Commit Graph
3 Commits
Author SHA1 Message Date
RRFZandmorluto 5aa1fc5737 fix(process): resolve Windows root-relative commands on the selected drive (#11763)
* 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>
2026-11-18 10:23:14 -05:00
morluto 998faa4118 refactor(ida): launch stdio registrations through the owned MCP transport (#10739)
* refactor(ida): launch stdio registrations through the owned MCP transport

IDA stdio servers now run as an owned process group (POSIX) or job (Windows) through OwnedMcpStdioTransport instead of the SDK's unowned child. Windows bare commands resolve against the registration PATH/PATHEXT and .cmd/.bat launchers run through the shared WindowsBatchCommand cmd.exe invocation. The frame limit is classified from the transport's typed OwnedMcpOutputLimitError instead of parsing SDK error text, detach no longer reaches into the SDK's private _process, and the per-operation byte budget is optional.

* fix(ida): keep owned stdio cleanup observable and resolve Windows interpreter paths

- StructuredContentMcpClient observes SDK-initiated closes like OwnedMcpClient,
  so a failed handshake with incomplete owned cleanup cannot reject unhandled.
- OwnedMcpStdioTransport attaches an owner-redacted captured output to cleanup
  failures; IDA passes its retained stderr tail.
- findWindowsExecutable resolves path-qualified commands at their own location
  with PATHEXT; IDA routes every Windows registration through it and refuses
  non-drive paths with an actionable remedy before the owned launcher.

* fix(process): end a stalled stdio launch when the owned MCP transport closes

OwnedMcpStdioTransport passes the launcher a signal it aborts with OwnerCancellation.ownerClosed when it stops, so IDA's Windows PATH/PATHEXT lookup stalled on an unreachable directory no longer holds close() open; IDA reports the interrupted lookup by its abort reason. The IDA environment tests use createTestTempDirectory for owned temporary roots.

* fix(ida): bound the Windows command lookup by the registration timeout

The PATH/PATHEXT search for a stdio registration now runs under a ProviderStartupDeadline of the registration's timeoutMs that also follows the transport's close signal. A lookup stalled on an unreachable PATH directory ends as an AnalysisTimeoutError (phase windows_path_lookup) instead of leaving connect() pending, matching Apktool's Windows lookup.

* test(ida): stub REA's PATH again now that process inspection runs ps from fixed locations

* test(ida): pass the required registration protocol in owned-stdio fixtures

* fix(ida): report an uninspectable Windows command candidate with its path and code

* refactor(ida): take the host platform from composition, not an ambient read

* fix(verify): pass the host platform to the real IDA upstream check

* fix(process): keep a failed launch's preparation cleanup owned by the stdio transport's close

OwnedMcpStdioTransport discarded the launch rejection before a supervisor
existed, so an ownership-helper preparation that transferred a cleanup owner
(or reported a cleanup failure without one) was dropped and close() reported
success. The transport's close now closes each transferred owner from
processInspectionCleanupFailures, and reports a ProviderCleanupError naming the
preparation resource until it does; a later close retries it.
2026-11-15 19:38:13 +08:00
morluto 422cff00ab refactor(process): give POSIX PATH search order one owner (#10891)
* refactor(process): give POSIX PATH search order one owner

Program lookup, the packet-tool locator and Jadx's Java lookup each split
PATH and resolved empty and relative entries themselves. They now take the
candidates from posixPathCandidates in src/process/executableSearch and keep
only their own probe and failure reporting. A counted architecture guard
keeps new PATH splits from returning; the remaining Windows and Ghidra
searches are baselined.

* refactor(process): move the colon-split PATH searches onto the search owner

REA's launcher-alias lookup and Ghidra's macOS Java lookup split PATH on
":" and skipped relative entries themselves; they now use
posixPathCandidates with relativeEntries: "skip", and the guard counts
colon splits of PATH as well.

* fix(process): produce PATH candidates lazily so an unused entry cannot fail an earlier match

* refactor(process): give Windows PATH/PATHEXT search order the executable search owner (#10916)

* refactor(process): give Windows PATH/PATHEXT search order the executable search owner

* fix(process): keep later PATH entries searchable after a stray quote or a removed working directory

* fix(reverse): refuse a Windows objdump or Rizin command that PATH does not find (#10953)

* fix(reverse): refuse an objdump or Rizin command Windows PATH does not find instead of launching the bare name

* fix(reverse): report a Windows lookup refusal after cancellation and to capability discovery

* fix(desktop): search a Windows host's PATH as Windows does when doctor looks up a program

programOnPath sent every host through the POSIX PATH search, so on Windows a
relative PATH entry reached absoluteCallerPath with platform linux and the
Windows working directory, which resolved against itself until the stack
overflowed. programOnPath now searches PATH and PATHEXT through
windowsPathCandidates on a Windows host, and absoluteCallerPath refuses a
non-POSIX working directory with a TypeError instead of recursing.
2026-11-15 18:24:04 +08:00