mirror of
https://github.com/morluto/rea.git
synced 2026-11-03 05:16:16 +00:00
1a48a82020
* feat(ghidra): bound one procedure's decompile without closing the session Fixes #5340 * fix(ghidra): report decompile limit as a request timeout Keeps BinarySession from marking every Ghidra operation unavailable after one procedure's decompile reaches the limit, and names the MCP tool-call deadline in the remediation. * fix(ghidra): budget the decompile limit and report cancellation-closed sessions Adds REA_GHIDRA_DECOMPILE_TIMEOUT_SECONDS to the derived Codex/Kimi tool deadline and forwards it through setup. Marks a remote failure that settles after caller cancellation so the timeout projection no longer promises session reuse. * feat(ghidra)!: default analyze_function to a retained summary analyze_function was the one analysis tool never wired into REA's detail: summary|complete convention, so it always returned the complete native dossier inline (0.2-10 MB, dominated by value-flow and references). Agents on tool-result-capped harnesses never reached the pseudocode. Add `detail` to analyze_function, defaulting to summary: return a compact dossier (procedure identity, pseudocode size, facet counts, value-flow availability) and retain the complete dossier Evidence in the session. Page pseudocode, references, assembly and value-flow through normalized_result.parent_evidence_id with inspect_analysis_view, which already projects every native facet. detail: complete returns the full dossier inline as before. Delivery detail never reaches the provider or Evidence identity. BREAKING CHANGE: analyze_function now defaults to the compact summary view instead of the complete dossier inline. Pass detail: "complete" for the previous behavior. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * feat(cli): expose analyze_function detail on the function command Wire `rea function` into the shared CLI detail-delivery convention (--detail, --evidence_output) so the CLI and MCP adapters offer the same summary | complete choice for analyze_function. Summary delivery writes the complete dossier Evidence to a bundle and returns the compact summary view; it fails before analysis when --evidence_output is missing, so no provider work is wasted. The CLI default stays complete because a one-shot command has no in-process Evidence retention. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * refactor(ghidra): harden analyze_function summary after independent review FlatlineRoundtable cross-model review (6 lanes) converged on the summary projection's availability semantics being underspecified. Document that facet_availability is authoritative over raw pseudocode length, that code_units are UTF-16 (the unit inspect_analysis_view pages by), and that value_flow.truncated is meaningful only when value-flow is available. Parenthesize the nullish-coalescing/comparison for clarity. No behavior change; declined claims (value-flow "truncated collapse", unresolved_calls dead code) were schema-impossible or already correct. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * test(ghidra): cover analyze_function detail wiring on both adapters Close the roundtable's top convergent finding: the summary path had only domain-projection coverage, not the server/CLI wiring where the breaking behavior lives. MCP server (registerEnhancedTools): the default delivers a retained summary whose complete parent is recorded for paging; delivery detail is stripped before the provider request; detail: complete returns the full dossier inline; a registration without retention rejects a summary request before any provider work. CLI (rea function): the default returns the complete dossier; --detail summary without --evidence_output fails before analysis; with the bundle sink it returns the compact summary and writes the complete Evidence bundle. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(ghidra): keep analyze_function delivery detail out of provider inputs Address the five automated review findings on #5660. P1: `detail` lived in the shared enhancedInputSchemas.analyze_function, so inspect_native_api composed `{procedure, detail: "summary"}` and the strict Ghidra analyze_function schema rejected it. Delivery detail is now owned by the analyze_function MCP tool contract; the handler separates it before any provider call, and executeFunctionAnalysisEvidence parses the provider input only. The MCP regression test validates provider input exactly as the production Ghidra client does and exercises inspect_native_api. P2: the analyze_function summary Evidence now carries the parent's relationship_coverage like native facet views, so partial or unknown caller/callee discovery is not reported as complete. P2: the summary reports the dossier's declared facet_availability, nulls the counts of facets the provider declared unavailable, and lists each reason in unknowns, so Kuna's empty references, comments and basic blocks are not mistaken for observed absence. P2: inspect-analysis-view evidence-file sources accept the Evidence bundle that --evidence_output writes, selecting a record with source.evidence_id when the bundle holds several; the CLI test reads the saved bundle back through that route. Tests that assert the complete dossier over public analyze_function now pass detail complete, and the oversized-dossier test asserts the summary view it previously expected to be unsupported. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(ghidra): keep omitted function facets unknown in the summary The analyze_function summary coerced an omitted unresolved_calls record to a count of 0 while the native facet view reports the same dossier as "not recorded", and it carried no counts at all for the pageable swift_string_literals facets. Count optional records as null when the dossier omits them, name them in unknowns, and add nullable counts for both Swift literal facets. The summary Evidence envelope also dropped the function address: createAnalysisViewEvidence now derives one function-view predicate for both the address location and relationship coverage. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(scripts): request complete analyze_function dossiers in real-provider checks The real-provider check scripts call analyze_function over MCP and then consume the complete dossier: requireFunctionDossier, functionDossierSchema, .procedure/.pseudocode reads, operation assertions, compare_functions inputs and the oversized-dossier rejection path. The summary default broke them. Every such call site now passes detail: "complete". Provider-direct callers are unchanged because detail is an MCP contract field; the admission-error sites stay unchanged because provider failures return before the summary branch. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(ghidra): keep unavailable value-flow counts unknown in the summary summarizeDossier coerced the five value_flow_* counts to zero when native_value_flow was null (IDA, Kuna) or declared unavailable (Ghidra), so the summary reported empty collections while value_flow.available said otherwise. The counts are now null unless value flow is available, with the reason already named in unknowns. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(evidence): budget authentication headroom by file shape and serialized size The evidence-file headroom reserved two parsed copies, calibrated for one record, and charged strings at two bytes per character. Bundle authentication additionally holds the schema-parsed bundle, the re-parsed records and two JSON.stringify results, which expand control characters and lone surrogates sixfold: a bundle whose raw_result is a million NUL characters estimated at 2 MB parsed but needed 12 MB of serialized text. The budget now has a parsed term and a serialized term per file shape, owned by EvidenceAuthenticationHeadroom and shared by the analysis-view reader and the bundle importer, which had no guard. Two transient peaks the budget did not name are removed at their producers: the canonical digest hashes a large leaf directly instead of concatenating it into its buffer, and the JSON validator walks containers by cursor so a wide array no longer queues one entry per member. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(mcp): route analyze_function summary replies to dossier consumers in docs and prompts The MCP guide, the version-comparison and crash-triage prompt contracts, the authored skill and the Kuna, IDA and function-matching docs still told agents to pass analyze_function replies straight to compare_functions or build_call_path. Under the summary default those consumers reject the reply. They now name the retained-evidence reference for compare_functions, the observations fields of match_native_functions, and detail: "complete" for build_call_path, which accepts only inline dossiers. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * test(evidence): pin validator member order and keep interrupted validation yielding Independent review of the cursor validator found the interrupted-validation fixture relied on the old last-first walk: its over-deep head was reached before the first cooperative yield. The padding now precedes the deep element so the walk yields first, which is the invariant the test guards. A new case pins first-defect-in-member-order reporting, skipped non-enumerable keys no longer count toward the yield cadence, and the import-headroom test loads the reader through a fresh module graph so it holds under the services project's shared workers. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AoJZDPcrgmMsCqLG4LfiFb * fix(ghidra): align deadline bounds and cancellation outcomes * refactor(ghidra): shorten deadline recovery guidance * fix(mcp): clarify the scope of inferred decompile deadlines * test(ghidra): verify deadline workflows and document measured costs Separate failure and registration cases by their distinct contracts. Record actual Linux Ghidra, CLI and MCP evidence with measured context, elapsed time, CPU and memory, and explicit remaining coverage limits. * docs(ghidra): pin integration and distinguish baseline format failures * fix(ghidra): own failure lifecycle wording in the provider * test(ghidra): budget timeout fixture generation before allocation * fix(analysis): preserve summary provenance and bound bundle guidance * fix(ghidra): align capability and traversal deadline guidance * style(config): align deadline guidance formatting * docs(ghidra): record capability and traversal verification * docs(analysis): use absolute paths for Evidence files * docs(analysis): record completed function summary measurements * docs(analysis): update main integration validation --------- Co-authored-by: Syed Rishad <rishadzzzz@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>
150 lines
8.6 KiB
Markdown
150 lines
8.6 KiB
Markdown
# Native function summary verification — 2026-10-23
|
|
|
|
## Behavior and source
|
|
|
|
MCP `analyze_function` defaults to a compact derived summary. The complete
|
|
function Evidence stays in the session and can be selected by the returned
|
|
`parent_evidence_id` through `inspect_analysis_view`. `detail: "complete"`
|
|
keeps the historical full reply. Delivery detail stays outside provider inputs
|
|
and parent Evidence identity. This MCP default is a breaking change.
|
|
|
|
CLI `rea function` still defaults to complete output. Opting into summary
|
|
requires an exclusive `--evidence_output` bundle destination. The authenticated
|
|
bundle can be paged by an absolute `evidence-file` source and explicit record
|
|
ID after the provider closes. Plain records also honor an explicit ID.
|
|
|
|
The implementation reuses the existing JSON authentication/headroom owner and
|
|
one bundle-authentication function for import and file views. It reads one file
|
|
once per request. Ambiguous-bundle guidance retains complete IDs and bounds
|
|
JSON-encoded reference previews to 2 KiB, omitting oversized operation labels.
|
|
Native summaries inherit function locations and relationship coverage; omitted
|
|
or unavailable facts stay unknown rather than becoming zero.
|
|
|
|
Measured source:
|
|
|
|
- Before: `fb6c88d518ab9b84ed396cf82dcb1c28a608c501`.
|
|
- After: `4e3d930ada93014efa43610265b58307de49a687`.
|
|
- Harness SHA-256: `d67b23db21024361b34f796add1ff91bf3fbcda1991ea9e369ef8bcf64c71ec4`.
|
|
- Linux x86-64; pinned Node 24.22.1 / npm 11.16.0; bring-your-own Ghidra 12.1.4
|
|
and JDK 21.0.12.1+1. No target execution.
|
|
- Target SHA-256:
|
|
`9c49d2eef9c2083f6f66f38de4ed0ee1bac17ca5b73dc8f3a90d9ba62a8d1d5d`.
|
|
Its 3,000-branch C source is the timeout verifier fixture, SHA-256
|
|
`f99074876faae8e18e45265dab007b6fcc396fe909d57efab72adc8925a960c7`.
|
|
Known entries `0x1010f9` and `0x128ed6` apply only to this exact artifact.
|
|
|
|
## Completed public MCP comparison
|
|
|
|
One fresh production stdio MCP client discovers all 221 tools, opens the target,
|
|
then obtains two answers: the small function's identity, availability/counts and
|
|
complete pseudocode; and the large function's identity, availability, assembly/
|
|
block counts and first 20 assembly rows. It repeats both reads in the same
|
|
session, performs one matched complete-small compatibility read and closes.
|
|
|
|
Before, each default function reply carries the full dossier. After, each
|
|
answer uses the default summary and one retained-Evidence page. Both lanes use
|
|
a one-second native decompile limit, an explicit 600-second SDK tool deadline
|
|
and matching 128 MiB server/client receive budgets. Selected facts match
|
|
between implementations and rounds. The large function has 45,007 assembly
|
|
rows and 9,001 blocks with pseudocode unavailable; the small pseudocode is
|
|
64 bytes, SHA-256
|
|
`c947ff22cab1ffef9b962b12d82e12d0e5f5e2a6c5634a2ed5ea3c353e1d5807`.
|
|
Each run retained the same observed JVM start identity through its tasks and
|
|
completed cleanup with an unchanged target digest.
|
|
|
|
| Whole benchmark | Before | After |
|
|
| --------------------------------- | ---------: | --------: |
|
|
| Estimated exchange tokens | 44,217,047 | 1,033,322 |
|
|
| Elapsed seconds | 237.715 | 128.600 |
|
|
| Waited process-family CPU seconds | 211.656 | 123.083 |
|
|
| Sampled family RSS MiB | 3951.7 | 2827.8 |
|
|
| Catalog/tool exchanges | 8 | 12 |
|
|
|
|
Estimated context falls 97.66% for this completed task, including the larger catalog and
|
|
four extra page calls. The catalog alone grows from 982,457 to 985,662 estimated
|
|
tokens. The complete-small compatibility read remains included in both totals.
|
|
|
|
| MCP phase elapsed seconds | Before | After |
|
|
| ---------------------------------------------------- | ------: | ------: |
|
|
| First small function, including deferred JVM startup | 80.3747 | 69.0745 |
|
|
| First large function, with the JVM already warm | 57.2394 | 19.8205 |
|
|
| Repeated small function | 0.0534 | 0.0601 |
|
|
| Repeated large function | 56.4148 | 19.7744 |
|
|
|
|
The repeated small task adds a page call and takes about 6.7 ms longer in this
|
|
pair. Large partial dossiers remain live and retry native analysis; healthy
|
|
small-function snapshot reuse remains available. This change reduces agent
|
|
output and its serialization/transport work while still producing and retaining
|
|
the complete parent dossier.
|
|
|
|
Tokens are `tokenx` 1.3.0 estimates over the actual catalog, arguments and whole
|
|
SDK results, including both text and structured projections. Protocol framing,
|
|
handshake, prompts, client rendering/replay and actual model usage are
|
|
unmeasured. The whole benchmark includes token estimation, hashing, resource
|
|
sampling and exchange-file writes. Summed measured MCP phases exclude the
|
|
subsequent counting/writing overhead: 217.327 → 127.538 seconds.
|
|
The whole-benchmark timings must not be presented as pure provider analysis time.
|
|
|
|
Node heaps were 2 GiB, Ghidra's heap 1 GiB, affinity CPUs 0/1 and Java active
|
|
processor count two. The observed-family early-stop guard allowed 6 GiB RSS,
|
|
600 seconds CPU and 900 seconds elapsed; neither completed run hit a guard.
|
|
RSS was sampled every 50 ms through recursive `/proc` observations. Shared
|
|
pages can be double-counted and brief/unobserved child transitions can be
|
|
missed; this is not a kernel cgroup peak or hard family memory limit. Waited
|
|
CPU includes reaped descendants. Baseline ran first, OS caches were not flushed
|
|
and the host was shared with other work. This is one paired investigation,
|
|
with cold/result-cache distinctions above, rather than a universal speed claim.
|
|
|
|
An earlier baseline attempt used a 240-second SDK deadline and began with the
|
|
large function. It timed out, then closed successfully; the whole attempt took
|
|
280.628 seconds and 219.562 seconds waited CPU, with no resource-guard stop.
|
|
Its failed receipt/exchanges remain separate. It is excluded from completed-task
|
|
comparisons. The final pair uses the same longer deadline and task order on
|
|
both implementations.
|
|
|
|
## Public CLI and retained files
|
|
|
|
The real default CLI command returned a complete small-function dossier with
|
|
the expected pseudocode (70.627 s elapsed, 70.034 s CPU). The first verification
|
|
harness assumed bare Evidence and failed to unwrap the successful public
|
|
`ok`/`data`/`meta` envelope. After correcting the harness, its completed receipt
|
|
and output were verified and reused without another provider analysis.
|
|
|
|
The summary CLI command analyzed the large function and wrote a 29,141,012-byte
|
|
bundle. It returned an 8,499-byte summary envelope in 78.163 seconds, with
|
|
94.196 seconds CPU and 2,381.2 MiB sampled family RSS. Two fresh CLI processes,
|
|
with the Ghidra installation path unavailable, authenticated that bundle and
|
|
returned its first and final 20 assembly rows. Coverage reports 45,007 total
|
|
rows, correct continuation and exhaustion. The target digest remained unchanged.
|
|
|
|
Those independent file-page commands took 17.218 / 18.175 seconds, 15.150 /
|
|
17.015 seconds CPU and about 666 / 669 MiB RSS. Each fresh command reads and
|
|
authenticates the complete bundle; this is not sublinear disk access or a
|
|
cross-process cache. Complete evidence remains available at the caller's file.
|
|
|
|
## Validation and limits
|
|
|
|
- Full `npm run check` passed after integration with main
|
|
`a062529bb2aa064be37e82df39c59c4fa798fe50` at source `b7c3f0cd79c4a02b135da56fff06de08dd456ff7`.
|
|
`docs:check` also passed.
|
|
- The latest focused contract/progress/snapshot/local-path run passed 139 tests;
|
|
the measured integration's selected run passed 136 and an earlier broader
|
|
CLI-selector/contract run passed 169. Overlaps are not summed. The earlier domain/Evidence run had 87 passes and one legacy
|
|
progress-fixture failure. The explicit-complete fixture and new summary's
|
|
increasing notifications subsequently passed in a separate eight-case MCP
|
|
run.
|
|
- Actual SDK/schema checks passed all 15 cases after regenerating ignored test
|
|
catalog metadata. The initial run had 14 passes and one stale-catalog
|
|
equality failure; its receipt is retained.
|
|
- The production CLI/MCP observations above use real Linux Ghidra. Hopper,
|
|
IDA, Kuna, macOS and Windows behavior is covered only to the extent of the
|
|
selected source/boundary tests, not new real-provider verification.
|
|
- Earlier source integration fixes and later main changes were not included in
|
|
every historical test/performance pin. The measured pair remains tied to the
|
|
explicit source commits above.
|
|
|
|
Raw exchanges, protocol observations, control failures, scripts and resource
|
|
receipts are retained under `/tmp/rea-maintenance-20261023/5660/`. Provider
|
|
projects, generated binaries, build outputs and local planning files are not
|
|
committed.
|