twig-sys 2.2.3

FFI bindings and native library for Twig (the Djot/Markdown/HTML/XML document engine). Used by the `twig-doc` crate.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
//! The authoring editor: a `Splicer` that knows how its format is SPELLED.
//!
//! ── The two layers ─────────────────────────────────────────────────────────
//! `Splicer` (`ast/splicer.zig`) is the engine: byte spans in, reparse,
//! rollback, undo. It is language-agnostic by construction and imports no
//! language module — hand it a `parse_fn` and it will edit djot, Markdown or
//! XML with the same code. What it cannot do is decide that bold is spelled
//! `**` here and `*` there.
//!
//! `Editor` is that decision, and nothing else: `Splicer` + a `*const Syntax`.
//! It hosts the gestures a caret editor actually performs — Cmd-B, H1, quote,
//! link — each of which is "consult the table, build the bytes, hand the
//! Splicer one span". The Splicer's invariant survives intact, because `Editor`
//! depends only on the `Syntax` INTERFACE and still names no format: it never
//! learns whether the table it was handed came from djot or Markdown.
//! `format.zig`'s registry is what binds the two.
//!
//! ── Why this exists ────────────────────────────────────────────────────────
//! All of this lived in `c_abi.zig`. Not by design — it accreted there because
//! the C ABI was the first (and only) caller with a caret to serve, and the
//! layer it needed didn't exist. The cost was steep: the knowledge that
//! `mailto:a@b.dev` is a `url` in Markdown but an `email` in djot could only be
//! reached through an `extern` function, could only be tested through a
//! `TwigEditor*` handle and a `TwigStatus` code, and could not be reached by
//! `twig edit` at all.
//!
//! So the C ABI's `TwigEditor` was never `Splicer` — it was always this type,
//! `{ editor, format }`, assembled by hand at the boundary. That is why this
//! module took the `Editor` name and the engine underneath was renamed to what
//! it always was: `TwigEditor` maps to `*Editor`, 1:1, and the ABI's job is
//! back to marshalling.
//!
//! ── Errors ─────────────────────────────────────────────────────────────────
//! Typed, so the ABI's mapping is mechanical and every other caller gets to
//! `switch` on something real. `error.UnsupportedFormat` is uniformly "the
//! `Syntax` table has a `null` where this gesture needed a spelling" — never a
//! hand-written per-format arm. See `syntax.zig`.

const std = @import("std");
const Allocator = std.mem.Allocator;

const AST = @import("ast.zig");
const Span = @import("../span.zig");
const locate = @import("locate.zig");
const syntax_mod = @import("../syntax.zig");

pub const Splicer = @import("splicer.zig").Splicer;

/// Used by the free functions below; the public vocabularies all hang off
/// `Editor` itself (`Editor.InlineKind`, `Editor.Error`, ...).
const Syntax = syntax_mod.Syntax;
const ContainerSpelling = syntax_mod.ContainerSpelling;

/// The `Node.Kind` tag an `InlineKind`/`ContainerKind` parses back as. The
/// vocabularies are named for their kinds, so this is a rename, not a mapping —
/// and it fails to compile rather than silently mis-mapping if one drifts.
fn kindTag(kind: anytype) Splicer.KindTag {
    return switch (kind) {
        inline else => |k| @field(Splicer.KindTag, @tagName(k)),
    };
}

/// Room for the widest marker/indent a list can produce (`999. ` and friends).
const container_indent = " " ** 24;

pub const Editor = struct {
    /// The kind vocabularies, re-exported so a caller needs only this type:
    /// `twig.Editor.InlineKind`. (The `Syntax` type itself is `twig.Syntax`.)
    pub const InlineKind = syntax_mod.InlineKind;
    pub const BlockKind = syntax_mod.BlockKind;
    pub const ContainerKind = syntax_mod.ContainerKind;

    pub const Error = error{
        /// `start > end`, or a range reaching past the source.
        InvalidRange,
        /// A heading level outside 1-6.
        InvalidLevel,
        /// A destination this format cannot hold (one containing a newline).
        InvalidDestination,
        /// The `Syntax` table has no spelling for this gesture in this format.
        UnsupportedFormat,
        /// No block covers the offset/range this gesture needs one for.
        NoBlock,
        /// The target node has no editable span/interior, or the gesture would
        /// corrupt something it refuses to touch.
        NotEditable,
        /// The edit produced a document that no longer parses; it was rolled
        /// back and nothing changed.
        EditConflict,
    } || Allocator.Error;

    splicer: Splicer,
    /// This format's spelling. Borrowed — `format.zig`'s registry entries are
    /// static, so it outlives any editor.
    syntax: *const Syntax,

    /// `parse_ctx`/`parse_fn` are the Splicer's contract (see its doc comment);
    /// `syntax` is the table every gesture below consults. Pair them from
    /// `format.zig`'s registry rather than by hand — an entry's `parseToAst` and
    /// `syntax` are two halves of one language, and crossing them would spell
    /// djot into a Markdown document.
    pub fn init(
        allocator: Allocator,
        source_bytes: []const u8,
        parse_ctx: *const anyopaque,
        parse_fn: Splicer.ParseFn,
        syntax: *const Syntax,
    ) !Editor {
        return .{
            .splicer = try Splicer.init(allocator, source_bytes, parse_ctx, parse_fn),
            .syntax = syntax,
        };
    }

    pub fn deinit(self: *Editor) void {
        self.splicer.deinit();
    }

    pub fn sourceBytes(self: *const Editor) []const u8 {
        return self.splicer.sourceBytes();
    }

    pub fn astView(self: *const Editor) *const AST {
        return self.splicer.astView();
    }

    pub fn lastChange(self: *const Editor) ?Splicer.Change {
        return self.splicer.last_change;
    }

    /// Validate a caller-supplied byte range. `Splicer.replaceAtSpan` ASSERTS on
    /// a bad range — fine for internal callers, but a range from a C caller or a
    /// stale caret is untrusted input, so it is checked into an error here,
    /// once, before any gesture can reach the assert.
    fn checkRange(self: *const Editor, start: usize, end: usize) Error!void {
        if (start > end or end > self.sourceBytes().len) return error.InvalidRange;
    }

    /// Splice rebuilt source in over `[start, end)`. Every gesture ends here.
    fn commitSplice(self: *Editor, start: usize, end: usize, text: []const u8) Error!void {
        self.splicer.replaceAtSpan(Span.init(start, end), text) catch |err| switch (err) {
            error.OutOfMemory => return error.OutOfMemory,
            // Anything else is the parser rejecting the edited document; the
            // splicer has already rolled it back.
            else => return error.EditConflict,
        };
    }

    // ── Inline marks ───────────────────────────────────────────────────────

    /// Wrap `[start, end)` in `kind`'s delimiters — the unconditional half of
    /// the inline toolbar (always adds a mark).
    pub fn wrapRange(self: *Editor, span: Span, kind: InlineKind) Error!void {
        try self.checkRange(span.start, span.end);
        const d = self.syntax.inline_delims.get(kind) orelse return error.UnsupportedFormat;
        self.splicer.wrapRange(span, d.open, d.close) catch |err| switch (err) {
            error.OutOfMemory => return error.OutOfMemory,
            else => return error.EditConflict,
        };
    }

    /// Toggle `kind` over `[start, end)`: strip the mark if the range already
    /// IS a node of `kind` (its whole span or its interior), else wrap it — a
    /// rich editor's Cmd-B.
    pub fn toggleInline(self: *Editor, span: Span, kind: InlineKind) Error!void {
        try self.checkRange(span.start, span.end);
        const d = self.syntax.inline_delims.get(kind) orelse return error.UnsupportedFormat;
        self.splicer.toggleInline(span, kindTag(kind), d.open, d.close) catch |err| switch (err) {
            error.OutOfMemory => return error.OutOfMemory,
            error.NoNodeSpan, error.NoContentSpan => return error.NotEditable,
            else => return error.EditConflict,
        };
    }

    // ── Block kind ─────────────────────────────────────────────────────────

    /// Convert the block at `offset` to `kind` (a `level`-N heading, or a
    /// paragraph) by rewriting its leading marker while keeping its inline
    /// content verbatim — the block half of the toolbar (H1 / Body).
    pub fn setBlock(self: *Editor, offset: usize, kind: BlockKind, level: u32) Error!void {
        const marker = self.syntax.heading_marker orelse return error.UnsupportedFormat;
        if (kind == .heading and (level < 1 or level > 6)) return error.InvalidLevel;

        const src = self.sourceBytes();
        if (offset > src.len) return error.InvalidRange;
        const ast = self.astView();
        const block = locate.innermostBlock(ast, offset, src.len) orelse return error.NoBlock;
        const node = ast.nodes[block];
        const cs = node.content_span orelse return error.NotEditable;
        const content = src[cs.start..cs.end];

        // Rewrite [block start, end-of-text): the leading marker region (a
        // heading) or nothing (a paragraph), plus the text — but NOT any
        // trailing newline the block span includes (Djot blocks do), so we don't
        // fuse with the next block. Rebuilding from `content_span` also
        // collapses a setext heading's underline line away for free.
        var end = node.span.end;
        if (end > node.span.start and src[end - 1] == '\n') end -= 1;
        if (end > node.span.start and src[end - 1] == '\r') end -= 1;

        const allocator = self.splicer.allocator;
        const prefix_len: usize = if (kind == .heading) level + 1 else 0; // marker*level + " "
        const buf = try allocator.alloc(u8, prefix_len + content.len);
        defer allocator.free(buf);
        if (kind == .heading) {
            @memset(buf[0..level], marker);
            buf[level] = ' ';
        }
        @memcpy(buf[prefix_len..], content);

        return self.commitSplice(node.span.start, end, buf);
    }

    // ── Block containers (quote / lists) ───────────────────────────────────
    // `setBlock` rewrites the leading marker of ONE block at one offset. A block
    // container is a different animal: it prefixes EVERY line of a possibly
    // multi-block range, it nests, and a list numbers its items — so it gets its
    // own gesture rather than another `BlockKind`. Everything below is line
    // surgery over the covered blocks, spliced in one shot.

    /// Toggle a block container over the blocks `[start, end)` covers.
    ///
    /// The already-in-container test walks the AST ancestors of `start` for a
    /// container of `kind`, and the toggle turns OFF only when the range covers
    /// every block that container holds — otherwise it turns ON, which is what
    /// makes a partial selection inside a quote nest (`> >`) instead of dragging
    /// the container's uncovered siblings out with it. Toggling a list kind
    /// while inside the other list kind converts in place rather than nesting.
    pub fn toggleBlockContainer(self: *Editor, span: Span, kind: ContainerKind) Error!void {
        try self.checkRange(span.start, span.end);
        const sp = self.syntax.container_spelling.get(kind) orelse return error.UnsupportedFormat;

        const allocator = self.splicer.allocator;
        const src = self.sourceBytes();
        const ast = self.astView();

        const blocks = coveredBlocks(allocator, ast, src, span.start, span.end) catch |err| switch (err) {
            error.OutOfMemory => return error.OutOfMemory,
            else => return error.NoBlock,
        };
        defer allocator.free(blocks.chain);

        const region_start = locate.lineStartAt(src, ast.nodes[blocks.first].span.start);
        const region_end = locate.lineEndAt(src, ast.nodes[blocks.last].span.end -| 1);

        var out: std.ArrayList(u8) = .empty;
        defer out.deinit(allocator);

        // The toggle-off / convert / nest decision, all from the ancestor chain.
        if (locate.innermostOfKind(ast, blocks.chain, kindTag(kind))) |target| {
            if (containerFullyCovered(ast, src, target, region_start, region_end)) {
                const t = ast.nodes[target].span;
                // The container's own lines, not the range's: its span can reach
                // past the last covered block (a quote's trailing `>` line).
                const splice_start = locate.lineStartAt(src, t.start);
                const splice_end = locate.lineEndAt(src, t.end -| 1);
                switch (kind) {
                    .block_quote => try buildQuoteStrip(
                        allocator,
                        src,
                        splice_start,
                        splice_end,
                        quoteDepthAbove(ast, blocks.chain, target),
                        &out,
                    ),
                    .bullet_list, .ordered_list => try buildListRewrite(
                        allocator,
                        src,
                        ast,
                        target,
                        splice_start,
                        splice_end,
                        null,
                        &out,
                    ),
                }
                return self.commitSplice(splice_start, splice_end, out.items);
            }
        }
        if (kind == .bullet_list or kind == .ordered_list) {
            const other: Splicer.KindTag = if (kind == .bullet_list) .ordered_list else .bullet_list;
            if (locate.innermostOfKind(ast, blocks.chain, other)) |target| {
                if (containerFullyCovered(ast, src, target, region_start, region_end)) {
                    const t = ast.nodes[target].span;
                    const splice_start = locate.lineStartAt(src, t.start);
                    const splice_end = locate.lineEndAt(src, t.end -| 1);
                    try buildListRewrite(allocator, src, ast, target, splice_start, splice_end, sp, &out);
                    return self.commitSplice(splice_start, splice_end, out.items);
                }
            }
        }

        try buildContainerAdd(allocator, src, ast, blocks, region_start, region_end, sp, &out);
        return self.commitSplice(region_start, region_end, out.items);
    }

    // ── Links ──────────────────────────────────────────────────────────────

    /// Link `[start, end)` to `destination`, or repoint the link already there.
    ///
    /// Decisions:
    ///   * An EXISTING link covering the range has its destination REPLACED, its
    ///     text kept. Re-linking is the common gesture (fix a URL), and it keeps
    ///     the op idempotent instead of nesting `[[t](a)](b)`. Removing a link is
    ///     already `Splicer.unwrapNode`, which peels a node to its interior.
    ///   * A RANGE INSIDE an existing autolink re-points it the same way, but
    ///     there is no text to keep: an autolink's text IS its destination, so
    ///     the node is respelled whole for the new one (canonically — see below
    ///     — so a `<url>` re-pointed at a relative path becomes `[dest](dest)`,
    ///     not a broken `<>`). Without this the op reads the URL as ordinary text
    ///     and splices a link into the middle of it:
    ///     `<https<https://y.dev>://x.dev>`.
    ///
    ///     This covers a caret AND any selection the autolink contains —
    ///     including one covering it exactly. An autolink's URL is not editable
    ///     text: no part of it can host a `[`, so "link half this URL" has no
    ///     spelling, and the selection carries no text a splice could keep.
    ///
    ///     A selection that starts or ends strictly INSIDE an autolink but isn't
    ///     contained by it (`see <https://x` … `.dev> ok`) is refused with
    ///     `error.NotEditable`: half of it is real text, so there is nothing to
    ///     re-point, and any splice would rewrite the URL. Refusing beats
    ///     silently changing the caller's URL, for the same reason a newline
    ///     destination is `error.InvalidDestination`.
    ///
    ///     A range inside BOTH a link and an autolink (`[<https://x.dev>](d)`)
    ///     re-points the link, not the autolink: a link's text is separable from
    ///     its destination, so re-pointing it keeps text that re-pointing the
    ///     autolink would discard.
    ///
    ///     A range that CONTAINS an autolink whole plus text around it is
    ///     untouched by all of the above — it splices at the autolink's edges,
    ///     corrupting nothing, and the autolink stays as the new link's text.
    ///   * A link with NO TEXT gets the canonical spelling for the destination it
    ///     was given, never `[](dest)`: a childless link has nothing to render,
    ///     so consumers fall back to showing the destination and the caret has
    ///     nowhere correct to sit. Where the format can spell an autolink it gets
    ///     `<dest>`; where it can't it gets `[dest](dest)`, the destination
    ///     doubling as text so it stays visible and editable. Which destinations
    ///     autolink, and how each format spells one, is twig's knowledge — a
    ///     consumer guessing would turn `<foo>` into raw HTML (Markdown) or
    ///     literal text (both). See `Syntax.spellsAutolink`.
    ///   * A destination is escaped per format (see `writeLinkDestination`); a
    ///     newline in one is `error.InvalidDestination`, since neither format can
    ///     hold it (Djot strips it, Markdown's `<…>` form forbids it) and
    ///     silently changing the caller's URL is worse than refusing.
    pub fn insertLink(self: *Editor, span: Span, dest: []const u8) Error!void {
        try self.checkRange(span.start, span.end);
        // A format with no link spelling refuses before anything else is read.
        if (self.syntax.link_text_escapes == null) return error.UnsupportedFormat;
        if (std.mem.indexOfAny(u8, dest, "\r\n") != null) return error.InvalidDestination;

        const start = span.start;
        const end = span.end;
        const allocator = self.splicer.allocator;
        const src = self.sourceBytes();
        const ast = self.astView();

        var chain: std.ArrayList(AST.Node.Id) = .empty;
        defer chain.deinit(allocator);
        try locate.ancestorChain(allocator, ast, start, src.len, &chain);

        // The text to sit in the brackets, and the span the rebuilt link
        // replaces. Re-pointing an existing link rebuilds the whole node: a
        // destination is a string payload with no span of its own, so there is
        // nothing smaller to splice (see `splicer.zig`'s module doc).
        var text: []const u8 = src[start..end];
        var target = Span.init(start, end);
        var repoint = locate.innermostCovering(ast, chain.items, &.{.link}, start, end);
        if (repoint == null) repoint = autolinkCovering(ast, chain.items, start, end);
        // Not covered by an autolink, but still landing inside one: the range
        // runs from ordinary text into the middle of a URL (either end can be the
        // one inside). There is nothing to re-point — half the selection is real
        // text — and no way to spell the result, so refuse rather than corrupt
        // the URL.
        if (repoint == null and start != end) {
            const splits =
                (try splitsAutolink(allocator, ast, src.len, start)) or
                (try splitsAutolink(allocator, ast, src.len, end));
            if (splits) return error.NotEditable;
        }
        if (repoint) |id| {
            const node = ast.nodes[id];
            if (node.span.start == 0 and node.span.end == 0) return error.NotEditable;
            // An autolink has no `[text]` half: the text it shows is the OLD
            // destination, so keeping it would spell the new link with the URL it
            // was meant to replace. Empty text sends it through the canonical
            // spelling below, exactly as a caret on bare text goes.
            text = switch (std.meta.activeTag(node.kind)) {
                .url, .email => "",
                else => if (node.content_span) |cs| src[cs.start..cs.end] else "",
            };
            target = node.span;
        }

        var out: std.ArrayList(u8) = .empty;
        defer out.deinit(allocator);

        // Keyed on the TEXT being empty, not the range: re-pointing an existing
        // `[](old)` is an empty range too, and it has the same childless link to
        // avoid. A non-empty range always carries text, so it never lands here.
        if (text.len == 0) {
            if (self.syntax.spellsAutolink) |spells| {
                try out.append(allocator, '<');
                try out.appendSlice(allocator, dest);
                try out.append(allocator, '>');
                // Ask about the exact bytes we would emit, so the test and the
                // output cannot disagree about what was spelled.
                if (spells(out.items)) return self.commitSplice(target.start, target.end, out.items);
                out.clearRetainingCapacity();
            }
        }

        try out.append(allocator, '[');
        if (text.len == 0) {
            // `dest` is a raw string being repurposed as text, so it needs
            // escaping for that position — unlike `text`, which is already source
            // the author (or a prior parse) spelled and which must be copied
            // through verbatim.
            try writeLinkText(allocator, self.syntax, dest, &out);
        } else {
            try out.appendSlice(allocator, text);
        }
        try out.appendSlice(allocator, "](");
        try writeLinkDestination(allocator, self.syntax, dest, &out);
        try out.append(allocator, ')');

        return self.commitSplice(target.start, target.end, out.items);
    }
};

// ── Block-container internals ──────────────────────────────────────────────

/// The blocks `[start, end)` touches: sibling `first`…`last` under the nearest
/// ancestor whose children are blocks. You cannot quote half a paragraph, so a
/// container op always widens to whole blocks first.
const BlockRange = struct {
    first: AST.Node.Id,
    last: AST.Node.Id,
    /// The ancestor chain down to `start`, reused for container detection.
    chain: []const AST.Node.Id,
};

/// Resolve `[start, end)` to the sibling blocks it touches. `end` is pulled back
/// off a trailing newline first: a block's span stops at its text in Markdown,
/// so a selection ending on the line break would otherwise resolve above the
/// block and drag the whole document in.
fn coveredBlocks(
    allocator: Allocator,
    ast: *const AST,
    src: []const u8,
    start: usize,
    end: usize,
) !BlockRange {
    var last_off = if (end > start) end - 1 else start;
    while (last_off > start and (src[last_off] == '\n' or src[last_off] == '\r')) last_off -= 1;

    var chain_a: std.ArrayList(AST.Node.Id) = .empty;
    errdefer chain_a.deinit(allocator);
    try locate.ancestorChain(allocator, ast, start, src.len, &chain_a);

    var chain_b: std.ArrayList(AST.Node.Id) = .empty;
    defer chain_b.deinit(allocator);
    try locate.ancestorChain(allocator, ast, last_off, src.len, &chain_b);

    var i: usize = 0;
    while (i + 1 < chain_a.items.len and i + 1 < chain_b.items.len and
        chain_a.items[i + 1] == chain_b.items[i + 1]) : (i += 1)
    {}
    // Climb to the nearest ancestor that holds blocks: the deepest shared node
    // may be an inline (a `str`), and a container wraps blocks, not words.
    var p = i;
    while (p > 0 and !locate.isBlockParent(std.meta.activeTag(ast.nodes[chain_a.items[p]].kind))) p -= 1;

    if (p + 1 >= chain_a.items.len) return error.NoBlock;
    const first = chain_a.items[p + 1];
    const last = if (p + 1 < chain_b.items.len) chain_b.items[p + 1] else first;
    return .{
        .first = first,
        .last = last,
        .chain = try chain_a.toOwnedSlice(allocator),
    };
}

/// True when the range's lines cover every block `target` holds — the condition
/// for toggling the container OFF rather than nesting inside it.
///
/// The test is "are all its blocks covered?", NOT "is its span inside the
/// region?": a container's span can run past its last block, because the blank
/// `>` line continuing a quote belongs to the quote and to no paragraph in it
/// (Djot spans `> > a\n>\n` as the inner quote, ending two bytes past its only
/// paragraph). Comparing spans there reads a fully-covered quote as partial and
/// nests forever.
fn containerFullyCovered(
    ast: *const AST,
    src: []const u8,
    target: AST.Node.Id,
    region_start: usize,
    region_end: usize,
) bool {
    const first = ast.nodes[target].first_child orelse return false;
    var last = first;
    var cur: ?AST.Node.Id = first;
    while (cur) |c| {
        last = c;
        cur = ast.nodes[c].next_sibling;
    }
    const lo = locate.lineStartAt(src, ast.nodes[first].span.start);
    const hi = locate.lineEndAt(src, ast.nodes[last].span.end -| 1);
    return region_start <= lo and region_end >= hi;
}

/// How many quotes enclose `target` on the chain — the number of `>` markers to
/// step over before the one that belongs to `target`.
fn quoteDepthAbove(ast: *const AST, chain: []const AST.Node.Id, target: AST.Node.Id) usize {
    var depth: usize = 0;
    for (chain) |id| {
        if (id == target) break;
        if (std.meta.activeTag(ast.nodes[id].kind) == .block_quote) depth += 1;
    }
    return depth;
}

/// Advance past one `>` quote marker — its optional indent, the `>`, and the one
/// optional space after it — or `null` if `line[i..]` doesn't start one.
fn skipQuoteMarker(line: []const u8, i: usize) ?usize {
    var j = i;
    var indent: usize = 0;
    while (j < line.len and line[j] == ' ' and indent < 3) : (indent += 1) j += 1;
    if (j >= line.len or line[j] != '>') return null;
    j += 1;
    if (j < line.len and line[j] == ' ') j += 1;
    return j;
}

/// The `[start, end)` of a list marker on `line` — `start` at the bullet/first
/// digit (so the indent before it stays put, keeping an enclosing container's
/// prefix intact) and `end` past the marker's trailing spaces. `null` if the
/// line doesn't open a list item.
fn listMarkerAt(line: []const u8) ?struct { start: usize, end: usize } {
    var j: usize = 0;
    while (j < line.len and (line[j] == ' ' or line[j] == '\t')) j += 1;
    const start = j;
    if (j >= line.len) return null;
    if (line[j] == '-' or line[j] == '*' or line[j] == '+') {
        j += 1;
    } else {
        if (line[j] == '(') j += 1;
        var digits: usize = 0;
        while (j < line.len and line[j] >= '0' and line[j] <= '9') : (digits += 1) j += 1;
        if (digits == 0) return null;
        if (j >= line.len or (line[j] != '.' and line[j] != ')')) return null;
        j += 1;
    }
    // A marker must be followed by whitespace (or end the line): `-x` is a
    // paragraph starting with a hyphen, not a bullet.
    if (j < line.len and line[j] != ' ' and line[j] != '\n' and line[j] != '\r') return null;
    while (j < line.len and line[j] == ' ') j += 1;
    return .{ .start = start, .end = j };
}

/// True if one of the covered blocks begins on `[line_start, line_end)` — the
/// test for "this line opens a new list item". Djot starts a quoted block at its
/// text (after `> `), Markdown at the line start; either way it lands on the
/// block's first line, which is all this asks.
fn blockStartsOnLine(ast: *const AST, blocks: BlockRange, line_start: usize, line_end: usize) bool {
    var cur: ?AST.Node.Id = blocks.first;
    while (cur) |id| {
        const s = ast.nodes[id].span.start;
        if (s >= line_start and s < line_end) return true;
        if (id == blocks.last) break;
        cur = ast.nodes[id].next_sibling;
    }
    return false;
}

/// True if one of `list`'s items begins on `[line_start, line_end)`.
fn itemStartsOnLine(ast: *const AST, list: AST.Node.Id, line_start: usize, line_end: usize) bool {
    var it = ast.children(list);
    while (it.next()) |item| {
        const s = item.span.start;
        if (s >= line_start and s < line_end) return true;
    }
    return false;
}

/// Wrap every line of `[region_start, region_end)` in `kind`'s prefix, one item
/// per covered block. The lines already carry any enclosing container's prefix,
/// so prefixing at column 0 nests naturally (`> a` -> `> > a`).
fn buildContainerAdd(
    allocator: Allocator,
    src: []const u8,
    ast: *const AST,
    blocks: BlockRange,
    region_start: usize,
    region_end: usize,
    sp: ContainerSpelling,
    out: *std.ArrayList(u8),
) !void {
    var ordinal: u32 = 1;
    var cont: []const u8 = sp.cont;
    var line_start = region_start;
    while (line_start < region_end) {
        const line_end = locate.lineEndAt(src, line_start);
        const line = src[line_start..line_end];
        const body = locate.lineBody(line);

        if (locate.isBlankLine(body)) {
            // A blank line inside the region: mark it for a quote (else the quote
            // ends here), leave it bare for a list (it separates items).
            if (sp.blank.len > 0) {
                try out.appendSlice(allocator, sp.blank);
                try out.appendSlice(allocator, line[body.len..]);
            } else {
                try out.appendSlice(allocator, line);
            }
            line_start = line_end;
            continue;
        }

        if (blockStartsOnLine(ast, blocks, line_start, line_end)) {
            var num_buf: [24]u8 = undefined;
            const marker = if (sp.numbered)
                std.fmt.bufPrint(&num_buf, "{d}. ", .{ordinal}) catch unreachable
            else
                sp.marker;
            if (sp.numbered) cont = container_indent[0..@min(marker.len, container_indent.len)];
            try out.appendSlice(allocator, marker);
            try out.appendSlice(allocator, line);
            ordinal += 1;
        } else {
            try out.appendSlice(allocator, cont);
            try out.appendSlice(allocator, line);
        }
        line_start = line_end;
    }
}

/// Strip the quote marker `target` contributes from each of its lines, leaving
/// any outer quote levels untouched: `depth` is how many quotes enclose it, so
/// the marker removed is the `depth`-th + 1 on every line. That's what makes
/// toggling off a nested quote peel exactly one level (`> > a` -> `> a`).
fn buildQuoteStrip(
    allocator: Allocator,
    src: []const u8,
    region_start: usize,
    region_end: usize,
    depth: usize,
    out: *std.ArrayList(u8),
) !void {
    var line_start = region_start;
    while (line_start < region_end) {
        const line_end = locate.lineEndAt(src, line_start);
        const line = src[line_start..line_end];

        var keep: usize = 0;
        var d: usize = 0;
        while (d < depth) : (d += 1) keep = skipQuoteMarker(line, keep) orelse break;
        if (d == depth) {
            if (skipQuoteMarker(line, keep)) |after| {
                try out.appendSlice(allocator, line[0..keep]);
                try out.appendSlice(allocator, line[after..]);
                line_start = line_end;
                continue;
            }
        }
        // A line with no marker at this level (a lazy continuation) is already
        // outside the level being removed — pass it through untouched.
        try out.appendSlice(allocator, line);
        line_start = line_end;
    }
}

/// Rewrite the list `target`'s item markers: `sp == null` removes the list
/// (toggle off), otherwise it converts one list kind to the other in place. The
/// text before a marker (an enclosing quote's `> `, a nesting indent) is kept
/// verbatim; a block's continuation lines are re-indented to the new marker's
/// width so they stay attached to their item.
///
/// Removing a list has to keep its items separate BLOCKS: a tight `- a\n- b\n`
/// would strip to `a\nb\n`, which is one two-line paragraph, not two. So a blank
/// line is injected between items that had none — the structure the items had is
/// what survives, not their tightness.
fn buildListRewrite(
    allocator: Allocator,
    src: []const u8,
    ast: *const AST,
    target: AST.Node.Id,
    region_start: usize,
    region_end: usize,
    sp: ?ContainerSpelling,
    out: *std.ArrayList(u8),
) !void {
    var ordinal: u32 = 1;
    var old_width: usize = 0;
    var new_width: usize = 0;
    var seen_item = false;
    var last_blank = true;
    var line_start = region_start;
    while (line_start < region_end) {
        const line_end = locate.lineEndAt(src, line_start);
        const line = src[line_start..line_end];
        const body = locate.lineBody(line);

        if (locate.isBlankLine(body)) {
            try out.appendSlice(allocator, line);
            last_blank = true;
            line_start = line_end;
            continue;
        }

        if (itemStartsOnLine(ast, target, line_start, line_end)) {
            // Only when the list is going away: a conversion keeps the items as
            // items, so it must not loosen a tight list.
            if (sp == null and seen_item and !last_blank) try out.append(allocator, '\n');
            const m = listMarkerAt(line) orelse {
                try out.appendSlice(allocator, line);
                line_start = line_end;
                continue;
            };
            var num_buf: [24]u8 = undefined;
            const marker: []const u8 = if (sp) |s|
                (if (s.numbered)
                    std.fmt.bufPrint(&num_buf, "{d}. ", .{ordinal}) catch unreachable
                else
                    s.marker)
            else
                "";
            try out.appendSlice(allocator, line[0..m.start]);
            try out.appendSlice(allocator, marker);
            try out.appendSlice(allocator, line[m.end..]);
            old_width = m.end - m.start;
            new_width = marker.len;
            ordinal += 1;
            seen_item = true;
            last_blank = false;
        } else {
            // A continuation line: swap the old marker's indent for the new one's
            // so the line stays inside its item.
            var j: usize = 0;
            while (j < line.len and j < old_width and line[j] == ' ') j += 1;
            try out.appendSlice(allocator, container_indent[0..@min(new_width, container_indent.len)]);
            try out.appendSlice(allocator, line[j..]);
            last_blank = false;
        }
        line_start = line_end;
    }
}

// ── Link internals ─────────────────────────────────────────────────────────
// `toggleInline` can't spell a link: its delimiters are a fixed `(open, close)`
// pair, and a link's closing half carries a payload (`](dest)`). Hence a
// dedicated gesture with a destination argument — and with the escaping that
// payload needs.

/// Write `dest` into `out` spelled so the format parses it back byte-for-byte.
///
/// This is the sharp edge of the whole gesture, and it is NOT one escape table:
///
///   * Markdown ends a destination at the first space — `[t](a b)` is not a link
///     at all, it is literal text — so a destination holding whitespace has to
///     move into the `<…>` form, where `<`/`>`/`\` are what need escaping.
///   * Djot takes spaces literally and gives `<…>` NO meaning: `[t](<a b>)`
///     links to the seven characters `<a b>`. Wrapping there would corrupt the
///     URL rather than protect it.
///
/// That difference is `DestEscapes.angle`: non-null means the format HAS an
/// angle form to escape into. The algorithm is the same either way, which is why
/// it lives here once and the alphabets live in `syntax.zig`.
///
/// Both formats honour a backslash escape inside the destination, which is what
/// keeps an unbalanced `)` from closing the link early.
fn writeLinkDestination(
    allocator: Allocator,
    syntax: *const Syntax,
    dest: []const u8,
    out: *std.ArrayList(u8),
) !void {
    const de = syntax.link_dest_escapes orelse return error.UnsupportedFormat;
    const angle: ?[]const u8 = if (de.angle) |a|
        (if (std.mem.indexOfAny(u8, dest, " \t") != null) a.escapes else null)
    else
        null;

    if (angle != null) try out.append(allocator, '<');
    const escapes = angle orelse de.plain;
    for (dest) |c| {
        if (std.mem.indexOfScalar(u8, escapes, c) != null) try out.append(allocator, '\\');
        try out.append(allocator, c);
    }
    if (angle != null) try out.append(allocator, '>');
}

fn writeLinkText(
    allocator: Allocator,
    syntax: *const Syntax,
    text: []const u8,
    out: *std.ArrayList(u8),
) !void {
    const escapes = syntax.link_text_escapes orelse return error.UnsupportedFormat;
    for (text) |c| {
        if (std.mem.indexOfScalar(u8, escapes, c) != null) try out.append(allocator, '\\');
        try out.append(allocator, c);
    }
}

/// The innermost autolink — the `<https://x.dev>` / `<a@b.dev>` form — on the
/// chain that wholly contains `[start, end)`.
///
/// Both node kinds are matched in both formats because the split is not the one
/// the names suggest — it follows the FORMAT, not just the destination.
/// `<mailto:a@b.dev>` parses as a `url` in Markdown and an `email` in djot, so
/// picking one kind per format would miss half the autolinks it was meant to
/// catch.
fn autolinkCovering(ast: *const AST, chain: []const AST.Node.Id, start: usize, end: usize) ?AST.Node.Id {
    return locate.innermostCovering(ast, chain, &.{ .url, .email }, start, end);
}

/// Whether writing at `pos` would land STRICTLY INSIDE an autolink's URL — an
/// autolink covers `pos`, and `pos` is neither of its edges. A splice at an edge
/// is safe (it lands beside the node); one strictly inside rewrites the URL
/// itself, which is never what any caller meant. See `Editor.insertLink`.
///
/// Builds its own chain because the caller's is rooted at `start`, and the offset
/// that lands inside can be `end` (a selection running from ordinary text into
/// the middle of a URL).
fn splitsAutolink(allocator: Allocator, ast: *const AST, source_len: usize, pos: usize) Allocator.Error!bool {
    var chain: std.ArrayList(AST.Node.Id) = .empty;
    defer chain.deinit(allocator);
    try locate.ancestorChain(allocator, ast, pos, source_len, &chain);
    const id = autolinkCovering(ast, chain.items, pos, pos) orelse return false;
    const span = ast.nodes[id].span;
    return span.start < pos and pos < span.end;
}

test {
    _ = @import("editor_test.zig");
}