* test(mcp): observe cleanup failures at their current delivery boundary
* fix(mcp): preserve workflow completion progress through delivery
* test(setup): keep the Windows Kimi profile fixture on a supported profile
The Windows-only case at `clientConfigurationEnvironment.test.ts:226` wrote
`rea_mcp_input_schema_profile: "full"`. That profile was retired: the reader
now refuses it ("REA_MCP_INPUT_SCHEMA_PROFILE accepts only \"compact\" …"),
so `configureClientConfiguration` returned `failed/readback` before writing
and the case failed on every Windows run, including main's own
`Windows capability and package lane`.
The case has no Windows dependency — it passes an explicit `configPath` and
the Kimi write dialect is the same everywhere — so it now runs on every
platform, which is what let the stale value rot unnoticed. The fixture keeps
the lowercase explicit key the case is about, with the supported value.
Verified on `main` `d6bcfc10`: the same fixture driven through
`configureClientConfiguration(..., "win32")` on this host reproduced
`{"status":"failed","reason":"readback"}` with the retired value and now
returns `{"status":"configured"}` with `compact`. The file's 75 cases pass
here (66 run, 9 platform-conditioned skips) and `check:fast` passes.
* test(ci): restore verifier and MCP contract regressions
* test(captures): assert adapter-owned input issue field names
* fix(ghidra): avoid Java probes for missing installations
---------
Co-authored-by: g0at <50488370+g1at@users.noreply.github.com>
Co-authored-by: IvenKooLab <276572622@qq.com>
Co-authored-by: ShuaixinHuang <197200901+ShuaixinHuang@users.noreply.github.com>
* fix(pcap): resolve tshark and capinfos on the Windows PATH
On Windows, resolvePcapTools kept the bare names "tshark" and "capinfos"
for the launcher to search PATH/PATHEXT. The owned Windows launcher
starts only absolute paths, so it refused the bare name (Win32 87) and
every pcap operation reported "Selected tool tshark could not be
launched (ERR_REA_WINDOWS_NATIVE)", even with tshark on PATH.
Resolve a PATH selection on Windows through the shared PATH/PATHEXT
search, from absolute entries only as on POSIX, and report the same
<tool>_missing failure when no image is found.
* test(pcap): assert typed missing-tool failures
* test(pcap): clarify PATH fixture platform
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(process): resolve Windows root-relative commands on the selected drive
* test(process): pin drive-less PATH entries and UNC root-relative commands
A Windows PATH entry rooted without a drive now counts as relative, so
programOnPath skips it. Pin that with a Windows case, test a
root-relative command under a UNC working directory, and say in the
relativeEntry and windowsPathCandidates docs that drive-less entries
count as relative.
---------
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(process): require a drive or UNC root for selected Windows tools
REA_ADB_PATH, REA_APKTOOL_COMMAND, REA_TSHARK_COMMAND and
REA_CAPINFOS_COMMAND must be absolute. On Windows the check used
win32.isAbsolute, which also accepts \tools\adb.exe: a path rooted on
the current drive. Such a setting names a different file depending on
REA's working drive, and the owned Windows launcher refuses it.
Add isQualifiedCallerPath to the caller-path owner, which on Windows
requires a drive root or a UNC or device root, and use it for all four
settings. POSIX keeps posix.isAbsolute.
* fix(filesystem): require a complete UNC root for a qualified Windows path
Two leading separators were treated as a UNC root, so ///tools/adb.exe,
\\\tools\adb.exe and a bare \\server passed, though win32 parses them
with a single-separator root on the current drive. Use win32's own root:
qualified only when the path is absolute and its root is longer than one
separator.
* test(process): use qualified paths for simulated Windows tools
The JADX bridge emits or removes direct_invokes and direct_invokes_exhaustive together. A protocol item carrying only the coverage flag reached trace_android_references as a reference that claimed true or false coverage over a null site list, while the coverage summary counted it as unknown. The protocol schema now rejects that combination.
Rebased onto current main from morluto/rea#2850. The companion change that removed the null default from the result schema already landed on main.
Co-authored-by: LcokHcak <268763621+LcokHcak@users.noreply.github.com>
* fix(artifacts): classify Gradle version selectors by their grammar
* fix(artifacts): match Gradle ASCII range whitespace
---------
Co-authored-by: ShuaixinHuang <197200901+ShuaixinHuang@users.noreply.github.com>
A tool selection that points at a file Windows cannot start (a DLL, a
script, any non-PE file) was reported as "could not be launched
(ERR_REA_WINDOWS_NATIVE)" by the owned launcher, and as an unrecognised
EFTYPE failure by Node's spawn. POSIX reports the same mistake as
ENOEXEC: "exists but could not be executed".
Treat EFTYPE as not executable, and read the native launcher's
ERROR_BAD_EXE_FORMAT (Win32 193) as EFTYPE, the code Node's own spawn
reports for the same file. Other native codes stay unclassified.
The Windows-only Fx case still read mcp.rea, but a new Fx profile is written
under mcpServers. Two Fx cases assert POSIX modes or a private-directory
close that Windows never performs; they now run off Windows, beside the
existing Windows case. Two release-exposure assertions spelled paths with a
POSIX separator; one now accepts either separator and one uses the reader's
host-file/member spelling.
* fix(native): report undecoded Objective-C category class properties
inspect_native_dispatch_metadata read a category's instance property
list and stopped. A category that also declares class properties kept
`decode.status: "decoded"` and `objc_properties_categories` stayed
complete, although its `_classProperties` list was never read.
When the image info declares the field and the list is present, the
category is now partial with `category_class_properties_not_decoded`
and coverage names the list field, as protocols already do for their
class property list.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* fix(native): keep a category whose class property field is unreadable
Reading the image info flags or the `_classProperties` pointer could
throw inside the category decode, which dropped the whole row and its
already decoded name, methods, protocols and properties.
Both reads now fail into a `category_class_properties_unreadable`
issue at the failing field, and the category stays as a partial row.
Same-named image infos are read together instead of taking the first.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* fix(native): report a truncated image info instead of reading it as absent
A present `__objc_imageinfo` section shorter than its 8 bytes was filtered
out, so a category in that image stayed decoded although whether its record
ends in `_classProperties` is unknown. The category now keeps its decoded
facts and is partial with `category_class_properties_unreadable`, located at
the image info.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* fix(native): resolve the category class property field and read image flags once
A bound `_classProperties` field read as zero and counted as a null list,
and so did a zero whose pointer fixups are unknown. The field is now resolved
through its fixup: a bind is a present list, and an unknown zero keeps the
category partial with `category_class_properties_unreadable`.
The image info declaration is read once per decode instead of once per
category; an image info that declares the field decides.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
* fix(native): list Swift types recorded in __swift5_types2
inspect_native_dispatch_metadata read nominal type records only from
`__swift5_types`. Swift records noncopyable types in `__swift5_types2`
with the same record layout, so those types were missing from
`swift_decls` while `swift_declarations` coverage stayed complete.
Both sections are now read, and each declaration's evidence names the
section its record came from.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* test(native): assert complete Swift declarations across both type sections
The regression fixture left each descriptor's parent unset, so both
declarations were partial and the test asserted only the examined
count. The descriptors now share a module context, and the test asserts
decoded rows, their module and qualified names, and complete coverage.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
fix(tests): treat zombies as stopped in process liveness checks
Fixes#11819.
Killed descendants waiting to be reaped read as alive to
process.kill(pid, 0), so three boundary tests fail in containers
whose PID 1 does not reap orphans. The liveness checks now treat a
zombie as stopped, using the existing observePosixProcess helper
(ps -o stat=, comm-safe) with two small wrappers,
processHasStopped / processGroupHasStopped, swapped in at the three
call sites. Test-only change; production code untouched.
Co-authored-by: morluto <76467478+morluto@users.noreply.github.com>
The ExtraData TerminalBlock ends a Shell Link, but bytes after it were
never examined or mentioned, and coverage stayed complete. Appending a
payload to a shortcut is a common way to hide one.
Record a malformed_structure issue at the first appended byte, with the
byte count, so coverage becomes partial. Shortcuts that end at the
TerminalBlock are unchanged.
Members under package/ are OPC packaging parts, like _rels/ and
[Content_Types].xml; nuget pack writes the core properties .psmdcp there.
Dependencies come from every <dependencies> element. When any <group>
exists, ungrouped <dependency> elements are omitted with a limitation that
counts them, because NuGet's restore-time NuspecReader ignores them.
The desktop-build archive-packing recovery message still sends readers
to docs/installation.md#run-rea-from-a-checkout. The documentation move
put that section in docs/getting-started/installation.md, so the
repository pointer resolved to no file.
Point at the existing section instead; the public VitePress route is
unchanged.
`cliDecimalField` published the domain range on the integer branch and only
the fixed `^[0-9]+$` guard on the decimal-token branch, and JSON Schema
bounds do not constrain strings. A client validating raw argv text against
`--schema` therefore admitted out-of-range tokens such as `--row-limit=0` or
`--block-size=255` that the parser rejects with `VALIDATION_ERROR`.
Publish the domain range on the token branch through the schema-constraint
owner, keeping the CLI's decimal leading-zero convention and the existing
parser diagnostics: the declaration is metadata, so accepted values and
`too_small`/`too_big` messages are unchanged. Non-range refinements such as
the power-of-two entropy block size stay parser-only, and an open-ended
domain still publishes its floor. The getting-started guide states the same
contract.
Fixes#9348
* fix(zip): account 64-bit data descriptors that MSIX packages write
MSIX and APPX packages defer each member's sizes to a signed data
descriptor with 8-byte sizes, but neither header carries a ZIP64 extra
field. The member-region account only tried the width the directory
record implied, so every descriptor counted as unaccounted bytes and the
listing's coverage became unknown.
Try both size widths, the record's own first. A layout still has to
match the recorded CRC and both sizes.
* fix(zip): account an empty member's 64-bit descriptor
An empty member's zero CRC and sizes also match the first 16 bytes of a
signed 24-byte descriptor read as the 32-bit layout, leaving 8 bytes
unaccounted. When several layouts match, prefer the one that ends
exactly at the next accounted byte.
inspect_elf_notes and inspect_elf_debug_link now examine up to 128 note
entries by default instead of 32; the ceiling stays 256. Ubuntu 26.04's
libc.so.6 is examined as 85 entries, 79 of them SystemTap probes, and both
tools refused it with default arguments.
The .nuspec, its OPC relationships and the Maven POM now parse with
lineEndings normalize, as XML 1.0 section 2.11 and the NuGet and Maven
readers do. Text from CRLF files no longer keeps carriage returns.