2380 walk_math_script(s, scope, out);
2381 }
2382}
2383
2384fn walk_math_script(
2385 s: &rustyfi_syntax::cst::ast::MathScript,
2386 scope: &mut XverScope,
2387 out: &mut FreeGlobals,
2388) {
2389 use rustyfi_syntax::cst::ast::MathScript;
2390 match s {
2391 MathScript::Super { group, .. } | MathScript::Sub { group, .. } => {
2392 walk_math_group_arg(group, scope, out)
2393 }
2394 MathScript::Primes(_) => {}
2395 }
2396}
2397
2398fn walk_math_group_arg(
2399 g: &rustyfi_syntax::cst::ast::MathGroupArg,
2400 scope: &mut XverScope,
2401 out: &mut FreeGlobals,
2402) {
2403 use rustyfi_syntax::cst::ast::MathGroupArg;
2404 match g {
2405 MathGroupArg::Group { elems, .. } => {
2406 for m in elems {
2407 walk_math_elem(&m.0, scope, out);
2408 }
2409 }
2410 MathGroupArg::Bot(b) => walk_math_bot(b, scope, out),
2411 }
2412}
2413
2414fn walk_math_bot(
2415 b: &rustyfi_syntax::cst::ast::MathBot,
2416 scope: &mut XverScope,
2417 out: &mut FreeGlobals,
2418) {
2419 use rustyfi_syntax::cst::ast::MathBot;
2420 match b {
2421 MathBot::Cmd { name, args } => {
2422 walk_any_math_cmd_ref(name, scope, out);
2423 for a in args {
2424 walk_math_arg(a, scope, out);
2425 }
2426 }
2427 MathBot::Chars(_) => {}
2428 MathBot::Embed(v) => {
2429 if v.mods.is_empty() {
2430 emit_value(scope, out, &v.name);
2431 }
2432 }
2433 MathBot::Sep(_) => {}
2434 MathBot::Group { elems, .. } => {
2435 for m in elems {
2436 walk_math_elem(&m.0, scope, out);
2437 }
2438 }
2439 }
2440}
2441
2442fn walk_math_arg(
2443 a: &rustyfi_syntax::cst::ast::MathArg,
2444 scope: &mut XverScope,
2445 out: &mut FreeGlobals,
2446) {
2447 use rustyfi_syntax::cst::ast::MathArg;
2448 match a {
2449 MathArg::Optional { body, .. } => walk_math_arg_body(body, scope, out),
2450 MathArg::Omission(_) => {}
2451 MathArg::Plain(body) => walk_math_arg_body(body, scope, out),
2452 }
2453}
2454
2455fn walk_math_arg_body(
2456 b: &rustyfi_syntax::cst::ast::MathArgBody,
2457 scope: &mut XverScope,
2458 out: &mut FreeGlobals,
2459) {
2460 use rustyfi_syntax::cst::ast::MathArgBody;
2461 match b {
2462 MathArgBody::Math { elems, .. } => {
2463 for m in elems {
2464 walk_math_elem(&m.0, scope, out);
2465 }
2466 }
2467 MathArgBody::Inline { elems, .. } => {
2468 for el in elems {
2469 walk_inline_elem(el, scope, out);
2470 }
2471 }
2472 MathArgBody::Block { elems, .. } => {
2473 for el in elems {
2474 walk_block_elem(el, scope, out);
2475 }
2476 }
2477 MathArgBody::ParenEscape { inner, .. } => walk_paren_body(inner, scope, out),
2478 MathArgBody::ListEscape { items, .. } => {
2479 for it in items {
2480 walk_expr(&it.value.0, scope, out);
2481 }
2482 }
2483 MathArgBody::RecordEscape { body, .. } => walk_record_body(body, scope, out),
2484 }
2485}
2486
2487fn walk_type_expr(
2488 te: &rustyfi_syntax::cst::ast::TypeExpr,
2489 scope: &mut XverScope,
2490 out: &mut FreeGlobals,
2491) {
2492 use rustyfi_syntax::cst::ast::TypeExpr;
2493 match te {
2494 TypeExpr::Fun { opts, dom, cod, .. } => {
2495 for o in opts {
2496 walk_type_prod(&o.ty, scope, out);
2497 }
2498 walk_type_prod(dom, scope, out);
2499 walk_type_expr(cod, scope, out);
2500 }
2501 TypeExpr::Atom(prod) => walk_type_prod(prod, scope, out),
2502 TypeExpr::OptRowFun {
2503 opt_dom, dom, cod, ..
2504 } => {
2505 for e in &opt_dom.entries {
2506 walk_type_expr(&e.ty.0, scope, out);
2507 }
2508 walk_type_prod(dom, scope, out);
2509 walk_type_expr(cod, scope, out);
2510 }
2511 }
2512}
2513
2514fn walk_type_prod(
2515 tp: &rustyfi_syntax::cst::ast::TypeProd,
2516 scope: &mut XverScope,
2517 out: &mut FreeGlobals,
2518) {
2519 walk_type_app(&tp.first, scope, out);
2520 for st in &tp.rest {
2521 walk_type_app(&st.ty, scope, out);
2522 }
2523}
2524
2525fn walk_type_app(
2526 ta: &rustyfi_syntax::cst::ast::TypeApp,
2527 scope: &mut XverScope,
2528 out: &mut FreeGlobals,
2529) {
2530 // Every atom of a postfix application `arg1 … ctor` — including the final
2531 // constructor — is a `TypeAtom`, and `walk_type_atom` already emits a bare
2532 // `Name` (and skips a module-qualified `NameMod`) as an
2533 // export-boundary type reference, so walking the whole run reproduces the
2534 // old per-arg-then-ctor behavior exactly.
2535 walk_type_atom(&ta.head, scope, out);
2536 for a in &ta.rest {
2537 walk_type_atom(a, scope, out);
2538 }
2539}
2540
2541fn walk_type_atom(
2542 atom: &rustyfi_syntax::cst::ast::TypeAtom,
2543 scope: &mut XverScope,
2544 out: &mut FreeGlobals,
2545) {
2546 use rustyfi_syntax::cst::ast::TypeAtom;
2547 match atom {
2548 TypeAtom::Cmd { args, .. } => {
2549 for a in args {
2550 for l in &a.opt_labels {
2551 walk_type_expr(&l.ty.0, scope, out);
2552 }
2553 walk_type_expr(&a.ty.0, scope, out);
2554 }
2555 }
2556 TypeAtom::Paren { inner, .. } => walk_type_expr(&inner.0, scope, out),
2557 TypeAtom::Record { fields, .. } => {
2558 for f in fields {
2559 walk_type_expr(&f.ty.0, scope, out);
2560 }
2561 }
2562 // A bound type variable — never a forked-name candidate.
2563 TypeAtom::Var(_) => {}
2564 TypeAtom::Name(n) => emit_type(scope, out, &n.name),
2565 // `Mod.t` — already qualified, not a free unqualified global.
2566 TypeAtom::NameMod(_) => {}
2567 TypeAtom::RecordOpen { inner, .. } => {
2568 for f in &inner.fields {
2569 walk_type_expr(&f.ty.0, scope, out);
2570 }
2571 }
2572 }
2573}
2574
2575/// One `block-frame-breakable` frame currently between its `FrameStart`/
2576/// `FrameEnd` markers on the page being walked.
2577struct OpenFrame {
2578 id: DecoId,
2579 /// The `FrameStart` marker's own `PlacedLine.x` — the frame's left edge.
2580 x: Length,
2581 /// The `FrameStart` marker's own baseline — the degenerate-rect fallback
2582 /// used at close time when NO real line ever appeared between Start/End
2583 /// (an empty frame). NOT used to seed `top`/`bottom` directly (a
2584 /// marker's own baseline is just wherever the previous line happened to
2585 /// end, unrelated to real content extent).
2586 marker_baseline: Length,
2587 /// Running (top, bottom) extent in page (y-down) coordinates, `None`
2588 /// until the first real line is seen between this frame's Start/End.
2589 top: Option<Length>,
2590 bottom: Option<Length>,
2591 /// Insertion order — used to sort same-page fires back into outer-before-
2592 /// inner document order (see the ordering note below).
2593 open_seq: usize,
2594 /// `true` once this frame has already emitted a head (`decoH`) or middle
2595 /// (`decoM`) fragment on an EARLIER page — i.e. its `FrameStart` landed on
2596 /// a previous page and it is still open. Drives the S/H/M/T choice: a
2597 /// non-carried frame closing on its start page fires `decoS`; a carried one
2598 /// fires `decoT`. At each page boundary a still-open frame fires `decoH`
2599 /// (first spanned page) or `decoM` (subsequent) and its per-page extent is
2600 /// reset. `false` for the common single-page frame (unchanged behaviour).
2601 carried: bool,
2602}
2603
2604/// The inline twin of [`OpenFrame`]: one `inline-frame-breakable` that is open
2605/// at some point during the placed-line walk. Where a block frame's fragments
2606/// are delimited by PAGE boundaries, an inline frame's are delimited by LINE
2607/// boundaries (upstream `append_framed_lines`, `lineBreak.ml:695`), so one of
2608/// these can open and close within a single line — the common case, and the
2609/// one that fires `decoS`.
2610#[derive(Clone)]
2611struct OpenInlineFrame {
2612 id: DecoId,
2613 /// Left edge of the fragment currently being accumulated, in absolute page
2614 /// coordinates: the start marker's own x on the line that opened the
2615 /// frame, and the line's own left edge on every continuation line.
2616 x: Length,
2617 /// Baseline of the line the current fragment sits on.
2618 baseline_y: Length,
2619 /// The frame's padded vertical extent, carried on its markers — see
2620 /// `PureHorzBox::InlineFrameMarker` for why it is the whole frame's rather
2621 /// than this fragment's.
2622 height: Length,
2623 depth: Length,
2624 /// `true` once an earlier fragment of this frame has already fired, i.e.
2625 /// the frame really did split — same S/H/M/T choice as `OpenFrame`'s
2626 /// `carried` field above.
2627 carried: bool,
2628}
2629
2630/// The absolute x just past a placed line's last box — where a fragment that
2631/// runs off the end of this line has to stop.
2632///
2633/// Uses each box's NATURAL width rather than its justified advance: the only
2634/// boxes whose two differ are glue, `line_content` trims a line's trailing
2635/// glue away, and an `inline-fil` that survives (the `… ++ inline-fil`
2636/// flush-left idiom) is exactly the case where the fragment should stop at the
2637/// ink rather than at the stretched fil.
2638fn placed_line_right_edge(line: &rustyfi_backend::PlacedLine) -> Length {
2639 let mut edge = Length::ZERO;
2640 for (dx, bx) in &line.contents {
2641 let right = *dx + bx.natural_width();
2642 if right > edge {
2643 edge = right;
2644 }
2645 }
2646 line.x + edge
2647}
2648
2649/// Fire one placed `inline-frame-breakable` fragment's decoration
2650/// (`decoS`/`decoH`/`decoM`/`decoT` picked by `deco_idx` — 0/1/2/3,
2651/// evalUtil.ml:169 `get_decoset` order), spanning `frame.x` to `right`.
2652///
2653/// The vertical padding is already folded into `frame.height`/`frame.depth`
2654/// (the markers carry the padded extent), so unlike the block twin this takes
2655/// no per-fragment pad selection: upstream's `append_vert_padding`
2656/// (`lineBreak.ml:74`) applies `paddingT`/`paddingB` to EVERY fragment of an
2657/// inline frame, not just the first and last — the split is horizontal, so
2658/// each fragment has its own full-height top and bottom edge.
2659fn fire_inline_frame_fragment(
2660 interp: &mut eval::Interp,
2661 paper_height: Length,
2662 page: usize,
2663 frame: &OpenInlineFrame,
2664 right: Length,
2665 deco_idx: usize,
2666) -> Result<(), eval::EvalError> {
2667 // A decoration, not a hook: the hooks-only half of the walk skips it.
2668 if interp.fire_pass == eval::FirePass::HooksOnly {
2669 return Ok(());
2670 }
2671 let (deco, deco_version) = match &interp.decos[frame.id.0] {
2672 eval::DecoEntry::InlineBreakable {
2673 decoset, version, ..
2674 } => (decoset[deco_idx].clone(), *version),
2675 _ => {
2676 return eval::eval_error("BUG: non-breakable deco behind an inline frame marker");
2677 }
2678 };
2679 let width = right - frame.x;
2680 let pt = (frame.x, paper_height - frame.baseline_y);
2681 // See the block-frame call site's identical comment — `annot.satyh`'s
2682 // `\href` fires `register-link-to-uri` from exactly this closure.
2683 interp.current_deco_id = Some(frame.id);
2684 let gr = primitives::apply_deco(
2685 interp,
2686 deco_version,
2687 deco,
2688 pt,
2689 width,
2690 frame.height,
2691 frame.depth,
2692 )?;
2693 interp.current_deco_id = None;
2694 interp.page_graphics[page].extend(gr);
2695 Ok(())
2696}
2697
2698/// One walk over a document's placed geometry, page by page, firing every
2699/// `hook-page-break` closure and every frame decoration with the final
2700/// coordinates the backend gave them.
2701///
2702/// This is the port's callback architecture: `make_hook` +
2703/// `handlePdf.ml:234/337`'s invocation (hooks) and `EvHorzFrame`/`EvVertFrame`
2704/// (decos), relocated to the one place that legally holds `&mut Interp` — the
2705/// backend produced the geometry (POD `HookId`/`DecoId` tokens riding inside
2706/// placed boxes, per `hbox.rs`); this reads them back and re-enters the
2707/// evaluator.
2708///
2709/// The state here is exactly what has to survive a page boundary, which is why
2710/// it is a struct rather than four locals: a `block-frame-breakable` whose
2711/// `FrameStart` and `FrameEnd` straddle a page break stays in `open` between
2712/// pages so its head/middle fragments fire at each boundary and its tail fires
2713/// when the `FrameEnd` finally arrives; inline frames persist across LINES the
2714/// same way, and across pages too (the line a frame continues onto can be the
2715/// first line of the next page).
2716///
2717/// Two callers drive it, and between them they cover one document exactly once
2718/// — see [`eval::FirePass`] for why the walk is split in half:
2719///
2720/// - `primitives::page_break_core`, per COLUMN, in upstream's own position
2721/// (`pageBreak.ml:748`'s `add_column_to_page`, before `columnendhookf` and
2722/// before this page's `pagepartsf`), running [`FirePass::HooksOnly`];
2723/// - [`fire_hooks`], over the finished `DocumentValue`, running
2724/// [`FirePass::DecosOnly`] — or [`FirePass::All`] when nothing fired the
2725/// hooks already, which is how a hand-built document behaves.
2726#[derive(Default)]
2727struct PlacedWalk {
2728 /// Document-order sequence number handed to each block frame as it opens,
2729 /// so the fragments fired on one page can be re-sorted into open order
2730 /// before they reach the page's underlay.
2731 next_open_seq: usize,
2732 open: Vec<OpenFrame>,
2733 open_inline: Vec<OpenInlineFrame>,
2734 /// (open_seq, graphics) per block-frame fragment fired on the CURRENT
2735 /// page, sorted by open order before being appended to the page's
2736 /// underlay — see the doc comment on the ordering this preserves.
2737 closings: Vec<(usize, Vec<GraphicsElem>)>,
2738}
2739
2740impl PlacedWalk {
2741 /// Start a page: frames carried over from a previous page start a fresh
2742 /// per-page extent, so their fragment on THIS page spans only this page's
2743 /// lines.
2744 fn begin_page(&mut self) {
2745 for f in &mut self.open {
2746 f.top = None;
2747 f.bottom = None;
2748 }
2749 self.closings.clear();
2750 }
2751
2752 /// Walk one contiguous run of this page's placed lines. `body` is what
2753 /// `Page::body_lines` encodes: the header and footer are appended after
2754 /// the columns and belong to their own line-break runs, so a frame
2755 /// straddling the body/header boundary must not paint across them.
2756 fn lines(
2757 &mut self,
2758 interp: &mut eval::Interp,
2759 paper_height: Length,
2760 page: usize,
2761 lines: &[rustyfi_backend::PlacedLine],
2762 body: bool,
2763 ) -> Result<(), eval::EvalError> {
2764 let page_number = (page + 1) as i64; // 1-based, = pbinfo#page-number
2765 for line in lines {
2766 // An inline frame that was still open when the previous line ended
2767 // continues here: re-anchor it to THIS line's left edge and
2768 // baseline, so its next fragment measures from where it resumes.
2769 if body {
2770 for f in &mut self.open_inline {
2771 f.x = line.x;
2772 f.baseline_y = line.baseline_y;
2773 }
2774 }
2775 for (dx, bx) in &line.contents {
2776 match bx {
2777 PureHorzBox::HookPageBreak { id } => {
2778 fire_page_break_hook(
2779 interp,
2780 paper_height,
2781 page_number,
2782 line.x + *dx,
2783 line.baseline_y,
2784 *id,
2785 )?;
2786 }
2787 // `Tabular`/`Graphics` for the same reason `Frame` is here:
2788 // a cell's boxes, and the inline run a `draw-text` carries,
2789 // never appear in the page flow, so a `\href` or a
2790 // `hook-page-break` inside one is only reachable through
2791 // this recursion.
2792 PureHorzBox::Frame { .. }
2793 | PureHorzBox::Tabular(_)
2794 | PureHorzBox::Graphics { .. } => {
2795 fire_inline_frame(
2796 interp,
2797 paper_height,
2798 page,
2799 line.x + *dx,
2800 line.baseline_y,
2801 bx,
2802 )?;
2803 }
2804 PureHorzBox::InlineFrameMarker {
2805 id,
2806 end: false,
2807 height,
2808 depth,
2809 } => {
2810 self.open_inline.push(OpenInlineFrame {
2811 id: *id,
2812 x: line.x + *dx,
2813 baseline_y: line.baseline_y,
2814 height: *height,
2815 depth: *depth,
2816 carried: false,
2817 });
2818 }
2819 PureHorzBox::InlineFrameMarker { id, end: true, .. } => {
2820 // Innermost still-open frame with this id (well-nested
2821 // by construction — `prim_inline_frame_breakable`
2822 // always splices a matched pair around its own
2823 // contents). Closing on the line it opened on is an
2824 // UNBROKEN frame: `decoS`. Closing on a later line
2825 // makes this the last of several fragments: `decoT`.
2826 if let Some(pos) = self.open_inline.iter().rposition(|f| f.id == *id) {
2827 let frame = self.open_inline.remove(pos);
2828 let deco_idx = if frame.carried { 3 } else { 0 };
2829 fire_inline_frame_fragment(
2830 interp,
2831 paper_height,
2832 page,
2833 &frame,
2834 line.x + *dx,
2835 deco_idx,
2836 )?;
2837 }
2838 }
2839 PureHorzBox::FrameMarker { id, end: false } => {
2840 self.open.push(OpenFrame {
2841 id: *id,
2842 x: line.x + *dx,
2843 marker_baseline: line.baseline_y,
2844 top: None,
2845 bottom: None,
2846 open_seq: self.next_open_seq,
2847 carried: false,
2848 });
2849 self.next_open_seq += 1;
2850 }
2851 PureHorzBox::FrameMarker { id, end: true } => {
2852 // Close the innermost still-open frame with this id
2853 // (well-nested by construction: `prim_block_frame_
2854 // breakable` always emits a matched Start/End pair
2855 // around its own contents). A frame carried over from
2856 // an earlier page fires its TAIL fragment (`decoT`,
2857 // bottom pad only); one that opened on this page fires
2858 // the single-fragment `decoS` (both pads).
2859 if let Some(pos) = self.open.iter().rposition(|f| f.id == *id) {
2860 let frame = self.open.remove(pos);
2861 // EVERY fragment carries the top pad, not just the
2862 // first: `chop_page` re-applies a still-open
2863 // frame's `paddingT` at the top of each
2864 // continuation page (upstream `pageBreak.ml:322`),
2865 // so the pad is real space on this page too and the
2866 // rect has to cover it — upstream's
2867 // `handlePdf.ml:325-330` spans the rect from the
2868 // fragment's `ypos`, above the `-% paddingT` shift
2869 // it lays the contents out at.
2870 let deco_idx = if frame.carried { 3 } else { 0 };
2871 let incl_top = true;
2872 let gr = fire_block_frame_fragment(
2873 interp,
2874 paper_height,
2875 &frame,
2876 deco_idx,
2877 incl_top,
2878 true,
2879 )?;
2880 self.closings.push((frame.open_seq, gr));
2881 }
2882 // An End with no matching open frame can't happen given
2883 // well-nested markers; ignored if it did.
2884 }
2885 PureHorzBox::EmbeddedBlock {
2886 block, anchor_last, ..
2887 } => {
2888 // A `block-frame-breakable` can also hide INSIDE an
2889 // `embed-block-breakable` (figbox's inline
2890 // `\fig-on-right`/`\fig-on-left`, which draw their image
2891 // from the frame's deco): its `FrameStart`/`FrameEnd`
2892 // markers live in this atomic box's own placed lines, not
2893 // the page flow, so the walk above never sees them. Fire
2894 // those nested decos with absolute coordinates.
2895 fire_embedded_block_frames(
2896 interp,
2897 paper_height,
2898 page,
2899 line.x + *dx,
2900 line.baseline_y,
2901 block,
2902 *anchor_last,
2903 &mut self.next_open_seq,
2904 &mut self.closings,
2905 )?;
2906 }
2907 _ => {}
2908 }
2909 }
2910 // Every REAL line (one with non-marker content) extends every
2911 // currently-open frame's (top, bottom) — pad Skips don't create
2912 // a `PlacedLine` at all, so they're naturally excluded here; the
2913 // ±pad compensation happens once, at close time, above.
2914 //
2915 // BODY lines only (`Page::body_lines`). The header and footer are
2916 // appended after the columns, and a frame carried across a page
2917 // boundary is open for this whole walk, so counting them stretched
2918 // every such frame's fragment from the header baseline to the
2919 // footer — easytable's `+code` blocks painted their grey background
2920 // over entire pages (4, 11, 12) instead of over their own lines.
2921 if body {
2922 if let Some((height, depth)) = placed_line_extent(line) {
2923 let top = line.baseline_y - height;
2924 let bottom = line.baseline_y + depth;
2925 for f in &mut self.open {
2926 f.top = Some(f.top.map_or(top, |t| t.min(top)));
2927 f.bottom = Some(f.bottom.map_or(bottom, |b| b.max(bottom)));
2928 }
2929 }
2930 }
2931 // An inline frame still open at the end of a line really did split
2932 // (upstream `append_framed_lines`' non-final `PureLine` arms): fire
2933 // this line's fragment — `decoH` for the first, `decoM` for every
2934 // later one. The frame stays open; the re-anchor at the top of the
2935 // next body line's turn moves it on.
2936 if body && !self.open_inline.is_empty() {
2937 let right = placed_line_right_edge(line);
2938 let pending: Vec<OpenInlineFrame> = self.open_inline.clone();
2939 for frame in &pending {
2940 let deco_idx = if frame.carried { 2 } else { 1 };
2941 fire_inline_frame_fragment(
2942 interp,
2943 paper_height,
2944 page,
2945 frame,
2946 right,
2947 deco_idx,
2948 )?;
2949 }
2950 for f in &mut self.open_inline {
2951 f.carried = true;
2952 }
2953 }
2954 }
2955 Ok(())
2956 }
2957
2958 /// Close page `page`: fire the fragments of every frame still open, then
2959 /// append this page's accumulated deco graphics to its underlay.
2960 fn end_page(
2961 &mut self,
2962 interp: &mut eval::Interp,
2963 paper_height: Length,
2964 page: usize,
2965 ) -> Result<(), eval::EvalError> {
2966 // Frames still open at page end straddle the following page break: fire
2967 // this page's fragment — a HEAD (`decoH`, top pad only) the first time a
2968 // frame spans, a MIDDLE (`decoM`, no pads) on every later page — and
2969 // keep the frame open so its remaining fragments (and eventual `decoT`)
2970 // fire on the pages ahead. A frame that never accumulated a real line
2971 // on this page (top/bottom still `None`) contributes nothing and does
2972 // NOT advance its fragment state: it stays `carried` as it was, so a
2973 // frame that opened at the very bottom of a page (no room for a line)
2974 // still fires its HEAD (or, if it also closes with content on a single
2975 // later page, a `decoS`) on the first page that actually holds its
2976 // content. Collect fires first (can't hold `&open` across the `&mut
2977 // interp` deco call), then mark exactly the frames that fired.
2978 let mut page_end_fires: Vec<(usize, Vec<GraphicsElem>)> = Vec::new();
2979 let mut fired_seqs: Vec<usize> = Vec::new();
2980 for frame in &self.open {
2981 if frame.top.is_none() && frame.bottom.is_none() {
2982 continue;
2983 }
2984 // Top pad on every fragment — see the `FrameMarker { end: true }`
2985 // arm above for why a carried fragment has one too.
2986 let deco_idx = if frame.carried { 2 } else { 1 };
2987 let gr = fire_block_frame_fragment(interp, paper_height, frame, deco_idx, true, false)?;
2988 page_end_fires.push((frame.open_seq, gr));
2989 fired_seqs.push(frame.open_seq);
2990 }
2991 for f in &mut self.open {
2992 if fired_seqs.contains(&f.open_seq) {
2993 f.carried = true;
2994 }
2995 }
2996 self.closings.extend(page_end_fires);
2997
2998 self.closings.sort_by_key(|(seq, _)| *seq);
2999 // `get_mut` rather than `[page]`: the HOOKS-ONLY pass runs while the
3000 // document is still being assembled, so `page_graphics` has no slot
3001 // for this page yet — and it needs none, since every fragment fire it
3002 // just made returned nothing. `fire_hooks` sizes the table up front.
3003 for (_, gr) in std::mem::take(&mut self.closings) {
3004 if let Some(slot) = interp.page_graphics.get_mut(page) {
3005 slot.extend(gr);
3006 }
3007 }
3008 Ok(())
3009 }
3010}
3011
3012/// Fire every placed decoration — and, for a document this crate did not page-
3013/// break itself, every placed page-break hook too — now that final page
3014/// numbers and points are known.
3015///
3016/// Sets `interp.current_page` to `Some(i)` for the duration of page `i`'s
3017/// walk (the "during page break" window: `register-destination`/
3018/// `register-link-to-*` — called directly by a hook or, more commonly,
3019/// transitively by a fired deco closure, e.g. `annot.satyh`'s `\href` —
3020/// only succeed inside this window) and back to `None` once every page is
3021/// done.
3022///
3023/// Which half of the walk runs here depends on
3024/// `Interp::page_break_hooks_fired`: after `page_break_core` has driven its
3025/// own per-page [`eval::FirePass::HooksOnly`] pass this is the
3026/// [`eval::FirePass::DecosOnly`] remainder, and for a `DocumentValue` built by
3027/// hand it is the undivided [`eval::FirePass::All`].
3028///
3029/// **Known scope cuts** (documented deviations):
3030/// - Frames nested inside a `Tabular` cell or an `EmbeddedBlock`'s stacked
3031/// lines are NOT discovered by this walk — their placed positions would
3032/// need the writers' cell/stack arithmetic replicated lang-side. No
3033/// bundled package puts an `\href`/frame inside one today.
3034/// - A `block-frame-breakable` frame whose `FrameStart` and `FrameEnd` land
3035/// on DIFFERENT pages now fires per-page fragments: `decoS` if it opens and
3036/// closes on one page, else `decoH` on its first (opening) page, `decoM` on
3037/// each fully-contained middle page, and `decoT` on its closing page — the
3038/// `pageBreak.ml` fragment split. Each fragment's rect spans only that
3039/// page's content extent, plus the frame's TOP pad (which `chop_page`
3040/// re-applies on every continuation page, `pageBreak.ml:322`) and, on the
3041/// tail/single fragment, the bottom pad. This is
3042/// what lets `figbox`'s `+fig-on-right`/`+fig-on-left` (which draw their
3043/// image in `decoH`) render on figures whose surrounding text wraps across a
3044/// page break.
3045///
3046/// `pub` (rather than crate-private) so unit tests can drive it directly
3047/// against a hand-built `DocumentValue`, without going through a full
3048/// `compile_document_cst` fixpoint.
3049pub fn fire_hooks(interp: &mut eval::Interp, doc: &DocumentValue) -> Result<(), eval::EvalError> {
3050 interp.page_graphics = doc.pages.iter().map(|_| Vec::new()).collect();
3051 let paper_height = doc.geometry.paper_height;
3052 interp.fire_pass = if interp.page_break_hooks_fired {
3053 eval::FirePass::DecosOnly
3054 } else {
3055 eval::FirePass::All
3056 };
3057 let mut walk = PlacedWalk::default();
3058 let result = (|| {
3059 for (i, page) in doc.pages.iter().enumerate() {
3060 interp.current_page = Some(i);
3061 walk.begin_page();
3062 let split = page.body_lines.min(page.lines.len());
3063 walk.lines(interp, paper_height, i, &page.lines[..split], true)?;
3064 walk.lines(interp, paper_height, i, &page.lines[split..], false)?;
3065 walk.end_page(interp, paper_height, i)?;
3066 }
3067 Ok(())
3068 })();
3069 interp.fire_pass = eval::FirePass::All;
3070 interp.current_page = None;
3071 result
3072}
3073
3074/// Fire one placed `block-frame-breakable` fragment's decoration (`decoS`/
3075/// `decoH`/`decoM`/`decoT` picked by `deco_idx` — 0/1/2/3, evalUtil.ml:169
3076/// `get_decoset` order) with its final geometry, returning the graphics it
3077/// draws (absolute page coordinates). `incl_top_pad`/`incl_bot_pad` select
3078/// which of the frame's `pads.t`/`pads.b` this fragment carries: the single
3079/// (`decoS`) fragment carries both, the head only the top, the tail only the
3080/// bottom, and a middle neither — matching `pageBreak.ml`'s per-fragment
3081/// padding. The rect spans the frame's accumulated (top, bottom) extent on the
3082/// current page; an empty frame (no real line between Start/End) falls back to
3083/// the Start marker's own baseline for a degenerate zero-height rect.
3084fn fire_block_frame_fragment(
3085 interp: &mut eval::Interp,
3086 paper_height: Length,
3087 frame: &OpenFrame,
3088 deco_idx: usize,
3089 incl_top_pad: bool,
3090 incl_bot_pad: bool,
3091) -> Result<Vec<GraphicsElem>, eval::EvalError> {
3092 // A decoration, not a hook: the hooks-only half of the walk skips it, and
3093 // the empty result leaves this page's underlay untouched.
3094 if interp.fire_pass == eval::FirePass::HooksOnly {
3095 return Ok(Vec::new());
3096 }
3097 let (pads, width, deco, deco_version) = match &interp.decos[frame.id.0] {
3098 eval::DecoEntry::Block {
3099 pads,
3100 width,
3101 decoset,
3102 version,
3103 } => (*pads, *width, decoset[deco_idx].clone(), *version),
3104 eval::DecoEntry::Inline { .. } | eval::DecoEntry::InlineBreakable { .. } => {
3105 return eval::eval_error("BUG: inline deco behind a block-frame marker")
3106 }
3107 };
3108 let top = frame.top.unwrap_or(frame.marker_baseline);
3109 let bottom = frame.bottom.unwrap_or(frame.marker_baseline);
3110 let frame_top = if incl_top_pad { top - pads.t } else { top };
3111 let frame_bottom = if incl_bot_pad {
3112 bottom + pads.b
3113 } else {
3114 bottom
3115 };
3116 let pt = (frame.x, paper_height - frame_bottom);
3117 // Record which DecoId is firing so a `register-destination` call
3118 // inside the deco (annot.satyh's `register-location-frame`) can tag
3119 // itself with it — see `Interp::current_deco_id`'s doc comment.
3120 interp.current_deco_id = Some(frame.id);
3121 let height = frame_bottom - frame_top;
3122 let gr = primitives::apply_deco(interp, deco_version, deco, pt, width, height, Length::ZERO)?;
3123 interp.current_deco_id = None;
3124 // A WHOLE-frame decoration (`decoS`, index 0), recorded box-local for the
3125 // reflow backend — see `FrameDecoration`. The deco drew at `pt`, so
3126 // shifting by `-pt` puts the origin back at the frame's own bottom-left.
3127 // A frame split across pages fires `decoH`/`decoM`/`decoT` instead and is
3128 // deliberately not recorded: there is no single drawing to scale.
3129 if deco_idx == 0 && !gr.is_empty() {
3130 let back = (Length::ZERO - pt.0, Length::ZERO - pt.1);
3131 interp.frame_decos.push((
3132 frame.id,
3133 rustyfi_backend::FrameDecoration {
3134 width,
3135 height,
3136 pads: (pads.l, pads.r, pads.t, pads.b),
3137 elems: gr.iter().map(|e| shift_graphics(back, e)).collect(),
3138 },
3139 ));
3140 }
3141 Ok(gr)
3142}
3143
3144/// Fire block-frame decorations that live INSIDE an `EmbeddedBlock` inline box.
3145///
3146/// figbox's inline `\fig-on-right`/`\fig-on-left` wrap a `block-frame-breakable`
3147/// (whose deco draws the figure image) in `embed-block-breakable`, so the
3148/// frame's `FrameStart`/`FrameEnd` markers end up in the embedded block's OWN
3149/// placed lines, never the page flow that [`fire_hooks`] walks — the image
3150/// would silently never render. Replicate `place_embedded_block`'s transform
3151/// (`rustyfi-pdf`): `place_block_at` from a zero origin, then shift so the
3152/// anchor line (first for top-anchor, last for bottom) sits at the box's inline
3153/// baseline. Converting that writer y-up geometry back to page y-down, inner
3154/// line `i`'s absolute baseline is `baseline_ydown + (bl_i - anchor_offset)` and
3155/// its x is `tx + line.x + dx`. Over those absolute lines we run the same
3156/// frame-open/close tracking and `fire_block_frame_fragment` as the main walk.
3157/// The box is atomic (one inline box, not page-broken), so every nested frame
3158/// opens and closes within it and fires a single-fragment `decoS`. Nested
3159/// `EmbeddedBlock`s (a figbox inside a figbox) recurse.
3160#[allow(clippy::too_many_arguments)]
3161fn fire_embedded_block_frames(
3162 interp: &mut eval::Interp,
3163 paper_height: Length,
3164 page: usize,
3165 tx: Length,
3166 baseline_ydown: Length,
3167 block: &[VertBox],
3168 anchor_last: bool,
3169 next_open_seq: &mut usize,
3170 out: &mut Vec<(usize, Vec<GraphicsElem>)>,
3171) -> Result<(), eval::EvalError> {
3172 let placed = place_block_at((Length::ZERO, Length::ZERO), block.to_vec());
3173 let anchor = if anchor_last {
3174 placed.last()
3175 } else {
3176 placed.first()
3177 };
3178 let Some(anchor) = anchor else {
3179 return Ok(());
3180 };
3181 let anchor_offset = anchor.baseline_y;
3182 let mut open: Vec<OpenFrame> = Vec::new();
3183 // `inline-frame-breakable` inside an embedded block, same story as the
3184 // `PureHorzBox::Frame` arm below: latexcmds' `\fbox`/`\doublebox`/
3185 // `\ovalbox`/`\shadowbox` all go through the BREAKABLE primitive, and a
3186 // `+listing` item's lines live here rather than in the page flow.
3187 let mut open_inline: Vec<OpenInlineFrame> = Vec::new();
3188 for pl in &placed {
3189 let abs_baseline = baseline_ydown + (pl.baseline_y - anchor_offset);
3190 for f in &mut open_inline {
3191 f.x = tx + pl.x;
3192 f.baseline_y = abs_baseline;
3193 }
3194 for (dx, bx) in &pl.contents {
3195 match bx {
3196 PureHorzBox::InlineFrameMarker {
3197 id,
3198 end: false,
3199 height,
3200 depth,
3201 } => {
3202 open_inline.push(OpenInlineFrame {
3203 id: *id,
3204 x: tx + pl.x + *dx,
3205 baseline_y: abs_baseline,
3206 height: *height,
3207 depth: *depth,
3208 carried: false,
3209 });
3210 }
3211 PureHorzBox::InlineFrameMarker { id, end: true, .. } => {
3212 if let Some(pos) = open_inline.iter().rposition(|f| f.id == *id) {
3213 let frame = open_inline.remove(pos);
3214 let deco_idx = if frame.carried { 3 } else { 0 };
3215 fire_inline_frame_fragment(
3216 interp,
3217 paper_height,
3218 page,
3219 &frame,
3220 tx + pl.x + *dx,
3221 deco_idx,
3222 )?;
3223 }
3224 }
3225 PureHorzBox::FrameMarker { id, end: false } => {
3226 open.push(OpenFrame {
3227 id: *id,
3228 x: tx + pl.x + *dx,
3229 marker_baseline: abs_baseline,
3230 top: None,
3231 bottom: None,
3232 open_seq: *next_open_seq,
3233 carried: false,
3234 });
3235 *next_open_seq += 1;
3236 }
3237 PureHorzBox::FrameMarker { id, end: true } => {
3238 if let Some(pos) = open.iter().rposition(|f| f.id == *id) {
3239 let frame = open.remove(pos);
3240 let gr = fire_block_frame_fragment(interp, paper_height, &frame, 0, true, true)?;
3241 out.push((frame.open_seq, gr));
3242 }
3243 }
3244 // An INLINE frame (`inline-frame-outer`/`-inner`/`-breakable`)
3245 // can hide in here too — latexcmds' `\fbox`/`\doublebox`/
3246 // `\ovalbox`/`\shadowbox` used inside `+listing` items, whose
3247 // lines live in an embedded block rather than the page flow
3248 // (26 of 144 inline frames in one document went undrawn
3249 // without this). `Tabular`/`Graphics` too — see the identical
3250 // arm in `fire_hooks`.
3251 PureHorzBox::Frame { .. }
3252 | PureHorzBox::Tabular(_)
3253 | PureHorzBox::Graphics { .. } => {
3254 fire_inline_frame(interp, paper_height, page, tx + pl.x + *dx, abs_baseline, bx)?;
3255 }
3256 PureHorzBox::EmbeddedBlock {
3257 block: inner,
3258 anchor_last: al,
3259 ..
3260 } => {
3261 fire_embedded_block_frames(
3262 interp,
3263 paper_height,
3264 page,
3265 tx + pl.x + *dx,
3266 abs_baseline,
3267 inner,
3268 *al,
3269 next_open_seq,
3270 out,
3271 )?;
3272 }
3273 _ => {}
3274 }
3275 }
3276 if let Some((height, depth)) = placed_line_extent(pl) {
3277 let top = abs_baseline - height;
3278 let bottom = abs_baseline + depth;
3279 for f in &mut open {
3280 f.top = Some(f.top.map_or(top, |t| t.min(top)));
3281 f.bottom = Some(f.bottom.map_or(bottom, |b| b.max(bottom)));
3282 }
3283 }
3284 if !open_inline.is_empty() {
3285 let right = tx + placed_line_right_edge(pl);
3286 let pending: Vec<OpenInlineFrame> = open_inline.clone();
3287 for frame in &pending {
3288 let deco_idx = if frame.carried { 2 } else { 1 };
3289 fire_inline_frame_fragment(interp, paper_height, page, frame, right, deco_idx)?;
3290 }
3291 for f in &mut open_inline {
3292 f.carried = true;
3293 }
3294 }
3295 }
3296 Ok(())
3297}
3298
3299/// Apply one `hook-page-break` closure to `(pbinfo, point)`.
3300///
3301/// Extracted so the walk can fire a hook wherever it is found, not only at
3302/// the top level of a placed line: `stdja`'s `+section` appends its
3303/// `hook-page-break` to the heading's inline boxes, wrapped in the title
3304/// deco's inline FRAME, so all 7 of this manual's hooks sat one level down
3305/// and none fired without this. `stdja.satyh:448` registers `<label>:page`
3306/// from inside this closure, so an unfired hook left `get-cross-reference`
3307/// rendering `?`: 11 such `?` for easytable and 21 for enumitem on pages
3308/// 1-2 alone, where SATySFi emits none.
3309fn fire_page_break_hook(
3310 interp: &mut eval::Interp,
3311 paper_height: Length,
3312 page_number: i64,
3313 x: Length,
3314 baseline_y: Length,
3315 id: rustyfi_backend::HookId,
3316) -> Result<(), eval::EvalError> {
3317 // The decos-only half of the walk retraces the same lines a moment after
3318 // `page_break_core` already ran every hook on them, in upstream's own
3319 // position — firing them again would run each closure twice.
3320 if interp.fire_pass == eval::FirePass::DecosOnly {
3321 return Ok(());
3322 }
3323 let closure = interp.hooks[id.0].clone();
3324 let mut fields = BTreeMap::new();
3325 fields.insert("page-number".to_string(), Value::Int(page_number));
3326 let pbinfo = Value::Record(fields);
3327 // PDF page space is y-up; placed geometry (`baseline_y`) is page space
3328 // y-down from the paper top — the same flip the writers apply.
3329 let point = Value::Tuple(vec![
3330 Value::Length(x),
3331 Value::Length(paper_height - baseline_y),
3332 ]);
3333 let applied = interp.apply(closure, pbinfo)?;
3334 match interp.apply(applied, point)? {
3335 Value::Unit => Ok(()),
3336 other => eval::eval_error(format!(
3337 "hook-page-break closure returned {}, expected unit",
3338 other.type_name()
3339 )),
3340 }
3341}
3342
3343/// Fire one placed inline frame's deco (and any frames nested in its
3344/// contents) with its final geometry — the port of `EvHorzFrame`'s
3345/// `deco (xpos, yposbaseline) wid hgt dpt` (handlePdf.ml:123-129), point
3346/// pre-flipped to PDF y-up exactly like the hook point above. The returned
3347/// `graphics list` (absolute page coordinates, `make_frame_deco`'s contract)
3348/// is accumulated onto this page's underlay.
3349///
3350/// `interp.current_page` is already `Some(page)` here (set by `fire_hooks`'s
3351/// caller), so a deco body calling `register-link-to-uri` (exactly
3352/// `annot.satyh:11-14`) lands its `Annot` on the right page — this is the
3353/// entire `\href` unlock.
3354///
3355/// Also the entry point for firing whatever is nested INSIDE a placed box:
3356/// besides an inline frame's own contents this descends into a
3357/// `Tabular`'s cells, since a cell's boxes never reach the page flow that
3358/// [`fire_hooks`] walks. Everything a cell can carry — a `\href`, a `\ref`,
3359/// a `hook-page-break` — was silently inert before: a `\href` in an
3360/// easytable cell produced no `/Link` annotation at all, which is 3 of
3361/// slydifi's 4 links. `bx` that is neither a frame nor a tabular is a no-op,
3362/// so callers can hand it every box on a line.
3363fn fire_inline_frame(
3364 interp: &mut eval::Interp,
3365 paper_height: Length,
3366 page: usize,
3367 x: Length,
3368 baseline_y: Length,
3369 bx: &PureHorzBox,
3370) -> Result<(), eval::EvalError> {
3371 let contents = match bx {
3372 PureHorzBox::Frame {
3373 width,
3374 height,
3375 depth,
3376 deco,
3377 contents,
3378 } => {
3379 // The frame's own decoration is skipped by the hooks-only half of
3380 // the walk, but its CONTENTS are still descended into below: a
3381 // `hook-page-break` one level down (`stdja`'s `+section` puts one
3382 // inside the title deco's frame) is reachable nowhere else.
3383 if interp.fire_pass != eval::FirePass::HooksOnly {
3384 let (deco_v, deco_version) = match &interp.decos[deco.0] {
3385 eval::DecoEntry::Inline { deco, version } => (deco.clone(), *version),
3386 eval::DecoEntry::Block { .. } | eval::DecoEntry::InlineBreakable { .. } => {
3387 return eval::eval_error("BUG: block deco behind an inline frame")
3388 }
3389 };
3390 let pt = (x, paper_height - baseline_y);
3391 // See the block-frame call site's identical comment —
3392 // `annot.satyh`'s `\href` fires `register-link-to-uri` from
3393 // exactly this closure.
3394 interp.current_deco_id = Some(*deco);
3395 let gr = primitives::apply_deco(
3396 interp,
3397 deco_version,
3398 deco_v,
3399 pt,
3400 *width,
3401 *height,
3402 *depth,
3403 )?;
3404 interp.current_deco_id = None;
3405 interp.page_graphics[page].extend(gr);
3406 }
3407 contents
3408 }
3409 PureHorzBox::Tabular(tab) => {
3410 // Each cell is its own placed run on its own baseline. The
3411 // writers' convention (`rustyfi-pdf`'s `emit_box`, `ty +
3412 // cell.baseline_y` in PDF y-UP space) means `cell.baseline_y` is
3413 // measured upward from the tabular box's own baseline, so in this
3414 // walk's y-DOWN page coordinates it subtracts.
3415 for cell in &tab.cells {
3416 fire_nested_in_contents(
3417 interp,
3418 paper_height,
3419 page,
3420 x + cell.x,
3421 baseline_y - cell.baseline_y,
3422 &cell.contents,
3423 )?;
3424 }
3425 return Ok(());
3426 }
3427 PureHorzBox::Graphics {
3428 elems,
3429 origin_independent,
3430 ..
3431 } => {
3432 // `draw-text` runs (`GraphicsElem::Text`) carry real inline boxes,
3433 // and figbox's `textbox` puts whole tables in one — slydifi's
3434 // theme table reaches its `\link`s only through here. Element
3435 // coordinates are box-local PDF y-up from the box's placed anchor,
3436 // except for an `origin_independent` box whose callback already
3437 // produced page-absolute ones: exactly `rustyfi-pdf`'s own
3438 // `(ax, ay)` choice, kept in step with it.
3439 let anchor_y = if *origin_independent {
3440 paper_height
3441 } else {
3442 baseline_y
3443 };
3444 let anchor_x = if *origin_independent { Length::ZERO } else { x };
3445 fire_nested_in_graphics(interp, paper_height, page, anchor_x, anchor_y, elems)?;
3446 return Ok(());
3447 }
3448 _ => return Ok(()),
3449 };
3450 fire_nested_in_contents(interp, paper_height, page, x, baseline_y, contents)
3451}
3452
3453/// Fire every hook, decoration and deferred destination carried by a graphics
3454/// box's elements — the inline runs a `draw-text` (`GraphicsElem::Text`) holds,
3455/// and the `GraphicsElem::Destination` markers an `inline-graphics` callback
3456/// left behind — recursing through `Group`/`Clip`. `anchor_x`/`anchor_y` are
3457/// the box's placed origin in this walk's (x, y-DOWN) page coordinates.
3458///
3459/// A `Text` element's own `transform` (from `rotate-graphics`/
3460/// `scale-graphics`) is deliberately NOT applied: a decoration's rect is an
3461/// axis-aligned rectangle, so a rotated run has no faithful rect to report,
3462/// and firing at the untransformed anchor at least puts a `\href`'s link
3463/// where the run starts rather than nowhere at all.
3464fn fire_nested_in_graphics(
3465 interp: &mut eval::Interp,
3466 paper_height: Length,
3467 page: usize,
3468 anchor_x: Length,
3469 anchor_y: Length,
3470 elems: &[GraphicsElem],
3471) -> Result<(), eval::EvalError> {
3472 for elem in elems {
3473 match elem {
3474 GraphicsElem::Text { pt, contents, .. } => {
3475 // `pt` is PDF y-UP relative to the anchor; this walk is y-down.
3476 fire_nested_in_contents(
3477 interp,
3478 paper_height,
3479 page,
3480 anchor_x + pt.0,
3481 anchor_y - pt.1,
3482 contents,
3483 )?;
3484 }
3485 GraphicsElem::Group(inner) | GraphicsElem::Clip(_, inner) => {
3486 fire_nested_in_graphics(interp, paper_height, page, anchor_x, anchor_y, inner)?;
3487 }
3488 // `pt` is y-UP from the anchor, as a `Text`'s is, and a
3489 // `NamedDest`'s `y` is y-up too, so lifting the anchor out of this
3490 // walk's y-DOWN frame is the only arithmetic needed. An
3491 // `origin_independent` box arrives with `anchor_x = 0`/`anchor_y =
3492 // paper_height` (see the caller), making that the identity — the
3493 // same page-absolute reading its ink gets from the writers.
3494 GraphicsElem::Destination { key, pt } => {
3495 // Placement bookkeeping, not a hook — recorded once, by the
3496 // half of the walk that also fires the decorations.
3497 if interp.fire_pass == eval::FirePass::HooksOnly {
3498 continue;
3499 }
3500 let name = interp.dest_name(key);
3501 let x = anchor_x + pt.0;
3502 let y = paper_height - anchor_y + pt.1;
3503 // As in a direct call: the reflow backend resolves a
3504 // destination to its Frame through this id.
3505 if let Some(deco_id) = interp.current_deco_id {
3506 interp.dest_decos.push((deco_id, name.clone()));
3507 }
3508 interp.destinations.push(rustyfi_backend::NamedDest {
3509 page,
3510 name,
3511 x,
3512 y,
3513 });
3514 }
3515 GraphicsElem::Fill(..) | GraphicsElem::Stroke(..) | GraphicsElem::DashedStroke(..) => {}
3516 }
3517 }
3518 Ok(())
3519}
3520
3521/// Fire every hook and decoration inside one placed content run — an inline
3522/// frame's contents or a tabular cell's — with `x0`/`baseline_y` as the run's
3523/// own absolute origin.
3524///
3525/// An `inline-frame-breakable` reached through here is spliced into the run as
3526/// a marker pair (see `prim_inline_frame_breakable`). Such a run is a single
3527/// `fit_cell` line, so the frame is always unbroken and always fires `decoS` —
3528/// which is also what upstream does in this position, since a breakable frame
3529/// reached through a *pure* box degrades to an atomic `LBOuterFrame`
3530/// (`convert_list_for_line_breaking_pure`, lineBreak.ml:335).
3531fn fire_nested_in_contents(
3532 interp: &mut eval::Interp,
3533 paper_height: Length,
3534 page: usize,
3535 x0: Length,
3536 baseline_y: Length,
3537 contents: &[(Length, PureHorzBox)],
3538) -> Result<(), eval::EvalError> {
3539 let mut open_inline: Vec<OpenInlineFrame> = Vec::new();
3540 for (dx, child) in contents {
3541 // A `hook-page-break` can sit INSIDE the frame — `stdja`'s `+section`
3542 // appends one to a heading that the title deco then wraps in a frame —
3543 // and the top-level walk never sees it. See `fire_page_break_hook`.
3544 if let PureHorzBox::HookPageBreak { id } = child {
3545 fire_page_break_hook(interp, paper_height, (page + 1) as i64, x0 + *dx, baseline_y, *id)?;
3546 }
3547 match child {
3548 PureHorzBox::InlineFrameMarker {
3549 id,
3550 end: false,
3551 height,
3552 depth,
3553 } => open_inline.push(OpenInlineFrame {
3554 id: *id,
3555 x: x0 + *dx,
3556 baseline_y,
3557 height: *height,
3558 depth: *depth,
3559 carried: false,
3560 }),
3561 PureHorzBox::InlineFrameMarker { id, end: true, .. } => {
3562 if let Some(pos) = open_inline.iter().rposition(|f| f.id == *id) {
3563 let frame = open_inline.remove(pos);
3564 fire_inline_frame_fragment(interp, paper_height, page, &frame, x0 + *dx, 0)?;
3565 }
3566 }
3567 _ => {}
3568 }
3569 fire_inline_frame(interp, paper_height, page, x0 + *dx, baseline_y, child)?;
3570 }
3571 Ok(())
3572}