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