Delete kill(signal?) from SubprocessHandle: consumers stop a process only
through terminate()'s tree-scoped SIGTERM→graceMs→SIGKILL escalation
(idempotent, also driven by the spec's abort signal, a no-op once the tree
is gone). The single-signal verb had exactly one consumer family —
lsp-local — and what it bought there was a private re-implementation of
the same escalation. The internal kill closure stays in spawn.ts as the
dispose ladder's tier primitive; terminate() now routes through it too.
lsp-local collapses onto the seam's escalation:
- LspConnection replaces its terminate()/kill() pair with one terminate()
that delegates to handle.terminate(). Behavior change: the
framing-failure path terminates instead of instant SIGKILL, so a
misbehaving server now gets SIGTERM plus the killGraceMs window to
flush before SIGKILL.
- ConnectionSpec.pipeDrainGraceMs becomes killGraceMs: one grace, the
spawn spec's graceMs, drives both the escalation window and post-exit
pipe draining (the provider already passed killGraceMs for it).
- LspInstance.forceTerminate() drops its hand-rolled bounded first wait
(LSP_KILL_GRACE) and escalateProcessTree (deleted with its export and
unit test): the seam's escalation already commits to SIGKILL after
killGraceMs, so only the unbounded quiescence awaits stay load-bearing.
Tests: kill()-shaped spawn specs become terminate()-shaped or fold into
the terminate() suites (group-wide delivery; the settled no-op case was
already pinned by 'terminate() after the tree died'); tree-survivor
coverage is intact. A stderr-'inherit' disposition test completes the
stdout/stderr symmetry so the scoped subprocess+lsp coverage gate stands
alone instead of leaning on subagent-acp's cross-package runs.
Docs: SubprocessHandle type-equiv block, seam/impl/group READMEs, and the
consumer-migration Agent Note lose the kill(signal?) vocabulary (zh pairs
re-recorded); cordis api/services catalogs regenerated.
Drop SubprocessSpawnSpec.dshEnv and splitEnvChannels(); childEnv() is now
scrubbed-base + explicit entries with no namespace validation. The invariant
dropped is the reserved-namespace check on explicit entries (DSH_* rejected
from env, non-DSH_* rejected from dshEnv). Explicit-entry trust already
covers it: an explicit credential-shaped entry has always merged after the
scrub as a deliberate caller opt-in, and an explicit DSH_* entry is the same
deliberate act — the staleness invariant lives entirely in scrubbedParentEnv
dropping AMBIENT credential-shaped and DSH_* names, which stays. The
validation's only observed effect was rejecting legitimate explicit entries:
both recent CI breakages (DSH_GATE_CONCURRENCY exported into every job
crashing lsp specs, DSH_PERMISSION_MODE in acp config.env crashing the
child spawn) were this check firing on values a caller meant to pass, each
fixed by routing around the bureaucracy the seam itself imposed.
The bash seam keeps its own request/spec dshEnv field: that is bash-owned
trusted-plugin vocabulary (the ctx.bashEnv collected overlay) whose merge-last
position guarantees a caller env entry cannot displace a managed fact;
bash-local now flattens ENV_OVERRIDES -> spec.env -> spec.dshEnv into the
seam's one env map. subagent-acp and lsp-local pass their single config env
map straight through. DshEnvironment/DshEnvironmentKey/DSH_ENV_PREFIX stay on
the subprocess seam as the namespace vocabulary (bash re-exports them;
scrubbedParentEnv filters on the prefix).
Tests: the two channel-rejection specs and the splitEnvChannels partition
spec are deleted; one spawn spec now proves an explicit DSH_* env entry
reaches the child while an ambient one is scrubbed; the acp/lsp forwarding
specs keep their MOCK_ECHO_ENV / LSP_FAKE_ECHO_ENV assertions with the split
comments rewritten to merge-after-scrub. Docs (en+zh, re-recorded) and the
owning Agent Notes updated; cordis api/services catalogs regenerated.
Codex review of the acp-only fix found lsp-local carries the identical
defect: its server config's unrestricted env merges into the connection's
ordinary spawn channel, so a configured DSH_* fact crashed the spawn with
the reserved-namespace rejection. The partition now lives on the seam as
splitEnvChannels() beside the scrub it complements; the ACP run and the
LSP connection both use it, and each proves child delivery end-to-end
(MOCK_ECHO_ENV / LSP_FAKE_ECHO_ENV fixture knobs). Seam + consumer README
rows updated (en+zh, re-recorded). bash-local is already two-channel;
mcp/pty/sdk bypass the seam and only share the scrub.
Review direction (tianyicui, PR #660): in a stacked PR, change all other
process-running places to use the new service.
- lsp-local: LspConnection spawns through ctx.subprocess (piped protocol
streams + a no-spill collected stderr tail); its private process-tree
helpers (POSIX group signalling, Windows taskkill, liveness polling) are
deleted in favor of the seam's handle verbs, and its buildChildEnv now
rides scrubbedParentEnv (LSP children also stop inheriting stale DSH_*).
The plugin injects 'subprocess'; compositions/tests mount
dsh-subprocess-local.
- subagent-acp: the ACP child spawns through the seam (piped ndjson streams,
inherited stderr); spawn failure surfaces through done-rejection into the
same startup race; disposal is handle.dispose with the plugin's configured
graces. dsh-subagent-subprocess is DELETED — its dispose ladder and scrub
are the seam's, and the isolated-config-dir helper had no consumer.
- mcp-client, pty-local, sdk-helper: adopt scrubbedParentEnv as the one
scrub definition (their spawns stay put by ownership: the MCP SDK and
node-pty own those calls; the SDK wizard runs outside any composition).
- Coverage: per-file 100% over every touched src file, with each v8 ignore
carrying a platform or contract reason; new suites cover stdio
dispositions, the dispose ladder tiers, injected-win32 tree semantics,
waitForExit, settled-kill/terminate no-ops, and spawn-failure disposal.
- Docs: consumer-migration Agent Note (en; zh follows in this PR), seam note
updated in place, subprocess.md rewritten for the reshaped vocabulary
(type-equiv re-registered), READMEs and SERVICE_ROLES updated, taskkill
added to knip ignoreBinaries.
Final review pass on the local provider:
- Tear the instance down when `initialize` REJECTS (utf-8 negotiation, malformed
result), not only on abort, so a permanently-rejecting `ready` is never pooled.
- Use the group-aware SIGKILL on a framing failure so helpers are reached.
- Validate maxMessageBytes and maxDocumentBytes positive at load alongside the
other byte caps.
- Fix the location renderer's outside-workspace check to match a `..` segment
exactly, so an in-workspace path like `..generated/a.ts` stays relative.
- Document the accepted ancestor-directory symlink-swap TOCTOU under the
trusted-host model (O_NOFOLLOW guards only the final component).
Further lifecycle/safety hardening of the local provider:
- Tear the instance down when the initialize handshake is aborted, so a
poisoned pending `ready` can't make later queries for that workspace re-wait.
- Make the serialized-queue wait itself abortable, so a query blocked behind
hung earlier work can still observe its own timeout.
- Spawn the server detached and signal the whole process group on teardown, so
helper processes (e.g. tsserver) can't outlive dispose().
- Open the source with O_NOFOLLOW and cap the read at maxDocumentBytes+1, closing
the symlink-swap and concurrent-grow windows the fd-based read left open.
- Honor an already-aborted signal before any host I/O or startup.
- Validate maxStderrBytes positive at load; surface the retained stderr tail in
the "language server exited" error so a fatal startup diagnostic is visible.
Lifecycle and safety fixes from the external review:
- Observe abort while awaiting the initialize handshake, so a server that never
replies can't defeat the tool-timeout signal.
- On an aborted request the server won't cancel, tear the instance down after a
bounded grace instead of releasing the serialized queue with work still live
(prevents overlapping document lifecycles).
- Re-check provider disposal after the canonicalize/read awaits so a query can't
spawn an unowned server after disposeAll().
- Read the source through one open handle (stat + read on the same fd) to close
the realpath-vs-read TOCTOU; decode with a fatal UTF-8 decoder so a legitimate
U+FFFD is not misclassified as invalid.
- Validate and read the source BEFORE spawning a server (pre-start rejection).
- Require an explicit openClose for option-form textDocumentSync.
- Reject nonpositive teardown budgets and non-executable absolute commands at
load; surface unsupported operations as structured LSP_UNSUPPORTED_OPERATION.
- Retain the stderr tail (fatal diagnostics land at exit), not the prefix.
- Catalog the seam vocabulary in docs/core-data-structures/lsp.md.
Implements the LSP capability seam RFC as three packages: dsh-lsp (the
ctx.lsp interface — provider registry by branded id + exclusive extension
mapping, per-query order-independent selection, closed request/result
vocabulary, LspError taxonomy), dsh-lsp-local (a generic stdio language-server
provider — Content-Length JSON-RPC framing, per-(provider, workspace) process
single-flight, transient didOpen/query/didClose, an abortable per-instance
queue, UTF-16 negotiation, host-namespace source reads outside ctx.fs, and
bounded shutdown/kill teardown), and dsh-tool-lsp (the model-facing lsp tool —
four operations, one-based UTF-16 cursor conversion, workspace-grouped location
rendering, hover capping, a required session workspace, and a timeout budget).
Why: an agent had text search and file reads but no way to identify a program
symbol — follow an alias, connect an interface to implementations, or read an
inferred type — before changing code. Splitting model contract, seam, and local
subprocess behavior keeps the four semantic queries stable across future remote
or sandbox-native providers without leaking a JSON-RPC escape hatch.
Tianyi Cui
Dudu-0223