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