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