Commit Graph
7 Commits
Author SHA1 Message Date
morluto 3ccb35a8b5 fix(kuna): admit regex settings before acquisition 2026-11-11 20:16:24 +08:00
morluto 3602b64c45 fix(providers): retain cleanup ownership across probes and queued work (#9468)
* fix(providers): retain cleanup ownership across probes and queued work

* fix(jeb): retain primary discovery failures during cleanup

* test(providers): bound cleanup and readiness barriers

* test(jeb): scope availability cleanup fault injection

* test(jeb): retain caller error through cleanup recovery
2026-11-09 06:06:10 +08:00
9512a80804 feat(process): make the regex worker heap request configurable (#6301)
* feat(process): make the regex worker heap request configurable

REA_REGEX_WORKER_HEAP_REQUEST_MIB keeps the 128 MiB default and refuses
values outside 8 to 16384 MiB. A positive Node --max-old-space-size still
overrides the request and is not clamped by the setting.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(process): set the regex worker heap floor to the measured startup bound

On Node 22.23.4 a 5 MiB old-generation request aborts the regex worker
during startup, and 6 MiB starts it and returns a match. The setting floor
is that measured bound.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(process): raise the regex worker heap floor to 7 MiB

6 MiB passes configuration but returns ERR_WORKER_OUT_OF_MEMORY for a
trivial match on supported Linux Node 24 and Node 26.0.0. 7 MiB returned
that match on the measured 22, 24, and 26 lines. A clean-process test
runs the production worker at the floor.

Co-authored-by: morluto <morluto@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
2026-10-22 23:43:13 +08:00
c7efd3a388 feat(config): make the regex matching deadline configurable (#6267)
* feat(config): make the regex matching deadline configurable

REA_REGEX_MATCHING_DEADLINE_MS defaults to 5000 ms and refuses values
outside 1..2147483647. Recovery and capability text name the effective
deadline, its source, and how to change the setting.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* fix(setup): preserve the regex matching deadline in registrations

Setup validated REA_REGEX_MATCHING_DEADLINE_MS but left it out of the
selected registration environment, so a caller override never reached
the launched server and matching fell back to 5000 ms.

Co-authored-by: morluto <morluto@users.noreply.github.com>

* test(config): restore a stubbed regex matching deadline on failure

Kuna deadline assertions now unstub the environment in finally, so a
thrown construction or expectation cannot leave the setting stubbed.

Co-authored-by: morluto <morluto@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
2026-10-22 17:10:51 +08:00
morluto 5927937179 fix(kuna): preserve cancellation and cleanup ownership
Reuse shared startup ownership for query waits and preserve opaque cancellation reasons, captured command facts and typed primary failures. Bind retained command errors to exact process and workspace leases; share close attempts and update earlier caller and aggregate errors after cleanup retry.

Refs: #4700, #4702
2026-10-17 14:38:50 +08:00
Kaoru0822andClaude Opus 5.5 8ab5024397 fix(kuna): stream string scans instead of capping them at 16 MiB (#4318)
* fix(kuna): stream string scans instead of capping them at 16 MiB

`list_strings` and `search_strings` buffered Kuna's whole string JSON and
refused anything above 16 MiB. Kuna writes about 290 bytes per row, so
every app above roughly 10 MB failed with a non-retryable memory error:
a 20 MB Ghostty binary produces 31 MB, Raycast 30 MB and an arm64
Firefox XUL 179 MB.

Parse the scan as Kuna writes it. `runOwnedCommand` accepts a stdout
sink, and `KunaStringInventoryReader` owns the interpretation: it
tokenizes synchronously with stream-json, keeps exact u64 address
tokens, validates each row before retaining it, skips members it does
not interpret, and admits retained rows against live heap headroom
rather than a fixed scan size. The first contradiction stops Kuna
instead of waiting for the rest; an incomplete document is judged only
after Kuna's own exit status. Diagnostics keep the 8 MiB Kuna budget,
now shared with the project export.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(kuna): admit string scan text as it streams and drop sinks with their command

Review found that packed string tokens were assembled whole before the
reader could discard them or check heap headroom, and that a stdout sink
could outlive a cancelled or timed-out command.

Move the token mechanics into `KunaJsonDocumentReader`: keys and numbers
(Kuna's own fields) stay packed, while strings, which carry target-derived
text, arrive in chunks. Ignored strings are dropped unassembled and kept
ones are admitted against live heap headroom as they grow. Kept text is
copied out of the decoded chunk it was sliced from, so retained rows no
longer pin whole chunks (XUL: 278 MB retained before, 156 MB after).
`KunaStringInventoryReader` keeps only the string-scan semantics.

`runOwnedCommand` now ties the stdout pipeline to the command's deadline
signal and aborts it when collection fails, destroying the sink.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(kuna): type the stream-json tokenizer where it is imported

The Reverse tools quality lane type-checks a partial project that reaches
KunaJsonStream without the separate streamJsonParser.d.ts augmentation,
so jsonParser had no declaration there. Declare the tokenizer's shape at
its only import site and remove the augmentation file, so every
compilation of the reader is self-contained. Split two helpers out of
the token dispatcher and simplify the stdout sink wait in
runOwnedCommand, which keeps the added complexity minimal.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(kuna): bound key and number assembly in streamed Kuna JSON

Review found that packed keys (and packed numbers) were assembled whole
by the tokenizer before the reader could discard or constrain them, so a
malformed scan with one enormous key could still exhaust the heap.

Stream every scalar. Keys are assembled only up to 64 code units, past
every field name the reader keeps, so a longer key is skipped unmatched;
kept numbers longer than that are rejected, and ignored numbers are
dropped unassembled. XUL's 611k-row scan parses in the same time.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(kuna): stream the call graph instead of capping it at 32 MiB (#4329)

* fix(kuna): stream the call graph instead of capping it at 32 MiB

Call relationships buffered `kuna decompile-graph` and refused output
above 32 MiB. A 53 MB Raycast binary produces a 157 MB graph (180k
functions, 773k edges), so every call query failed with a non-retryable
memory error after Kuna's minute-long whole-program walk.

`KunaCallGraphReader` validates the graph through the shared streaming
document reader. It retains only the facts queries use (126 MB for
Raycast) and checks edges against the complete node set once Kuna
exits. Unrequested bodies are refused before any of their text is
assembled, through new must-be-null row fields; address fields may be
null, as `forwardsTo` is. Malformed graphs are now `AnalysisOutputError`
like the string scan.

`KunaQueryWork.scan` owns the run-and-validate sequence both producers
shared: diagnostics budget, early rejection and retention of
deterministic overflow.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(kuna): index call-graph edges as they stream

Review found that the reader kept every parsed edge and built the edge
indexes only after streaming, outside any heap admission, so a large
graph could exhaust the heap instead of returning a memory constraint.

Index each edge when it arrives: duplicate pairs and callee orders are
checked per caller, call edges go straight into the caller/callee
indexes, and no edge array is kept, so the whole working set grows under
the per-row admission. Endpoints and order completeness are checked once
all nodes are known, and the validation-only indexes are released then.
Raycast's graph now peaks at 346 MB instead of 503 MB and retains 93 MB
instead of 126 MB, in the same time.

Also drop imports the shared scan made unused.

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* docs(kuna): state the graph's retained size after incremental indexing

Refs #1957

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-15 21:53:59 -07:00
7054bbb9ab feat(kuna): add cached string inventory and search (#4095)
Add cached, reference-free Kuna string inventories through existing CLI/MCP operations. Preserve exact addresses, typed facts, selected Evidence and bounded query lifecycles; validate native fields and caller-specific failures. Verified with real Linux SQLite, zlib, printf and Unicode/empty ELF targets, 194 focused regressions and all 38 CI checks.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>
2026-10-15 13:04:27 +00:00