Skip to main content

harn_vm/
stdlib.rs

1//! Standard library builtins for the Harn VM.
2//!
3//! Every builtin is declared with the `#[harn_builtin]` proc-macro
4//! (`crate::stdlib::macros::harn_builtin`). Each annotation emits a sibling
5//! `static <FN>_DEF: VmBuiltinDef` carrying the signature, aliases, handler,
6//! and metadata, and registers it into the workspace-global
7//! [`macros::ALL_BUILTIN_DEFS`] distributed slice at link time. The CLI / LSP /
8//! lint / serve / dap binaries call [`force_link`] to defeat rlib dead-code
9//! stripping (linkme issue #36) so every static lands in the slice. Modules
10//! still expose a `register_<module>_builtins(vm)` helper for ordered eager
11//! registration (e.g. so `clock::timestamp` can override `process::timestamp`).
12//! `register_vm_stdlib` calls those helpers in order and then installs the
13//! aggregated signatures into the parser registry.
14//!
15//! See `CONTRIBUTING.md` ("Adding a stdlib builtin") for the full template.
16
17pub mod macros;
18
19mod agent_sessions;
20pub mod agent_state;
21pub(crate) mod agents;
22mod agents_daemon;
23mod artifact_emit;
24pub(crate) mod assemble;
25pub mod asset_paths;
26mod bytes;
27mod calendar;
28mod channel_guardrails;
29mod channels;
30pub(crate) mod clock;
31pub(crate) mod collections;
32mod command_policy;
33pub(crate) mod compaction;
34mod compression;
35mod concurrency;
36mod connectors;
37mod cookies;
38mod cron;
39mod crypto;
40mod csv;
41mod datetime;
42mod document;
43mod durable_step;
44mod event_log;
45mod external_agent;
46pub(crate) mod files;
47mod flow;
48mod fs;
49mod git;
50pub(crate) mod git_topology;
51mod grounding;
52pub(crate) mod harn_entry;
53pub(crate) mod hitl;
54mod hitl_read;
55pub mod host;
56pub mod http_response;
57pub(crate) mod io;
58mod iter;
59pub(crate) mod json;
60mod json_query;
61pub(crate) mod json_stream;
62mod jsonrpc;
63mod junit;
64mod lifecycle_receipts;
65mod logging;
66pub mod long_running;
67mod math;
68pub(crate) mod memory;
69mod monitors;
70mod multipart;
71mod net;
72mod net_policy;
73mod oauth_dynreg;
74mod oauth_storage;
75pub(crate) mod observability;
76pub(crate) mod options;
77mod package_snapshot;
78pub(crate) use package_snapshot::PackageSnapshotRegistry;
79mod path;
80pub(crate) mod path_scope_guard;
81pub(crate) mod pool;
82#[cfg(feature = "postgres")]
83mod postgres;
84#[cfg(feature = "postgres")]
85pub use postgres::install_shared_pool_registry;
86pub mod process;
87pub(crate) mod process_spawn;
88mod project;
89mod project_catalog;
90mod project_enrich;
91mod regex;
92mod review;
93mod runtime_scope;
94pub(crate) mod sandbox;
95pub mod secret_scan;
96pub(crate) mod session_store;
97mod sets;
98pub(crate) mod shapes;
99mod skills;
100#[cfg(feature = "sqlite")]
101mod sqlite;
102pub(crate) mod strings;
103pub(crate) mod supervisor;
104pub mod template;
105mod testbench;
106mod testing;
107mod timing;
108pub mod token_redaction;
109pub(crate) mod tool_hooks;
110pub(crate) mod tools;
111pub mod tracing;
112mod transcript_compact;
113pub(crate) mod transcript_project;
114mod triggers_stdlib;
115mod tui;
116mod types;
117mod url_parse;
118mod vision;
119pub(crate) mod waitpoint;
120mod waitpoints;
121mod web;
122pub mod workflow_messages;
123pub(crate) mod xml;
124
125use crate::http::register_http_builtins;
126use crate::llm::register_llm_builtins;
127use crate::mcp::register_mcp_builtins;
128use crate::mcp_server::register_mcp_server_builtins;
129use crate::vm::Vm;
130
131pub(crate) use crate::schema::{json_to_vm_value, schema_result_value};
132pub(crate) fn set_thread_source_dir(dir: &std::path::Path) {
133    process::set_thread_source_dir(dir);
134}
135
136/// Register core builtins: pure/deterministic, no I/O.
137pub fn register_core_stdlib(vm: &mut Vm) {
138    crate::runtime_context::register_runtime_context_builtins(vm);
139    types::register_type_builtins(vm);
140    math::register_math_builtins(vm);
141    strings::register_string_builtins(vm);
142    json::register_json_builtins(vm);
143    json_stream::register_json_stream_builtins(vm);
144    xml::register_xml_builtins(vm);
145    datetime::register_datetime_builtins(vm);
146    document::register_document_builtins(vm);
147    calendar::register_calendar_builtins(vm);
148    cron::register_cron_builtins(vm);
149    regex::register_regex_builtins(vm);
150    bytes::register_bytes_builtins(vm);
151    compression::register_compression_builtins(vm);
152    command_policy::register_command_policy_builtins(vm);
153    runtime_scope::register_runtime_scope_builtins(vm);
154    crypto::register_crypto_builtins(vm);
155    csv::register_csv_builtins(vm);
156    junit::register_junit_builtins(vm);
157    multipart::register_multipart_builtins(vm);
158    url_parse::register_url_builtins(vm);
159    web::register_web_builtins(vm);
160    cookies::register_cookie_builtins(vm);
161    path::register_path_helper_builtins(vm);
162    sets::register_set_builtins(vm);
163    collections::register_collection_builtins(vm);
164    iter::register_iter_builtins(vm);
165    event_log::register_event_log_builtins(vm);
166    durable_step::register_durable_step_builtins(vm);
167    channels::register_channel_builtins(vm);
168    channel_guardrails::register_channel_guardrail_builtins(vm);
169    shapes::register_shape_builtins(vm);
170    testing::register_testing_builtins(vm);
171    flow::register_flow_builtins(vm);
172    lifecycle_receipts::register_lifecycle_receipt_builtins(vm);
173    net_policy::register_net_policy_builtins(vm);
174    http_response::register_http_response_builtins(vm);
175}
176
177/// Register I/O builtins (requires OS access).
178pub fn register_io_stdlib(vm: &mut Vm) {
179    io::register_io_builtins(vm);
180    host::register_host_builtins(vm);
181    fs::register_fs_builtins(vm);
182    package_snapshot::register_package_snapshot_builtins(vm);
183    files::register_file_builtins(vm);
184    git::register_git_builtins(vm);
185    vision::register_vision_builtins(vm);
186    agent_state::register_agent_state_builtins(vm);
187    memory::register_memory_builtins(vm);
188    session_store::register_session_store_builtins(vm);
189    net::register_net_builtins(vm);
190    process::register_process_builtins(vm);
191    process::register_path_builtins(vm);
192    sandbox::register_sandbox_builtins(vm);
193    // Clock builtins overlay process::timestamp/elapsed so they honor
194    // mock_time / advance_time. Register AFTER process to take precedence.
195    clock::register_clock_builtins(vm);
196    crate::durable_rate_limit::register_durable_rate_limit_builtins(vm);
197    testbench::register_testbench_builtins(vm);
198    project::register_project_builtins(vm);
199    grounding::register_grounding_builtins(vm);
200    tracing::register_tracing_builtins(vm);
201    observability::register_observability_builtins(vm);
202    timing::register_timing_builtins(vm);
203    tui::register_tui_builtins(vm);
204}
205
206fn register_agent_stdlib_before_llm(vm: &mut Vm) {
207    concurrency::register_concurrency_builtins(vm);
208    connectors::register_connector_builtins(vm);
209    review::register_review_builtins(vm);
210    secret_scan::register_secret_scan_builtins(vm);
211    tools::register_tool_builtins(vm);
212    tool_hooks::register_tool_hooks_builtins(vm);
213    crate::composition::register_composition_builtins(vm);
214    skills::register_skill_builtins(vm);
215    agents_daemon::register_daemon_builtins(vm);
216    triggers_stdlib::register_trigger_builtins(vm);
217    #[cfg(feature = "postgres")]
218    postgres::register_postgres_builtins(vm);
219    #[cfg(feature = "sqlite")]
220    sqlite::register_sqlite_builtins(vm);
221    waitpoints::register_waitpoint_builtins(vm);
222    monitors::register_monitor_builtins(vm);
223    hitl::register_hitl_builtins(vm);
224    hitl_read::register_hitl_read_builtins(vm);
225    waitpoint::register_waitpoint_builtins(vm);
226    supervisor::register_supervisor_builtins(vm);
227    agents::register_agent_builtins(vm);
228    pool::register_pool_builtins(vm);
229    oauth_storage::register_oauth_storage_builtins(vm);
230    oauth_dynreg::register_oauth_dynreg_builtins(vm);
231    token_redaction::register_token_redaction_builtins(vm);
232    agent_sessions::register_agent_session_builtins(vm);
233    artifact_emit::register_artifact_emit_builtins(vm);
234    external_agent::register_external_agent_builtins(vm);
235    path_scope_guard::register_path_scope_guard_builtins(vm);
236    workflow_messages::register_workflow_message_builtins(vm);
237    transcript_compact::register_transcript_compaction_builtins(vm);
238    compaction::register_compaction_builtins(vm);
239    transcript_project::register_transcript_projection_builtins(vm);
240    assemble::register_assemble_context_builtin(vm);
241    crate::egress::register_egress_builtins(vm);
242    crate::security::register_security_builtins(vm);
243    register_http_builtins(vm);
244    jsonrpc::register_jsonrpc_builtins(vm);
245}
246
247fn register_agent_stdlib_after_llm(vm: &mut Vm) {
248    register_mcp_builtins(vm);
249    register_mcp_server_builtins(vm);
250    crate::step_runtime::register_step_builtins(vm);
251}
252
253/// Register agent builtins (requires network access and async runtime).
254pub fn register_agent_stdlib(vm: &mut Vm) {
255    register_agent_stdlib_before_llm(vm);
256    register_llm_builtins(vm);
257    register_agent_stdlib_after_llm(vm);
258}
259
260/// Register all standard builtins on a VM (core + io + agent). Also
261/// installs the macro-emitted signature slice into the parser registry
262/// (idempotent under repeat calls with the same slice pointer).
263pub fn register_vm_stdlib(vm: &mut Vm) {
264    register_core_stdlib(vm);
265    register_io_stdlib(vm);
266    register_agent_stdlib(vm);
267    if vm.global("harness").is_none() {
268        vm.set_harness(crate::harness::Harness::real());
269    }
270    harn_builtin_registry::install_builtin_signatures(all_builtin_signatures());
271}
272
273pub(crate) fn rebind_execution_state_builtins(vm: &mut Vm) {
274    concurrency::register_concurrency_builtins(vm);
275}
276
277fn stdlib_probe_vm() -> Vm {
278    let mut vm = Vm::new();
279    register_vm_stdlib(&mut vm);
280    // Name-only/metadata introspection never accesses this path, but passing
281    // a real per-platform temp dir keeps registration logic honest if a
282    // callee someday validates its parent.
283    let tmp = std::env::temp_dir();
284    crate::store::register_store_builtins(&mut vm, &tmp);
285    crate::checkpoint::register_checkpoint_builtins(&mut vm, &tmp, "default");
286    crate::metadata::register_metadata_builtins(&mut vm, &tmp);
287    // Install the macro-emitted signatures into the parser registry so any
288    // probe-driven name/metadata query (e.g. the alignment test) sees the
289    // post-migration sig set. Idempotent under repeat install with the same
290    // pointer (which `all_builtin_signatures()` guarantees).
291    harn_builtin_registry::install_builtin_signatures(all_builtin_signatures());
292    vm
293}
294
295/// Aggregate of every `#[harn_builtin]`-emitted `VmBuiltinDef` in the stdlib.
296///
297/// Backed by the `linkme::distributed_slice` declared on
298/// [`crate::stdlib::macros::ALL_BUILTIN_DEFS`] — every annotated fn
299/// contributes one entry automatically at link time. Keep builtin registration
300/// on this distributed slice instead of per-module arrays plus a central
301/// hand-maintained aggregator.
302///
303/// **Force-link warning** (linkme issue #36): rlib dead-code stripping
304/// can drop these statics when `harn-vm` is linked transitively. Every
305/// binary that exercises builtins (`harn-cli`, `harn-lsp`, `harn-lint`,
306/// `harn-serve`, `harn-dap`) calls [`force_link`] near `main()` to defeat
307/// the stripping. The alignment test
308/// `linkme_distributed_slice_populates_with_all_builtins` catches a silent
309/// regression by asserting the slice is non-empty.
310pub fn all_builtin_defs() -> &'static [&'static macros::VmBuiltinDef] {
311    &macros::ALL_BUILTIN_DEFS
312}
313
314/// Force-link entry point: a `pub fn` that touches `ALL_BUILTIN_DEFS` so
315/// the linker keeps every `#[harn_builtin]`-emitted static. Drivers
316/// (`harn-cli`, `harn-lsp`, etc.) call this once at startup. Doing nothing
317/// at runtime is fine — the side effect is purely a link-time signal.
318///
319/// See [`linkme issue #36`](https://github.com/dtolnay/linkme/issues/36)
320/// for why the explicit touch is necessary on every supported target.
321pub fn force_link() {
322    // `black_box` prevents LLVM from constant-folding the length read away.
323    // The `>= 1` guard never trips at runtime but is a load-bearing safety
324    // net: it converts a silent slice-empty regression into a panic that
325    // surfaces at the first builtin call instead of a confusing
326    // `HARN-NAM-002` somewhere down the line.
327    let len = std::hint::black_box(macros::ALL_BUILTIN_DEFS.len());
328    assert!(
329        len >= 1,
330        "linkme distributed_slice ALL_BUILTIN_DEFS is empty — \
331         the binary is missing `harn_vm::stdlib::force_link()` at startup, \
332         or the linker stripped the harn-vm rlib statics (see linkme issue #36)"
333    );
334}
335
336/// Driver-facing helper: flatten the macro-emitted `BuiltinDef`s into a
337/// `&'static [&'static BuiltinSignature]` slice suitable for
338/// [`harn_builtin_registry::install_builtin_signatures`].
339///
340/// Aliases are expanded into their own `BuiltinSignature` entries (the
341/// allocation is leaked once at startup — process-lifetime is appropriate
342/// for a global registry).
343pub fn all_builtin_signatures() -> &'static [&'static harn_builtin_meta::BuiltinSignature] {
344    use std::sync::OnceLock;
345    static AGG: OnceLock<Vec<&'static harn_builtin_meta::BuiltinSignature>> = OnceLock::new();
346    AGG.get_or_init(|| {
347        let mut out: Vec<&'static harn_builtin_meta::BuiltinSignature> = Vec::new();
348        for def in all_builtin_defs() {
349            if def.runtime_only {
350                continue;
351            }
352            out.push(&def.sig);
353            for alias in def.aliases {
354                let aliased = harn_builtin_meta::BuiltinSignature {
355                    name: alias,
356                    ..def.sig
357                };
358                out.push(Box::leak(Box::new(aliased)));
359            }
360        }
361        out
362    })
363    .as_slice()
364}
365
366/// Register every `#[harn_builtin]`-emitted def on the given VM. Drivers
367/// that build the full stdlib via `register_vm_stdlib` get this for free —
368/// each module's `register_*_builtins` walks its `MODULE_BUILTINS` slice.
369/// This helper is exposed for embedders / tests that want a one-call entry.
370pub fn register_all_macro_builtins(vm: &mut Vm) {
371    for def in all_builtin_defs() {
372        vm.register_builtin_def(def);
373    }
374}
375
376/// Return the canonical list of all stdlib builtin names. Used by
377/// harn-lint and harn-lsp to avoid hardcoded duplicate lists.
378pub fn stdlib_builtin_names() -> Vec<String> {
379    let vm = stdlib_probe_vm();
380    let mut names = vm.builtin_names();
381    // Special opcodes/keywords, not registered builtins, but linter
382    // should recognize them as valid function calls.
383    for extra in [
384        "spawn",
385        "await",
386        "cancel",
387        "cancel_graceful",
388        "__signal_interrupted",
389        "__signal_off_interrupt",
390        "__signal_on_interrupt",
391        "__signal_raise",
392        "is_cancelled",
393    ] {
394        names.push(extra.to_string());
395    }
396    names
397}
398
399/// Return discoverable metadata for registered stdlib builtins.
400pub fn stdlib_builtin_metadata() -> Vec<crate::vm::VmBuiltinMetadata> {
401    stdlib_probe_vm().builtin_metadata()
402}
403
404/// Reset thread-local stdlib state. Call between test runs.
405///
406/// Note: `long_running::reset_state()` is intentionally NOT called here
407/// because that store is process-global, not thread-local. Wiping it
408/// from a per-test reset hook lets one test cancel another test's
409/// in-flight worker thread (and lose its `agent_inbox::push`
410/// notification), which surfaces as `walk_dir_long_running` /
411/// `glob_long_running` timing out under parallel test load. The two
412/// call sites that genuinely need a clean handle store —
413/// `stdlib::fs::tests::{walk_dir_long_running,glob_long_running}` — call
414/// `long_running::reset_state()` explicitly while holding
415/// `LONG_RUNNING_TEST_LOCK`.
416pub fn reset_stdlib_state() {
417    logging::reset_logging_state();
418    process::reset_process_state();
419    clock::reset_clock_state();
420    io::reset_io_state();
421    sandbox::reset_sandbox_state();
422    git::reset_git_state();
423    fs::reset_fs_state();
424    json::reset_json_state();
425    json_stream::reset_json_stream_state();
426    host::reset_host_state();
427    host::reset_scoped_host_state();
428    observability::reset_observability_state();
429    timing::reset_timing_state();
430    durable_step::reset_durable_step_state();
431    crate::egress::reset_egress_policy_for_host();
432    hitl::reset_hitl_state();
433    crate::http::reset_http_state();
434    crate::external_agent::reset_external_agent_state();
435    jsonrpc::reset_jsonrpc_state();
436    monitors::reset_monitor_state();
437    waitpoints::reset_waitpoint_state();
438    waitpoint::reset_waitpoint_state();
439    triggers_stdlib::reset_auto_resume_timeouts();
440    compaction::reset_compaction_state();
441    agents::reset_agent_worker_state();
442    agents::workflow::reset_workflow_run_states();
443    pool::reset_pool_state();
444    #[cfg(feature = "postgres")]
445    postgres::reset_postgres_state();
446    #[cfg(feature = "sqlite")]
447    sqlite::reset_sqlite_state();
448    supervisor::reset_supervisor_state();
449    agents::records::reset_eval_metrics();
450    agents::records::reset_friction_events();
451    tools::clear_current_tool_registry();
452    tools::clear_tool_synthesis_cache();
453    vision::reset_vision_state();
454    crate::skills::clear_current_skill_registry();
455    template::reset_prompt_registry();
456    crate::triggers::clear_webhook_intake_state();
457    crate::llm::cache::reset_in_process_cache_state();
458}
459
460#[cfg(test)]
461mod tests {
462    use super::*;
463
464    #[tokio::test(flavor = "current_thread")]
465    async fn register_vm_stdlib_installs_default_harness_handle() {
466        let chunk = crate::compile_source(
467            r"
468fn __probe_global_harness_clock() {
469  const now = harness.clock.now_ms()
470  return now >= 0
471}
472
473fn main(harness: Harness) {
474  return __probe_global_harness_clock()
475}
476",
477        )
478        .expect("compile harness clock probe");
479        let mut vm = Vm::new();
480        register_vm_stdlib(&mut vm);
481
482        assert!(vm.global("harness").is_some());
483        let result = vm
484            .execute(&chunk)
485            .await
486            .expect("execute harness clock probe");
487        assert!(matches!(result, crate::value::VmValue::Bool(true)));
488    }
489
490    /// `harn_stdlib::builtin_reexports` names builtins from a crate that
491    /// cannot see the builtin registry, so nothing there can catch a typo or a
492    /// rename. This is that check: every re-exported name must resolve to a
493    /// real builtin, or `import { … } from "std/…"` binds a reference to
494    /// nothing and fails at the call site instead of the import.
495    #[test]
496    fn every_stdlib_builtin_reexport_names_a_registered_builtin() {
497        let registered: std::collections::HashSet<&str> = all_builtin_defs()
498            .iter()
499            .flat_map(|def| std::iter::once(def.sig.name).chain(def.aliases.iter().copied()))
500            .collect();
501
502        let mut checked = 0;
503        for entry in harn_stdlib::STDLIB_SOURCES {
504            for name in harn_stdlib::builtin_reexports(entry.module) {
505                assert!(
506                    registered.contains(name),
507                    "std/{} re-exports '{name}', which is not a registered builtin",
508                    entry.module
509                );
510                checked += 1;
511            }
512        }
513        assert!(
514            checked > 0,
515            "no re-exports were checked — the table or the module list is not being read"
516        );
517    }
518}