The capability-seams graph derives its implementation lists from
SERVICE_ROLES in scripts/gen-doc-graphs.ts, which still listed only the
worker-thread backend. Add experimental-code-runtime-python so the
generated graph and table match the registered ctx.codeRuntime
implementations; regenerate docs/capability-seams.md, sync the zh pair,
and re-record the i18n pairing.
Review findings on the CPython backend: an explicit pythonBin path bypassed
the load-time checks (missing/non-executable/directory paths surfaced only
as a run-time worker-exit); a throwing binding member accessor escaped the
fd-3 data callback and terminated the host; the reply drain waited on
'drain' alone, so a pipe destroyed under the wait hung forever; and two
staging-leak assertions diffed a global tmpdir that parallel workers can
perturb.
resolvePythonBin now applies the same accessSync(X_OK) + isFile check to
explicit paths (resolved against the host CWD), and the load error message
distinguishes 'is not an executable regular file' from 'does not resolve on
PATH'. validateBindings snapshots callables into a plain record during run()'s
synchronous validation, turning an accessor throw into the seam-misuse
rejection and fixing the key set the boot frame and dispatch share. The reply
drain waits on drain/close/error together and short-circuits on
proto.destroyed. The staging-leak assertions check the exact paths this test
file staged (recorded by the mocked mkdtempSync) instead of a tmpdir diff.
docs(code-runtime-python): add the alternatives section to the hardening note
docs(config-catalog): refresh the code-runtime-python Config source line
test(code-runtime-python): cover the async spawn-error worker-exit path
The review's carry-over: 'each has a published backend' in the CodeRuntime
JSDoc and its projections (tool-cordis api-catalog, subsystems page) plus
'both shipped'/'backends ship' in the code-runtime README all claimed the
Python backend is released; it is private and experimental, excluded from the
release family. The wording now states the TypeScript backend is released and
the Python backend is experimental and private (not published), in the JSDoc
(api-catalog regenerated to match), the READMEs (paired), and the subsystems
page (paired).
The CPython code runtime's complete public contract is experimental, so it
moves to packages/experimental per the experimental-packages rules: npm name
@deepseek-ai/dsh-experimental-code-runtime-python, private: true, no
publishConfig. All references updated (code-runtime READMEs, config-catalog
and module-graph regenerated with zh alignment, tsconfig paths, doc-standard
and workspace-constraints scripts, the fd-3 and settlement Agent Notes, and
the package README links); md-links and translation pairing pass, and the
suite still runs green.
The earlier merge had adopted master's protocol-only package.json (peer/dev
limited to invariants and cordis, no dependencies), but src/index.ts imports
@deepseek-ai/dsh-code-runtime, dsh-session, dsh-timeout, and schemastery at
runtime — a published lib/index.js could not resolve those bare specifiers.
The manifest now mirrors code-runtime-worker-thread (the five peers, the
schemastery dependency, and the matching dev set); the lockfile, module graph,
and third-party notices are regenerated, and the module-graph zh pair is
re-synced.
The master merge brought a stale generated config-catalog that omitted the
dsh-code-runtime-python config section and mislisted the package. Regenerate
docs/config-catalog.md (verify-config-catalog passes), translate the python
config section into zh, keep the ts config-catalog code blocks verbatim
(untranslated, per the pairing rule), and drop the stray zh Library-packages
line. Corpus-wide verify-translation-pairing passes 1029.
`sendReply` ignored `proto.write`'s `false` return, so a program resolving
several large values in one `asyncio.gather` round encoded every reply in the
same turn and queued all of them in fd 3's writable buffer. Binding resolution
carries no seam-level byte cap to bound that, and the failure kills the host
process rather than failing the run: measured on a 64 KiB-highWaterMark pipe,
eight 4 MiB replies buffered 32.0 MiB at once against 0.0 MiB once paced.
Replies now go through a queue that encodes and writes one frame at a time,
awaiting `drain` when the pipe is full. The encode happens inside the loop, so a
queued reply the run no longer needs is dropped by the `settled` check without
ever being serialized.
This was previously deferred on the grounds that serializing would narrow the
seam's concurrency contract. That reasoning was wrong: the child matches each
reply to its `call` by id from a pump that reads fd 3 continuously, so arrival
order was never observable, and the bindings still run concurrently. Only the
host's peak memory and the flush timing change. The README entry recording the
deferral is removed and the Agent Note records the mechanism instead.
Adding a published Python backend and reordering `flush_line` left several
owning documents stating things that are no longer true.
`src/invariant.ts` justified its empty installer with "ships only the fd-3
wire-protocol codec", which the subprocess execution path contradicts. The
reason now states the actual one: every relation this backend maintains lives
in the CPython child or on the fd-3 wire, so no same-process event sequence is
observable from a listener -- the same shape the sibling worker-thread backend
uses.
The seam's `PORTABLE_RESERVED_WORDS` and `language` JSDoc, the code-runtime
README pair, and docs/subsystems/code-runtime both said only TypeScript has a
published backend. Corrected in all four, with the generated cordis catalog
regenerated for the `language` change.
The note attributed the 12x multiple to the settlement flush holding three
copies. That stopped being true when `flush_line` was reordered to drop the
pending chunks before its push: the binding worst case is the newline path's
single near-budget write. Corrected in the note (both sides) and in the test
comment that repeated it.
The note's Testing section now registers the cases this stack added, and the
Chinese side receives the O(depth) entry it never got plus the new ones -- it
had drifted from the English.
`INTERPRETER_BASELINE_BYTES` argued 64 MiB from a RESIDENT set while RLIMIT_AS
bounds address space. It now cites the bootstrap's own measurement (30.23 MiB
of mappings for `python3 -I`), making 64 MiB roughly twice the measured
baseline.
Also: a hardcoded `(:232-235)` comment reference becomes a reference by name,
a "which now walks in O(depth) too" change narrative becomes a current-state
statement, and a stray double blank line is removed.
The load-time output-budget/addressSpaceMb gate used a worst-case multiple of 8,
assuming two simultaneous ~4x astral copies (the built string and its encode).
Three are live at the peak: on the newline path a single write holds the caller's
text argument, the line slice handed to push, and push's encode copy; the
settlement flush_line path held the pending chunks, their join, and that encode
copy. A budget admitted at 8x (e.g. maxLogBytes 48 MiB against addressSpaceMb 512)
could still OOM the child. The multiple is now 12, the strict `>` is `>=` so a
budget whose peak exactly equals the room left after the interpreter baseline is
rejected (that peak plus the baseline is the whole address space), and flush_line
drops the pending chunks before its push to match the newline path's
join-clear-push order. The child re-check mirror and both note sides move in step;
config-catalog is regenerated from the updated field JSDoc.
The load-time addressSpaceMb gate used a 1/8 fraction derived for ASCII, but the
child ledgers trigger on character count against a serialized-byte budget: an
astral character is one character yet ~4 bytes stored and ~4 encoded, live at
once, so the true worst-case peak is ~8x the budget, not ~2x. Replace the
fraction with an explicit OUTPUT_BUDGET_WORST_CASE_ADDRESS_SPACE_MULTIPLE (8)
and a strict `>`, and gate maxValueBytes the same way as maxLogBytes — the value
path builds and encodes a near-budget completion under the same RLIMIT_AS, so
the incompatible pair was previously admitted there too.
Slice the newline branch's unterminated tail to a budget-sized prefix: it
buffered the whole text[pos:] before the flush trigger could bound it, so an
early newline plus a huge tail made a second full copy of the model's string —
an RLIMIT_AS death the config gate cannot cover since the tail can far exceed
maxLogBytes.
Disclose the cross-field constraint in the maxLogBytes/maxValueBytes/addressSpaceMb
JSDoc (regenerating config-catalog); refresh the note's stale
Buffer.byteLength(JSON.stringify) reference; reconcile the arrival-order rebuttal
with the seam's "in order" logs JSDoc (within-stream, cross-stream best-effort).
Extend the load-rejection test to both budgets and add a tail-copy regression;
sync the zh pair.
Land the PythonCodeRuntime implementation on top of the fd-3 protocol
seam: python3 -I per run, binding namespace over fd 3, RLIMIT_CPU/AS,
wall-clock timer, and SIGTERM->grace->SIGKILL process-group teardown,
with the real-subprocess integration suite.
Fixes three defects surfaced on the source PR's review before they ship:
- boot-write failure resolved a worker-exit through finish()/settle()
that read wallTimer/onAbort/live in their TDZ, rejecting run() instead;
the boot write now runs after those bindings and the v8-ignore that hid
the branch is removed.
- log capture serialized against settlement with no lock while model
daemon threads keep writing; LogBuffer now owns one shared re-entrant
lock taken by write/flush_line/push.
- the fd-3 line residual was a subarray view pinning the whole joined
frame; it is copied into a right-sized Buffer via detachResidual so
pendingBytes measures what is retained.
Review found `127.0.0.2` routed through the proxy. The bypass list carries four
literal loopback entries because that is all a consumer reading an environment
can match, and `proxyForUrl` matched only those — leaving the rest of
`127.0.0.0/8`, `0.0.0.0`, and the IPv4-mapped spellings routed through a proxy
that could then reach them. Loopback is now recognised structurally, which a
list entry cannot express; the published entries stay for the environment
readers.
The same review case exposed a wider one. `web_fetch` skips its address checks
on a proxied hop, because the proxy resolves the origin — but a literal needs no
resolution, so the skip bought nothing and let a proxy on this machine reach
every private range those checks refuse, `169.254.169.254` included. A literal
the checks would refuse now takes the validated path, where the existing
refusal already covers it.
Tests that proved a tunnelled hop used a loopback origin, which no policy can
route through a proxy any more. They name a host only the proxy can answer for
instead — closer to what a proxied request actually looks like.
The user guide promised the proxy carried every outbound request including
telemetry. It carries neither on an older Node, nor anything a model-authored
script sends, so the promise is narrowed and the exceptions listed. A password
in a proxy URL reaching every tool DSH runs is documented there too: it is how
the variable already behaves, and worth knowing before putting one in.
Tianyi Cui