feat(ipc): runner-aware task IPC infra + auto env tracking#430
Draft
wan9chi wants to merge 5 commits into
Draft
Conversation
Member
Author
This stack of pull requests is managed by Graphite. Learn more about stacking. |
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
wan9chi
added a commit
that referenced
this pull request
Jun 11, 2026
… command (#436) ## Motivation The planning context's env map — the full session environment, typically hundreds of entries — was copied at every boundary it crossed during planning: - once into `PlanContext` per plan (`plan_query`), - once **per `&&`-item** by `PlanContext::duplicate` (each command gets a scoped context), - once more per `&&`-item into `ScriptCommand::envs` for program lookup and plan-request callbacks, - and once per spawn into `all_envs`. Of these, only the last one does real work — the spawn env is filtered in place by `EnvFingerprints::resolve`. Every other copy duplicates an almost-always-identical map just to hand it to the next stage, and the cost scales with task count × command count, not with actual env changes. The immediate trigger is the upcoming runner-aware caching work (#430/#431): it saves each spawn's **full env context** into the plan (`SpawnCommand::full_envs`) so `getEnv`/`getEnvs` and tracked-env validation resolve against the plan rather than the live process env. Without sharing, that would add yet another full map copy per spawn; with this change it costs a pointer. ## Approach Wrap the map in `Arc` with copy-on-write semantics: - Duplication and hand-off points (`duplicate`, `ScriptCommand::envs`, `PlanContext::new`) share the map — O(1). - The two mutation points pay for a copy only when they change something: `prepend_path` once per task node (package `.bin` PATH prepend), and `add_envs` only when a command actually has prefix envs — an empty prefix set is guarded so it doesn't break sharing. - The per-spawn `all_envs` clone remains, with a comment marking it as the one place the map's contents are genuinely copied. No behavior change; `Session` already held the map in an `Arc`, so its call sites are untouched. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
wan9chi
added a commit
that referenced
this pull request
Jun 11, 2026
…ape (#437) ## Motivation `SpawnFingerprintChange` carried three env variants — `EnvAdded`, `EnvRemoved`, `EnvValueChanged` — that encoded the same concept three times over: an env var differing between a stored fingerprint and the current state. Every consumer had to spell out the same three-arm match, and the user-facing wording lived inline in `format_spawn_change`'s arms. This shape is about to be needed in more places. The runner-aware caching work (#430) detects env differences at two additional points — tool-tracked envs (`getEnv`) and env-glob match-sets (`getEnvs`) validated at cache lookup — and without a shared type each would re-invent the added/removed/changed triple plus its own formatting, with the wording drifting across three copies (the first draft of #430 had exactly that: an option-pair `{old: Option<Str>, new: Option<Str>}` with an impossible `(None, None)` state, a three-map `EnvGlobDiff`, and two hand-rolled renderers). ## Approach Introduce `EnvMismatch { Added, Removed, Changed }` next to the other mismatch vocabulary in `cache/`, with its `Display` impl as the single source of the user-facing wording. `SpawnFingerprintChange` folds its three variants into one `Env(EnvMismatch)`. No behavior change: detection logic, message strings, and snapshots are identical. #430 then reuses the enum for its post-run paths instead of defining parallel shapes. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Establishes the runner ↔ vite IPC channel and the auto env-tracking half of runner-aware caching: - `vite_task_ipc_shared`, `vite_task_server`, `vite_task_client`(+napi): the IPC protocol, Rust transport, and embedded napi client vite loads by absolute path (no-op when absent). - vite reports `getEnv`/`getEnvs`/`disableCache` to the runner; the fingerprint diffs the tracked env value/glob match-set across runs and reports `envs changed` on a miss (wording matches manual env config). - `ignoreInput`/`ignoreOutput` are accepted over IPC but applied as a no-op for now — they land with auto output tracking. - Auto output restoration stays disabled: `output: None` resolves to a disabled glob config (matching prior behavior), so a cache hit replays recorded output without restoring written files. The napi `type-def` emission for the embedded client cdylib is redirected to a discarded sink in its build.rs so it doesn't leak `RunnerClient`/`load` into consumer bindings. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The inline cache-miss reason now names the env var(s) that changed — `env 'NODE_ENV' changed` (or `envs 'A', 'B' changed` for several) instead of a bare `envs changed`. Applies to both manually-configured `env` changes and runner-reported tracked-env / tracked-env-glob changes, so the two render consistently. The detailed summary already named the vars. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… mismatches Prerun spawn-fingerprint changes and post-run tracked-env / env-glob mismatches each carried their own added/removed/changed shapes (SpawnFingerprintChange env variants, PostRunMismatch option pairs, EnvGlobDiff). Replace all three with a shared EnvMismatch enum whose Display impl is the single source of the user-facing wording. EnvGlobDiff is gone: glob validation now reports the first differing entry (sorted lockstep scan, like detect_globbed_input_change) instead of accumulating a full diff that rendering then flattened anyway. This also deletes the impossible (None, None) arm in the old tracked_env_to_spawn_change. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…apshot The IPC Recorder's env map and PostRunFingerprint::validate both read the live process env (std::env::vars_os / var_os). Tasks' envs come from the plan; std::env::vars_os is only meant to bootstrap the session snapshot (Session.envs) at init. Thread that snapshot through ExecutionContext / execute_spawn into Recorder::new (now Arc-shared) and through ExecutionCache::try_hit into validate, so getEnv/getEnvs and the tracked-env lookup validation resolve against the same map the plan was built from — and tests can inject envs via Session::init_with. Document the rule in CLAUDE.md (Code Constraints → Environment Variables). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…the plan A command-prefixed env (PREFIXED_ENV=x node tool.mjs) is part of the spawn's env, but getEnv resolved against the session snapshot only and answered (unset) — and recorded that wrong value into the post-run fingerprint. The plan now saves two env maps per spawn: all_envs (filtered, passed to the child — unchanged) and full_envs, the planning context's envs overlaid with the spawn's own additions. Nested expansions already thread prefix envs into the context, so full_envs picks up enclosing runs' prefixes too (taskA: PREFIXED_A=a vt run taskB → taskB's getEnv sees PREFIXED_A). full_envs is serde-skipped: it mirrors the ambient env and is not part of the plan's identity. The Recorder and tracked-env validation both use full_envs, keeping record and lookup symmetric per cache key (prefix envs are in the spawn fingerprint, so a changed prefix is a different entry). e2e: fetch_env_sees_command_prefix_env (was failing, now green) and fetch_env_sees_intermediate_prefix_envs (nested prefix accumulation + cache-hit validation). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR 1 of 2 — splits #410 into a Graphite stack.
Stack:
mainEstablishes the runner ↔ vite IPC channel and the auto env-tracking half of runner-aware caching:
vite_task_ipc_shared,vite_task_server,vite_task_client(+napi): the IPC protocol, Rust transport, and embedded napi client vite loads by absolute path (no-op when absent).getEnv/getEnvs/disableCacheto the runner; the fingerprint diffs the tracked env value/glob match-set across runs and reportsenvs changedon a miss (wording matches manual env config).ignoreInput/ignoreOutputare accepted over IPC but applied as a no-op here — they land in feat(cache): runner-aware auto output tracking + ignore consumption #431.output: Noneresolves to a disabled glob config (matching prior behavior), so a cache hit replays recorded output without restoring written files.The napi
type-defemission for the embedded client cdylib is redirected to a discarded sink in its build.rs so it doesn't leakRunnerClient/loadinto consumer bindings.🤖 Generated with Claude Code