attocode-code-intel 0.1.7__tar.gz → 0.1.8__tar.gz

{attocode_code_intel-0.1.7 → attocode_code_intel-0.1.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: attocode-code-intel
-Version: 0.1.7
+Version: 0.1.8
 Summary: Local code search and navigation for coding agents and people
 Project-URL: Homepage, https://github.com/eren23/attocode
 Project-URL: Repository, https://github.com/eren23/attocode
@@ -140,7 +140,8 @@ To use the latest source, install from a clone of this repository: `uv tool inst
 Version 0.1.7 also provides `changes` and `trace` commands: investigate a diff,
 follow source-backed caller paths, export JSON/Markdown, and check or refresh saved
 evidence. MCP equivalents `change_context` and `trace_flow` are local full-profile
-tools. See the [change investigation guide](https://github.com/eren23/attocode/blob/main/docs/guides/change-investigation.md).
+tools. Version 0.1.8 adds exact file/symbol selection and bounded changed-class
+summaries, with `--class-mode expand` for class-wide callers. See the [change investigation guide](https://github.com/eren23/attocode/blob/main/docs/guides/change-investigation.md).
 
 The agent sees 16 tools (the `daily` profile):
 

{attocode_code_intel-0.1.7 → attocode_code_intel-0.1.8}/README.md

@@ -27,7 +27,8 @@ To use the latest source, install from a clone of this repository: `uv tool inst
 Version 0.1.7 also provides `changes` and `trace` commands: investigate a diff,
 follow source-backed caller paths, export JSON/Markdown, and check or refresh saved
 evidence. MCP equivalents `change_context` and `trace_flow` are local full-profile
-tools. See the [change investigation guide](https://github.com/eren23/attocode/blob/main/docs/guides/change-investigation.md).
+tools. Version 0.1.8 adds exact file/symbol selection and bounded changed-class
+summaries, with `--class-mode expand` for class-wide callers. See the [change investigation guide](https://github.com/eren23/attocode/blob/main/docs/guides/change-investigation.md).
 
 The agent sees 16 tools (the `daily` profile):
 

attocode_code_intel-0.1.8/evals/designs/2026-10-17-change-workflow-replays.json

@@ -0,0 +1,98 @@
+{
+  "engine_version": "0.1.7",
+  "source_commit": "7b09e28946dad07448d4f8a155fa2be8c6defb4d",
+  "scope": "Selected production/regression files replayed on complete actual parent trees; other original PR files are excluded. No input-project code execution. Not an agent study or fresh unseen tasks.",
+  "known_labels": "Source-audited selected examples, not exhaustive relevance judgments; Python stdlib AST independently locates exact preselected call sites.",
+  "limits": {
+    "scratch_bytes": 1073741824,
+    "process_tree_rss_bytes": 2147483648,
+    "seconds": 600,
+    "warm_samples_per_task": 5
+  },
+  "tasks": [
+    {
+      "id": "workspace-cache",
+      "commit": "7b09e28946dad07448d4f8a155fa2be8c6defb4d",
+      "file": "packages/code-intel/src/attocode_intel/tools/search_tools.py",
+      "symbol": "_get_trigram_index",
+      "patch_files": [
+        "packages/code-intel/src/attocode_intel/tools/search_tools.py",
+        "packages/code-intel/tests/test_trigram_workspace.py"
+      ],
+      "known_callers": [
+        {
+          "file": "packages/code-intel/src/attocode_intel/tools/search_tools.py",
+          "caller": "fast_search"
+        },
+        {
+          "file": "packages/code-intel/src/attocode_intel/tools/search_tools.py",
+          "caller": "regex_search"
+        },
+        {
+          "file": "packages/code-intel/src/attocode_intel/tools/search_tools.py",
+          "caller": "frecency_search"
+        }
+      ],
+      "known_tests": [
+        "packages/code-intel/tests/test_trigram_workspace.py",
+        "packages/code-intel/tests/unit/code_intel/test_remote_local_fallback.py"
+      ],
+      "parent": "c3850fde3b2c11c94eb7d2e7c7161651ff75d0f4"
+    },
+    {
+      "id": "behavioral-negation",
+      "commit": "39883fefa1f9f2f4b8a9463817a710d50e50ad13",
+      "file": "packages/code-intel/src/attocode_intel/focused_evidence.py",
+      "symbol": "scope_exclusion_terms",
+      "patch_files": [
+        "packages/code-intel/src/attocode_intel/focused_evidence.py",
+        "packages/code-intel/src/attocode_intel/query_ranking.py",
+        "packages/code-intel/src/attocode_intel/_internal/integrations/context/semantic_search.py",
+        "packages/code-intel/tests/test_behavioral_negation.py"
+      ],
+      "known_callers": [
+        {
+          "file": "packages/code-intel/src/attocode_intel/query_ranking.py",
+          "caller": "rerank_broad_candidates"
+        },
+        {
+          "file": "packages/code-intel/src/attocode_intel/query_ranking.py",
+          "caller": "query_diagnostics"
+        },
+        {
+          "file": "packages/code-intel/src/attocode_intel/_internal/integrations/context/semantic_search.py",
+          "caller": "SemanticSearchManager.search_candidates"
+        }
+      ],
+      "known_tests": [
+        "packages/code-intel/tests/test_behavioral_negation.py"
+      ],
+      "parent": "6d407d656da83242e70ff8a37c03ce43893cfa54"
+    },
+    {
+      "id": "overload-source",
+      "commit": "6d407d656da83242e70ff8a37c03ce43893cfa54",
+      "file": "packages/code-intel/src/attocode_intel/search_evidence.py",
+      "symbol": "hydrate_selected_sources",
+      "patch_files": [
+        "packages/code-intel/src/attocode_intel/search_evidence.py",
+        "packages/code-intel/src/attocode_intel/_internal/integrations/context/semantic_search.py",
+        "packages/code-intel/tests/test_chunk_identity.py"
+      ],
+      "known_callers": [
+        {
+          "file": "packages/code-intel/src/attocode_intel/gateway.py",
+          "caller": "OperationGateway._execute_sync"
+        },
+        {
+          "file": "packages/code-intel/tests/test_chunk_identity.py",
+          "caller": "test_java_overloads_fuse_individually_and_reopen_without_losing_metadata"
+        }
+      ],
+      "known_tests": [
+        "packages/code-intel/tests/test_chunk_identity.py"
+      ],
+      "parent": "54f0e80a5e60144acee53b3b0d0f7c0e899a6419"
+    }
+  ]
+}

attocode_code_intel-0.1.8/evals/designs/2026-10-17-change-workflow-replays.md

@@ -0,0 +1,61 @@
+# Real-change workflow replays
+
+Exercise published 0.1.7 against three already known fixes in this repository:
+workspace trigram ownership (#228), behavioral exclusion terms (#225), and
+overload-specific source hydration (#224). This is an operational replay with
+source-audited examples, not a fresh task-completion or agent productivity study.
+
+The adjacent JSON pins each original commit/parent, selected patch files, target
+definition, eight known caller sites and four known relevant test files. Caller
+labels identify enclosing definitions; Python's standard-library AST finds exact
+call lines independently of the intelligence parser. An initial source audit
+corrected the hydration caller from `execute` to `_execute_sync` before any
+product report was collected. Labels are not exhaustive relevance judgments.
+
+For each task, reconstruct the complete actual parent tree, initialize disposable
+Git history, and write only the selected real production/regression files from
+the fixing commit. Other changes in the original PR are excluded. The input
+project is never executed. Contained tracked symlinks are preserved and recorded;
+escapes and other non-regular archive members are rejected. Input archive limits
+are 3,000 entries, 2 MiB per file and 48 MiB total regular-file content.
+
+Measure these in order:
+
+1. Installed CLI `changes`, with a new index directory, then its saved-report check.
+2. Installed CLI `trace`, with disk caches from the first command.
+3. Five trace calls in a new full-profile gateway. The first may load disk caches;
+   the following four describe reuse in that process. Report samples and their
+   median/range, without treating this small sample as a stable tail estimate.
+4. A real full-profile stdio MCP trace, including process startup, using existing
+   disk caches. Grade the delivered structured data and check its portability.
+5. For each independent source-edit, staging and branch-name mutation: an old
+   report must exit 1/stale; refresh must produce a report that exits 0/current;
+   restoring the original state must make the original report current again.
+
+Record changed-target presence, known direct caller-site coverage, known-test
+delivery and ranks, graph sizes/omissions, unresolved references, report bytes,
+latency, peak sampled process-tree RSS and managed scratch. Check scorer rejection
+of missing edges and wrong call-site/definition identities using copies of actual
+reports. A missed label is recorded rather than removed or retuned.
+
+The outer supervisor enforces 600 seconds, 1 GiB managed storage and 2 GiB sampled
+process-tree RSS. Runtime also retains the product's source/output/traversal
+bounds. Only the disposable snapshots/caches are removed; compressed observations
+and a hash-bound summary remain. No acquisition of corpora, models, infrastructure
+or input dependencies is needed. Daily/lean profiles and ranking defaults remain
+unchanged.
+
+The first preparation attempt rejected the repository's contained documentation
+symlink before collecting any product report. Its failure is preserved privately.
+The corrected runner supports contained symlinks; the task labels stayed fixed.
+
+Reproduction uses `eval/quality/change_workflow_trial.py`, an installed 0.1.7 Python
+environment and its matching executable. Run it under `eval.quality.runner` with
+the limits above. Scratch and output paths must be new. The result artifact pins
+the manifest hash; the result summary additionally pins the runner and artifact.
+
+These tasks were selected by the implementer from familiar fixes in one Python
+repository. No before/after product comparison, native timing arm, independent
+reviewer, repair execution or user outcome is measured. Recall on selected call
+sites and test files does not establish complete impact, test sufficiency,
+general relevance, agent speed or benefit in another repository/language.

attocode_code_intel-0.1.8/evals/results/2026-10-17-change-focus-decision.json

@@ -0,0 +1,160 @@
+{
+  "base_commit": "7b09e28946dad07448d4f8a155fa2be8c6defb4d",
+  "baseline": "Published 0.1.7 installed tool; new same-environment sequential replay",
+  "candidate": "Unreleased editable source, class summaries enabled by default",
+  "candidate_engine_sha256": "b4f6cfd5598db38e7e0137ba0cd4dce21f705b43b457aad009087d707ee50ddb",
+  "freeze": {
+    "labels_unchanged": true,
+    "manifest_sha256": "f0829a9ad0dd6bc2a152d747e412bf2456c477f6a7f6b7769b88dbdc0a2ea7f2",
+    "product_files_sha256": {
+      "packages/code-intel/src/attocode_intel/change_investigation.py": "9c18cda0b7adf6cba1c8c3e9e564c01eed2f39737e2b83540400f56876d7a75c",
+      "packages/code-intel/src/attocode_intel/investigation_cli.py": "be7a0aeb20a44634f0c94a5eb90bb70d5c15ac08aa557c26578b76e108a3d8c7",
+      "packages/code-intel/src/attocode_intel/tools/composite_tools.py": "9a8633f94ec633a88e03570f414c662e33262e280259a4051b871127333fc213"
+    },
+    "runner_sha256": "b957650dd1445367f0fcfcc18b07964df7346da42bebe2be2333b0fd647872e8",
+    "same_environment_baseline_then_candidate": true,
+    "final_review_correction": "Class-member summaries also require source containment to distinguish duplicate class names; clarify that only selected member flows are delivered. Known labels unchanged."
+  },
+  "arms": {
+    "baseline": {
+      "artifact": "2026-10-17-change-focus-baseline.json.gz",
+      "gzip_sha256": "f432dfb8477a3fb5eba28c891f6dd8eb74a5b30998a916c52d46ef9d56035bd5",
+      "raw_sha256": "507adf3664c74ec46387449276fad37e0e457411c1173a0e62da515236d527e8",
+      "raw_bytes": 1160967,
+      "gzip_bytes": 70763,
+      "supervision": {
+        "seconds": 80.079,
+        "wall_seconds": 80.08,
+        "peak_disk_bytes": 197632054,
+        "peak_rss_bytes": 430686208,
+        "cleanup_verified": true
+      }
+    },
+    "candidate": {
+      "artifact": "2026-10-17-change-focus-candidate.json.gz",
+      "gzip_sha256": "609cbb268242e5a64fc600c775767d30181b9642a790227fa625cc193ac1b8f5",
+      "raw_sha256": "39902ed3c4d7f7e8bca750aaa99f44dd7fc4ace7b2e2afdaea62da721894e979",
+      "raw_bytes": 953540,
+      "gzip_bytes": 60508,
+      "supervision": {
+        "seconds": 74.804,
+        "wall_seconds": 74.804,
+        "peak_disk_bytes": 206155830,
+        "peak_rss_bytes": 459620352,
+        "cleanup_verified": true
+      }
+    }
+  },
+  "tasks": [
+    {
+      "id": "workspace-cache",
+      "baseline_report_bytes": 101679,
+      "candidate_report_bytes": 102039,
+      "bytes_reduction_fraction": -0.003540554096716164,
+      "known_callers_retained": 3,
+      "nonclass_flows_compared": 11,
+      "nonclass_resolved_edges_identical": true,
+      "known_tests_retained": [
+        {
+          "file": "packages/code-intel/tests/test_trigram_workspace.py",
+          "found": true,
+          "rank": 5
+        },
+        {
+          "file": "packages/code-intel/tests/unit/code_intel/test_remote_local_fallback.py",
+          "found": true,
+          "rank": 3
+        }
+      ],
+      "candidate_class_summaries": 0,
+      "timings": {
+        "baseline": {
+          "new_index_changes_seconds": 6.283472000039183,
+          "cached_cli_trace_seconds": 1.1024351250380278,
+          "in_process_trace_median_seconds": 0.2697261249413714,
+          "stdio_trace_including_startup_seconds": 1.0779324999311939
+        },
+        "candidate": {
+          "new_index_changes_seconds": 4.20542483299505,
+          "cached_cli_trace_seconds": 3.2976214999798685,
+          "in_process_trace_median_seconds": 0.541220374929253,
+          "stdio_trace_including_startup_seconds": 1.0819442920619622
+        }
+      }
+    },
+    {
+      "id": "behavioral-negation",
+      "baseline_report_bytes": 257931,
+      "candidate_report_bytes": 174205,
+      "bytes_reduction_fraction": 0.32460619312917016,
+      "known_callers_retained": 3,
+      "nonclass_flows_compared": 14,
+      "nonclass_resolved_edges_identical": true,
+      "known_tests_retained": [
+        {
+          "file": "packages/code-intel/tests/test_behavioral_negation.py",
+          "found": true,
+          "rank": 3
+        }
+      ],
+      "candidate_class_summaries": 1,
+      "timings": {
+        "baseline": {
+          "new_index_changes_seconds": 4.668141957954504,
+          "cached_cli_trace_seconds": 3.1727850410388783,
+          "in_process_trace_median_seconds": 0.46924618748016655,
+          "stdio_trace_including_startup_seconds": 0.9772187080234289
+        },
+        "candidate": {
+          "new_index_changes_seconds": 4.666719374945387,
+          "cached_cli_trace_seconds": 3.319811374996789,
+          "in_process_trace_median_seconds": 0.5898817084962502,
+          "stdio_trace_including_startup_seconds": 1.102902540937066
+        }
+      }
+    },
+    {
+      "id": "overload-source",
+      "baseline_report_bytes": 406914,
+      "candidate_report_bytes": 212152,
+      "bytes_reduction_fraction": 0.4786318484987,
+      "known_callers_retained": 2,
+      "nonclass_flows_compared": 20,
+      "nonclass_resolved_edges_identical": true,
+      "known_tests_retained": [
+        {
+          "file": "packages/code-intel/tests/test_chunk_identity.py",
+          "found": true,
+          "rank": 3
+        }
+      ],
+      "candidate_class_summaries": 3,
+      "timings": {
+        "baseline": {
+          "new_index_changes_seconds": 5.089328333036974,
+          "cached_cli_trace_seconds": 3.240020291064866,
+          "in_process_trace_median_seconds": 0.4533247079816647,
+          "stdio_trace_including_startup_seconds": 3.1255698750028387
+        },
+        "candidate": {
+          "new_index_changes_seconds": 4.793609583983198,
+          "cached_cli_trace_seconds": 3.313361292006448,
+          "in_process_trace_median_seconds": 0.5379430835018866,
+          "stdio_trace_including_startup_seconds": 1.062626083032228
+        }
+      }
+    }
+  ],
+  "limitations": [
+    "Previously consumed selected caller/test labels from familiar Python fixes, not fresh held-out ranking or agent outcomes.",
+    "Three known tests are themselves changed patch inputs; only one preexisting test is a discovery example.",
+    "Sequential local arms have no isolation of OS caches, temperature or timing noise; no general speedup claim.",
+    "Unresolved diagnostic counts/ranks vary; fewer diagnostics are not established precision gains."
+  ],
+  "candidate_patch": {
+    "artifact": "2026-10-17-change-focus-candidate.patch.gz",
+    "gzip_sha256": "3bbb85f459e70c573ba3d9638a5d5cd7f0ae409f567b199f8784571a25d65f3c",
+    "raw_sha256": "249e0a7ac095744078107a660d1112eb98b2fded3f17d172c29df4131a75e743"
+  },
+  "gate_passed": true
+}

attocode_code_intel-0.1.8/evals/results/2026-10-17-change-workflow-replays.md

@@ -0,0 +1,109 @@
+# Published 0.1.7 on three real changes
+
+The installed release found all **eight preselected caller sites**, all **three
+changed target definitions**, and all **four known relevant test files** in these
+historical replays. Test-file suggestion ranks were 3, 5, 5 and 3. All nine
+saved-report mutation/refresh sequences passed. Seven caller labels are production
+sites; one is a newly added regression-test call. Three of the four test files
+are themselves changed patch inputs. Only the unchanged remote-fallback test
+(rank 3) is a preexisting-test discovery example. This establishes evidence on
+these examples; it does not measure broad test discovery, repair success or an
+agent benefit.
+
+## Inputs and measurements
+
+The [design](../designs/2026-10-17-change-workflow-replays.md) and
+[frozen task manifest](../designs/2026-10-17-change-workflow-replays.json) describe
+the selected production/regression files replayed on each complete actual parent
+tree. These are familiar fixes from this Python repository, not fresh unseen
+tasks. Input project code/tests were never executed. No models, cloud workers,
+new corpora or input dependency installs were used.
+
+| Real fix | Known caller sites | Known test files | CLI changes, new index | CLI trace, disk cache | In-process trace median | MCP trace with startup |
+|---|---:|---:|---:|---:|---:|---:|
+| Workspace cache isolation, #228 | 3/3 | 2/2 | 6.553 s | 1.076 s | 0.264 s | 1.020 s |
+| Behavioral negation, #225 | 3/3 | 1/1 | 4.636 s | 3.162 s | 0.612 s | 0.982 s |
+| Overload source recovery, #224 | 2/2 | 1/1 | 5.168 s | 3.204 s | 0.524 s | 1.080 s |
+
+The first CLI command has a new project index directory; this is not an OS
+cache-cold benchmark. The subsequent CLI trace reuses disk caches. Each new
+gateway made five sequential traces; medians above use the last four, excluding
+the initial cache load. Warm ranges were 0.241–0.301, 0.260–0.654 and 0.300–0.922
+seconds respectively. These small samples do not establish stable tail latency
+or a performance improvement. The stdio measurement includes startup, protocol
+initialization and tool discovery, with previously populated disk caches; it
+ends when the result arrives, before connection shutdown.
+
+The combined change reports, all five gateway traces per task and all three
+actual full-profile stdio MCP traces retained the labeled caller sites. The delivered MCP structured reports
+passed the CLI-compatible integrity/freshness check. This trial leaves the host's
+daily profile unchanged; the new operations were exercised in disposable full
+sessions.
+
+For each task, separately editing a caller file, staging the patch, or changing
+only the branch name made the saved report stale (exit 1). Refresh generated a
+current report (exit 0); restoring the initial state made the original current
+again. That is nine independent sequences, including actual CLI exits.
+
+Twelve post-collection scorer controls rejected empty edges, wrong call-site
+lines, wrong callee files and wrong caller identities in copies of actual
+reports. Labels were not changed after seeing results. This tests selected-site
+scoring, not the precision of every returned edge.
+
+## The practical limit
+
+The combined change reports were **101,671**, **257,930** and **406,951 bytes**.
+No changed files or definitions were omitted, but **1, 4 and 8 individual caller
+flows** hit their caps. Broad changed-class flows in particular produced many
+unresolved candidates: omitted-unresolved counts summed across changed flows
+were 0, 611 and 1,175. These sums may count the same candidate in several flows;
+they are not unique missing references.
+
+Standalone traces of the selected definitions used the larger 32-node bound and
+did not omit nodes or edges in these examples: graphs had 26/26, 18/38 and 5/6
+nodes/edges, with 24, 0 and 0 unresolved rows. Missing edges still do not establish
+absence. Two CLI snapshots reported phase `hydrating` despite parsed-file counts
+equaling their totals; neither is described here as proof of complete indexing.
+
+For a broad patch, use the combined report to select a definition, then use a
+focused trace. A useful next implementation would make individual-symbol focus
+and changed-class summaries easier to request and read. This trial does not
+change traversal defaults, class behavior or ranking.
+
+## Cost, storage and evidence
+
+The successful outer supervised run took **70.712 seconds**, with peak sampled
+process-tree RSS **448.938 MiB** and peak managed storage **197.363 MiB**, below
+the fixed 600-second / 2-GiB-RSS / 1-GiB-storage limits. Disposable snapshots and
+caches were removed; the managed root retains only its marker/empty directories.
+Shared studies, caches, model storage and Docker were untouched.
+
+The first attempt stopped during preparation on the repository's contained
+documentation symlink, before collecting any product report. Its failure is
+preserved privately. Preparation was corrected to preserve bounded contained
+symlinks, with unchanged task labels. This was a harness correction, not a product
+failure. Scoped Ruff and whitespace checks passed.
+
+The [machine-readable summary](2026-10-17-change-workflow-summary.json) pins the
+manifest, exact runner and compressed/raw observation hashes. The
+[compressed observations](2026-10-17-change-workflow-observations.json.gz) contain
+complete reports, source hashes, all timed samples, labels and freshness results
+(70,711 compressed bytes; 1,159,956 raw bytes). Private supervision logs and the
+initial preparation failure remain under the ignored local quality directory.
+
+Reproduce under an installed 0.1.7 environment, using new scratch/output paths,
+with the outer limits from the design:
+
+```sh
+/path/to/0.1.7/bin/python eval/quality/change_workflow_trial.py \
+  --repo "$PWD" \
+  --manifest packages/code-intel/evals/designs/2026-10-17-change-workflow-replays.json \
+  --scratch /private/tmp/new-workflow-replay \
+  --output /private/tmp/new-workflow-observations.json.gz \
+  --cli /path/to/0.1.7/bin/attocode-code-intel
+```
+
+There is no native timing arm, independent reviewer, general relevance judgment,
+cross-language sample or executed repair in this trial. Known-test delivery is
+not proof of test sufficiency. No product source, default, version or live MCP
+configuration changed.