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