mirror of
https://github.com/morluto/rea.git
synced 2026-11-16 03:33:25 +00:00
63b0426e409339a5efbc8ea7bbdea0b4502ec935
8
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
63b0426e40 |
fix(errors): rethrow internal invariant violations from plist consumers and the native output projection (#11029)
* fix(plist): rethrow internal invariant violations from plist consumers KeyedArchiveReader reported any non-limit decode failure as a format failure, InterfaceBuilderAnalysis listed it as an invalid archive, ReleaseExposureReader fell back to scanning raw bytes, and the native plist parser reported it as provider output. An InternalInvariantError is a REA defect, not evidence about the input, so each consumer now rethrows it. The native parser's binary path now shares its one cause projection instead of a duplicate. * fix(native): rethrow internal invariant violations from the native output projection nativeOutputFailure wrapped any producer failure as AnalysisOutputError, so a REA defect thrown while reading a plist or dispatching a command was reported as native output parsing. It now rethrows InternalInvariantError, which reaches the provider's adapter-failure path with the invariant as its cause. * fix(artifacts): rethrow internal invariant violations from the shared artifact failure mapper translateArtifactFailure turned every unrecognised cause into an ArtifactOperationError with reason io, so the keyed-archive and Interface Builder rethrows still reached users as artifact read failures. The shared mapper now rethrows InternalInvariantError, and the deferred client reports it as a ProviderAdapterError. Provider-level tests drive both operations through that client. * refactor(dotnet): report managed reader defects through invariantViolation (#11030) * refactor(dotnet): report managed reader defects through invariantViolation The managed readers threw plain TypeError and RangeError for states their callers rule out: an allowance or method page outside the input schema, a missing snapshot after identity admission, and a missing CLI header after the caller already returned for it. These are REA defects, not malformed targets, so they throw invariantViolation. The two CLI re-checks are gone: the callers pass the CLI header they already narrowed. The dotnet parser-plain-throw baseline drops from 6 to 0. * refactor(providers): refuse unusable tool paths and record marker mismatches without throwing to their own catch (#11031) * refactor(firmware): refuse unusable tool paths without throwing to its own catch resolveFirmwareCommand threw a plain TypeError for a configured engine or limiter path that is not a regular file, only to catch it in the next line and wrap it as an unavailable capability. A helper now returns why a path cannot run, and each caller builds the one refusal directly. Refusal reasons are unchanged; the directory tests now pin them exactly. The firmware parser-plain-throw baseline drops from 2 to 0. * refactor(pwndbg): refuse unusable debugger paths without throwing to its own catch inspectPwndbgCore threw plain Errors for a relative or wrong-kind configured path only to catch them and wrap them as an unavailable capability. A helper now returns why a path cannot be used, with any host error as its cause, and the loop builds the one refusal. Cancellation between host checks still propagates. Refusal reasons are unchanged. The PwndbgCoreContext parser-plain-throw baseline entry is removed. * refactor(pwntools): record a mismatched failure marker without throwing to its own catch readPwntoolsFailureEvidence threw a plain Error when the private bridge marker disagreed with the exit status, only to catch it and record its message as the marker failure. The mismatch is now recorded directly. The recorded evidence is unchanged; a unit test pins both the matching and the mismatched marker. The PwntoolsResourceLimits parser-plain-throw baseline entry is removed. * fix(android): report an unobserved Java probe as a REA defect, not a Java refusal (#11032) * fix(android): report an unobserved Java probe as a REA defect, not a Java refusal When runOwnedCommand resolved without observing the completed run, the probe threw a plain Error that its catch then classified as a Java runtime refusal, telling the caller to review their launcher. The state is a broken internal assumption, so the probe throws invariantViolation and projects it as a jadx adapter failure. The JadxJavaProbe parser-plain-throw baseline entry is removed. * refactor(native): report native provider defects through invariantViolation (#11033) * refactor(native): report native provider defects through invariantViolation Native providers threw plain Error and TypeError for states their own earlier steps rule out: a client with no target, a measured slice with no selection, a read longer than requested, a decoder command without its runtime, an LLDB root released before its children, and an IPS integer fact from another reader. These are REA defects, so they throw invariantViolation. NativeUiInterruptionControl no longer needs a throw: recording the deadline interruption returns the interruption its error is built from. Six native parser-plain-throw baseline entries drop. * refactor(native): report malformed lipo, swift-demangle and UI helper output through ParseFailure (#11034) * refactor(native): report malformed lipo and swift-demangle output through ParseFailure The lipo and swift-demangle output parsers threw plain TypeError for output that breaks their framing, and the lipo result classifier caught every TypeError as malformed output, so a host TypeError from a REA defect was reported as lipo output. Both parsers now throw malformedInput with their own format, and the lipo classifier accepts ParseFailureError instead of TypeError. Messages and the consumers' output and format failure categories are unchanged. Four native parser-plain-throw baseline entries drop. * refactor(native): report the UI helper's non-JSON output through ParseFailure The native UI helper runtime threw a plain Error when the helper's stdout was not JSON. It now throws malformedInput with its own format and the JSON failure as cause. The observation boundary still reports it as a native-macos adapter failure with the same reason. The NativeUiHelperRuntime parser-plain-throw baseline entry is removed. * refactor(verify): keep process-lifecycle owners out of the parser plain-throw guard (#11035) * refactor(verify): keep process-lifecycle owners out of the parser plain-throw guard JadxWorkspaceRoot and PwntoolsDecoder sit in parser folders but own a private helper process; their remaining plain throws report the helper's startup, handshake and cleanup outcomes, not parse faults. The guard now names them as process-lifecycle owners and leaves them out, so their five baseline entries drop without inventing parse errors for process failures. * refactor(parsers): report the remaining artifact, il2cpp, model and domain defects through invariantViolation (#11036) * refactor(parsers): report the remaining artifact, il2cpp, model and domain defects through invariantViolation Nine plain Error, TypeError and RangeError throws in parser folders reported broken internal assumptions or caller contracts (an uninitialized ZIP or DMG reader, a missing scan result, a non-byte stream chunk, an invalid traversal or marshal limit, a header count or linking fact the reader itself should have produced, a misordered Safetensors budget, an append after fingerprinting). None is an input fault, and no catch site depends on their class. They now throw invariantViolation with the same messages, and their parser-plain-throw baseline entries drop. * refactor(evidence): refuse an earlier Evidence format through ParseFailure (#11037) * refactor(evidence): refuse an earlier Evidence format through ParseFailure UnsupportedEvidenceFormatError restated the ParseFailure algebra as its own TypeError subclass. The format owner now throws unsupportedInput for the evidence-bundle format at the evidence_format field and exports the predicate its consumers use. The bundle reader's unsupported-format kind and message are unchanged. The snapshot reader surfaced only TypeError messages, so it now also surfaces the format refusal and keeps its recreate guidance; a boundary test pins that. * refactor(native): report Mach-O reader and inspection defects through invariantViolation (#11042) The Mach-O inspection plan reads the lipo, nm and dyld_info captures it just ran, the dispatch metadata reader is called only for Mach-O executables, and the command-name reader takes limits its callers derive. A missing capture, a non-Mach-O target or a malformed limit is a REA defect, not an input fault, so these six sites throw invariantViolation with their existing messages instead of TypeError. No catch site depends on their class. Their parser-plain-throw baseline entries drop. * fix(artifacts): join the Interface Builder cleanup failure to a catalog defect When catalog work failed while the concurrent Interface Builder reader failed to close, analyzeAppleAssetCatalogs built the primary through translateArtifactFailure before joining cleanup. That mapper now rethrows an InternalInvariantError, so a defect skipped the join and the reader's retained resources were lost. The defect now becomes the artifact graph provider's adapter failure at that site and joins the cleanup failure as before. |
||
|
|
cb4e5cfbff |
feat(analysis): make resource budgets recoverable across CLI and MCP (#9706)
* refactor(analysis): centralize resource budgets and recovery facts * feat(analysis): expose ASAR and Ghidra runtime admission controls * feat(artifacts): expose reader admission and preserve limited recognition * feat(javascript): expose recoverable semantic precision budgets * feat(web): add recoverable source-map budgets and compact retained views * feat(web): make historical capture decoding budgets recoverable * fix(desktop): enforce selected view policies across capture and export * feat(runtime): make Frida and LLDB retention budgets selectable * feat(native): expose diagnostic budgets and preserve exhaustion causes * feat(managed): expose recoverable snapshot and representation budgets * feat(metadata): make Go and Rust inspection budgets recoverable * feat(wasm): expose representation and producer admissions * feat(evm): expose per-call worker resource budgets * feat(android): make resource and string-pool omissions recoverable * feat(producers): expose command and recorded-capture budgets * docs(analysis): record recovery contracts and integrated validation * fix(docs): link calibration oracle from published verification pages * fix(docs): restore desktop task anchors and receipt links * test(analysis): align fixtures with selectable budget contracts * fix(analysis): remove source-location projector dependency cycle * test(analysis): refresh budget recovery and verifier regressions * test(analysis): preserve format and publication recovery expectations * test(windows): align conformance with owned runtime admission * fix(desktop): preserve selected alias kinds in exported state * test(analysis): align provider fixtures with recoverable budgets * test(desktop): validate emitted alias metadata structurally * fix(ghidra): classify bounded Windows endpoint refusals * test(cli): assert selectable artifact admission recovery * fix(cli): preserve capture provenance in environment builds |
||
|
|
371f3d91ae |
fix(managed): retain type discovery after member limits (#5312)
fix(managed): retain types after member limits Admit type identity and a bounded recovery envelope before member decoding so callers can discover smaller selections after capacity exhaustion. Fixes #5146 Co-authored-by: eco-null <eco-null@users.noreply.github.com> Co-authored-by: morluto <76467478+morluto@users.noreply.github.com> |
||
|
|
435efe7afc |
fix(frida): share exact JSON string byte accounting
Count the serialized JSON representation at its owner, including escaped lone UTF-16 surrogates. Frida now retains the correct bounded prefix and reports truncation in script results and subsequent session Evidence. Move the existing exact managed string counter into the domain layer and use it for Frida, managed projections and Ghidra inline PDB caveats. Remove the superseded counters while keeping Ghidra's bounded length guards and NativeAOT's distinct raw-byte admission before string allocation. Validated with 296 focused tests, check, docs and exact test discovery. The Unicode oracle checks every UTF-16 code unit and adjacent fragments against native JSON encoding. Frida execution uses simulated native handles through the real MCP handlers/client; real target instrumentation is unverified. Fixes #5097 Co-authored-by: Prabhuram Karki <prabhuramkarki07@gmail.com> |
||
|
|
bddeae5108 |
fix(managed): charge admitted member facts once in the final projection (#3487)
* fix(managed): charge admitted member facts once in the final projection inspect_managed_members admits method bodies, signatures, call edges, field accesses and user strings as they are produced, then admits the whole result again before returning it. The final pass re-counted every already-admitted record, so the 64 MiB representation budget was spent at roughly twice the result's JSON size. Ordinary assemblies such as MathNet.Numerics, MongoDB.Driver, Npgsql, ClosedXML, System.Reactive and Azure.Storage.Blobs returned resource-limit with no members. The budget now remembers each admitted projection. A containing projection spends that charge once instead of counting the subtree again; a value that serializes more than once still pays for every further occurrence. Results that already fit are byte-identical. Refs #3449 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * test(managed): check a repeated occurrence pays with fresh credit Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
23d50ea3be |
fix(managed): bind row budget charges to metadata identities (#1897)
Charge fixed row bookkeeping once per table and row within an admitted metadata layout. Repeated owner lookups no longer permanently consume capacity for temporary cursors; distinct rows and failed admissions retain the existing representation limit and typed failure behavior. Verify repeated local attribute-constructor lookups through public CLI metadata/member inspection and preserve the remaining capacity guards. Closes #1878 Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com> Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> |
||
|
|
30787297d4 |
fix(managed): charge retained member facts instead of decoding bounds (#1768)
* fix(managed): charge retained member facts instead of decoding bounds inspect_managed_members returned no types or methods for ordinary assemblies: Newtonsoft.Json 13.0.3 (netstandard2.0, 705 KB) exceeded the 64 MiB representation budget. Each method body reserved 256 bytes per IL byte and each signature 256 bytes per blob byte before decoding, and those worst-case bounds were never returned, so the budget scaled with total IL size rather than with the retained output. The final projection admission also charged every decoded string at six bytes per character and every scalar at 32 bytes, about eight times its JSON size. Bound IL, exception-clause and signature decoding with a transient admission that does not consume capacity, then reserve the decoded body or signature that is actually retained. Projection admission now charges the exact UTF-8 JSON bytes of decoded strings, keys and scalars; escaped and multibyte text still counts at its encoded size. Newtonsoft.Json now yields 496 types, 4185 methods and 18846 call edges. Assemblies whose retained facts exceed the budget, such as the 2.3 MB System.Private.ServiceModel 4.7.0, remain unavailable. * fix(managed): bound concurrent decoding and count JSON bytes in place Review follow-up. A method's IL bound and its exception-section bounds were each checked against the same remaining capacity although their bookkeeping is live at the same time, so 40 MiB of IL plus 40 MiB of exception clauses passed a 64 MiB budget. Transient admissions now accumulate until the method body or signature is built, and reserve respects bounds that are still live. Projection admission also serialized each decoded string to measure it, which could allocate about six times a near-limit string before the check. It now counts the exact escaped UTF-8 bytes without building the JSON text. * fix(managed): admit derived call and field edges as they are built Review follow-up. Released IL bounds no longer stand in for the call and field-access records that edges() derives from anchors. Each record is now admitted before the graph grows further, so a crafted assembly cannot materialize the whole derived graph before the representation guard fails. * fix(managed): admit derived caller names before building them Review follow-up. The caller name of call and field-access records was interpolated per anchor before any admission saw it. It is now built once per method, after its exact JSON size, counted from the type and member names, has been admitted. |
||
|
|
0f0d6e4d43 |
fix(managed): budget decoding before representation expansion
Validation: integrated changes passed formatting, typecheck, lint, resolved module-boundary verification and git diff --check. Tests and real-provider workflows were not run. |