[pull] master from GaijinEntertainment:master#976
Merged
Conversation
…egressions Closes a silent stale-state bug in the MCP server: every tool that compiles .das via compile_file (compile_check / aot / lint / describe_type / find_symbol / list_functions / list_module_api / list_types) was retaining the C++-side [function_macro] / [call_macro] registrations from its FIRST compile call, so module edits to macro-heavy code (decs / sql / dasImgui boost) were invisible to subsequent MCP calls until the server was restarted. The fix: each of those 8 tools is now a thin do_<tool>() wrapper that shells out via run_mcp_subtool() (utils/mcp/tools/common.das) to a fresh daslang.exe running utils/mcp/subtools/<tool>.das. Macro state is process- local, so a fresh subprocess starts with a clean slate every time. The dispatch contract (do_<tool>(...args)) is unchanged, so protocol.das needs no edit and the 270 existing tests in test_tools.das exercise the wiring end-to-end. Wins: - Stale-macro bug fixed for all 8 tools. - 4 new tests for the run_mcp_subtool helper itself: passthrough, missing subtool, timeout, panic-with-stdout-capture. - 3 fixture subtools (_test_passthrough / _test_panic / _test_sleep) drive the helper tests. cpp_dirs narrowing ------------------ Added a cpp_dirs parameter to do_find_symbol and do_goto_definition: an optional comma-separated list of repo-relative paths that scopes the with_cpp_source redirect lookup. Empty -> use the cached global C++ index over CPP_SEARCH_DIRS (today's behavior). Non-empty -> fresh scoped scan via the new cpp_lookup_by_name_scoped() helper in cpp_common.das, no global cache touched. The slow tests (test_find_symbol_with_cpp_source, test_goto_definition_with _cpp_source, test_cpp_find_symbol) now pass cpp_dirs="src/builtin" or directory="src/ast" so the scan walks ~33 files instead of the full ~947. test_tools.das suite drops from 132s to 89s (33% faster) at 279/279 green. daslib/json_boost.das STYLE018 fix ---------------------------------- Generic JV(struct) instantiation on a bool field generated `if (field == default<typedecl(field)>)` which lowers to `field == false` and trips the new STYLE018 lint. Split the workhorse path with a static_if on typeinfo stripped_typename(field) == "bool" so bool uses `if (!field)` and numeric workhorses keep the generic-zero compare. macro_call regressions ---------------------- - test_call_macro_cross_expr_emit.das: gap #3 from the dasImgui plan (cross-expression visibility of an auto-emitted module global). - test_widget_dispatch.das + _widget_dispatch_helper.das: end-to-end smoke for the dasImgui-style [widget] call_macro pattern (dotted-flag parsing, named-tuple argument workaround, ExprField chain walking, cross-module Type.tStructure construction). Lint sweep ---------- All 27 changed/new .das files pass mcp__daslang__lint cleanly (53 -> 0 warnings). Pre-existing PERF/STYLE warnings in touched files (cpp_common, goto_definition, json_boost) cleared too, per the imminent PR lint gate. Verification ------------ - 8001/8001 dastest tests pass. - 7395/7395 AOT tests pass (after cmake reconfigure to refresh the new macro_call AOT glob). - 279/279 mcp tests pass (270 existing + 4 new run_mcp_subtool + 5 new subtests inside those 4). - 246/246 json round-trip tests pass (sanity for the json_boost edit). mouse-data ---------- Cached the @capture(& var) lambda capture-by-reference syntax landmines (parens, move-into-storage, unsafe wrapping) discovered while building the [widget] runtime closures. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ess-tools mcp: per-tool subprocess refactor + cpp_dirs narrowing + macro_call regressions
Three issues in utils/mouse/:
- No automatic re-indexing — `git pull` of new docs left the SQLite
index stale until manual `mouse rebuild`.
- "Too similar" was a near-universal false positive — `dupe_check`
blocked any FTS5 hit, and the OR-joined query matched nearly any
natural-language question.
- Tool output rendered BM25 ranks raw (small negative floats, often
scientific notation) with vague headers.
Fixes
- Auto-reindex: persisted staleness signature in a new `index_meta`
table (migration v3). The signature compute lives in a new shared
module `utils/common/git_signature.das` (per-search-dir
`git rev-parse HEAD:<dir>` tree hash + filtered `git status
--porcelain` + per-changed-file mtimes, hashed; sorted recursive
filesystem walk on no-git fallback). Both utils/mouse and
utils/mcp/tools/cpp_common use the shared module — ~150 LOC of
duplicated git plumbing collapsed into one place.
- Per-tree HEAD: a `git pull` (or branch switch) that doesn't change
the indexed subtree no longer invalidates the cache. Drops a
whole class of spurious rebuilds on monorepo workflows.
- Sorted fs fallback: `_findfirst`/`readdir` ordering varies across
platforms; collected paths are now sorted before hashing so the
signature is deterministic.
- Jaccard scoring: dupe_check now sorts by Jaccard title similarity
(with stop-words filtered); add_doc hard-blocks only at >= 0.7,
surfaces the ranked list otherwise. Caller decides below threshold.
- Direct response messaging with fixed-decimal floats ({x:.3f}); ask
shows "Top N best matches" with both BM25 and sim columns; add
responses are mode-aware (block / created+similar / created-clean).
- New --raw-query / rawQuery flag bypasses sanitizer for FTS5
power-user syntax.
daslib promotions
- daslib/strings_boost: new `jaccard(table, table)` and
`jaccard(array, array)` next to `levenshtein_distance`.
- daslib/fio: `run_and_capture` promoted from utils/mcp/tools/common.das
(one canonical home; common.das now relies on the daslib re-export).
Tests: utils/common/tests/test_git_signature (17), utils/mouse/tests
(34), utils/mcp/test_tools (275), tests/strings/strings_jaccard (11);
all PASS in interpreter mode. Sphinx clean; das2rst regenerated with
new groups.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…reindex mouse: git-aware auto-reindex + Jaccard dupe gate + --raw-query
…_signature helper Commit 5839228 (mouse: git-aware auto-reindex) added two calls to `compute_index_signature(root)` in main.das — `cmd_rebuild` and the `tool_rebuild` MCP entrypoint. The actual helper exported from `utils/common/git_signature.das` is `compute_signature(root, predicate, search_dirs_abs)` (3 args). Both call sites failed to compile, breaking `mouse rebuild` from the CLI and `mouse__rebuild` over MCP. The mouse test suite didn't cover those entry points, so the regression slipped through to master. Rather than re-spell the 3-arg call at every signature-stamp site, factor the canonical lambda + dirs into `compute_docs_signature(root)` in index.das (same shape that `ensure_index_fresh` already computes for the read side). Use it in: - main.das `cmd_rebuild` (was broken) - main.das `tool_rebuild` (was broken) - index.das `ensure_index_fresh` (was inline) - index.das `add` flow at line 533 (was inline) Single source of truth for "what's the staleness signature shape" — keeps the read in `ensure_index_fresh` and every write in lockstep. Verification: - `mouse rebuild` from CLI: 21/21 docs reindexed clean - `utils/mouse/tests`: 61/61 pass - lint clean on touched files (the two PERF warnings come from inside daslib/linq.das + sqlite_linq.das macro-expansion contexts, not the mouse files; latent on master) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )