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