Qualified calls and api polish - #45
Merged
Merged
Conversation
…lish
This branch bundles two related streams of work for tscodesearch:
QUALIFIED-CALL INDEXING & VAR-TYPE RESOLVER
- New ``qualified_calls`` Tantivy field stores ``Type.Method`` tokens for
call sites: static-style receivers (``Foo.Save()`` → ``Foo.Save``) and
resolved-receiver forms (``store.Save()`` where ``store: IRepo`` →
``IRepo.Save``). The receiver type is pinned by a new method-scoped
var-type map (``_build_var_type_map``) with conflict suppression — names
that map to multiple types in one scope emit no qualified form.
- Block-level scoping: ``block`` / ``catch_clause`` / ``for_statement`` /
``foreach_statement`` / ``using_statement`` are scope nodes, so sibling
blocks (if/else, multiple catch arms, two for-loops) isolate their
declarations. Fixes prior false negatives where shadowed names
collapsed to ``None`` under method-wide flat scoping.
- Best-guess inference heuristics in ``_infer_var_type``: await unwrap,
generic-method type-argument (``c.Resolve<IService>()`` → ``IService``),
static-factory pattern (``Widget.Create()`` → ``Widget``), ternary
branch try-each, member-access (instance ``obj.Prop`` → ``Prop`` /
static ``Type.Member`` → ``Type``), and foreach-over-collection
element-type extraction across scope boundaries (handles
``foreach (var x in someField)`` where ``someField`` is in file scope).
- ``calls`` mode now searches ``call_sites,qualified_calls,path_tokens``
so bare ``Save`` and qualified ``IRepo.Save`` both hit. The AST
post-filter in ``q_calls`` is upgraded to match by resolved receiver
too — ``q_calls("IRepo.Save")`` finds ``store.Save()`` when ``store``
resolves to ``IRepo``.
- New ``var_type NAME`` listing mode: for each identifier occurrence,
report the resolved type (or ``(unresolved)`` / ``(conflicting)``).
Saves an ``at LINE:COL`` round-trip when you just want the type.
- ``q_at`` fix: ``field_declaration`` and ``event_field_declaration``
carry their name on a nested ``variable_declarator`` (no direct
``name`` field), so the position scope chain was silently dropping
the field. Now extracted via ``_field_declarator_name``.
- ``q_calls`` anchors chained calls at the method-name token. The
reported line and source text point at where the matched identifier
actually appears, not the start of the outer multi-line invocation.
- Windows file-handle flake fix: ``Backend.upsert_many`` retries the
whole add+commit cycle on transient Windows IO errors (Defender /
mmap-release races on a freshly-created collection's first commit).
``tests.helpers._delete_collection`` adds a parent-directory canary
on Windows so tearDownClass doesn't return until the next setUpClass
can write into ``.tantivy/``.
API POLISH (Phases 0-5 from the improvement plan)
- Phase 0 (docs/errors): ``uses_kind`` default-union now documented;
``accesses_of`` docstring cross-links to ``all_refs`` for implicit-this
reads; ``query_codebase`` redirect message expanded to cover file-only
modes (``body``/``at``/``params``/``var_type``) — not just listing
modes; CLAUDE.md notes the required ``mcp__tscodesearch__*`` prefix.
- Phase 1 (output trimming): replaced the box-drawing banner
(``── [kind] name (lines X–Y) ──``) on ``body`` / ``declarations``
output with a compact ASCII header (``[kind] name S-E:``) across all
five language modules. Saves ~30 chars per match and avoids
``cp1252`` codec errors on Windows consoles.
- Phase 4a: tier-2/3 follow-up suggestions and the docstring examples
drop the ``$SRC_ROOT/`` placeholder — the tool accepts the bare
relative path that ``query_codebase`` itself returns.
- Phase 5 (visibility filter): captures explicit C# access modifiers
plus language defaults (top-level types → internal, nested types →
private, interface members → public, enum body members → public,
class/struct/record members → private). Compound modifiers collapse to
their dominant role (``protected internal`` → ``protected``,
``private protected`` → ``private``). Exposed as a comma-separated
``visibility`` argument on ``query_codebase`` and ``query_single_file``
for the four declaration modes (declarations / classes / methods /
fields). Two parallel Tantivy fields (``type_visibilities``,
``member_visibilities``) so a "files with at least one public member"
query doesn't catch files whose only public thing is a top-level type.
- Diagnostic: new ``scripts/probe_resolved.py`` categorises lowercase-
receiver gap candidates so future tuning targets the biggest patterns.
TESTS: +173 new cases (1099 total, all green; 10× full-suite stability
sweep clean). Coverage spans var-type resolution, block scoping, inference
heuristics, qualified-receiver matching, ``var_type`` mode, ``q_at`` field
scope, output format, redirect routing, Windows IO retry, and visibility
extraction + filtering.
DOCS: CLAUDE.md updated with the new ``qualified_calls`` schema field,
revised ``calls`` semantics, ``var_type`` mode entry, ``visibility``
filter section, and the MCP-prefix note.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Strip non-ASCII codepoints (em dashes, arrows, box drawing, ellipsis,
bullets, math symbols, check marks) from every .py / .md / .mjs / .js
file under the repo and replace each with its ASCII analogue:
EM DASH (U+2014) / EN DASH (U+2013) -> '--' or '-'
RIGHTWARDS / LEFTWARDS ARROW -> '->' / '<-'
RIGHTWARDS DOUBLE ARROW (U+21D2) -> '=>'
HORIZONTAL ELLIPSIS (U+2026) -> '...'
BOX DRAWINGS LIGHT / DOUBLE HORIZONTAL -> '-' / '='
BOX DRAWINGS LIGHT VERTICAL -> '|'
BULLET (U+2022) -> '*'
GREATER-/LESS-THAN OR EQUAL TO -> '>=' / '<='
ELEMENT OF (U+2208) -> ' in '
MULTIPLICATION SIGN (U+00D7) -> 'x'
CHECK MARK / BALLOT X -> 'OK' / 'NO'
Touches 80 files and ~17k character positions. Mostly cosmetic
section-separator comments and prose dashes, but also includes the
``q_accesses_on`` output marker (``<-``) and a handful of f-strings
that the daemon and tools emit.
Adds CLAUDE.md "ASCII only" rule with the Unicode-name-keyed reference
table (so the rule's own text doesn't violate the rule). Adds two
helper scripts:
scripts/find_nonascii.py audits the tree and reports every
non-ASCII codepoint by file region
(docstring / string literal / comment /
code), separating hot-path modules
(query / indexserver / mcp_server /
tsquery_server) from cold.
scripts/replace_nonascii.py the mass-rewriter used for this commit;
dry-run by default, --apply to write.
All 1099 tests still pass; no test assertion depended on the old
Unicode forms.
Motivation: Windows console encoding (cp1252) and many downstream tools
misrender or fail on the previously-used codepoints, and the
``[method] name S-E:`` tool output had been triggering ``cp1252 codec
can't encode`` errors when piped to Python sub-processes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
| by_cat: dict[str, list] = {} | ||
| in_docstring = False | ||
| tag = None | ||
| for ln, line in enumerate(open(path, encoding="utf-8").read().splitlines(), 1): |
|
|
||
| for path in _walk_targets(): | ||
| try: | ||
| text = open(path, encoding="utf-8").read() |
| count = line.count(q) | ||
| if count % 2 == 1: | ||
| in_docstring = not in_docstring | ||
| tag = q if in_docstring else None |
| import os | ||
| import re | ||
| import sys | ||
| import unicodedata |
Comment on lines
+21
to
+24
| from query.cs import ( | ||
| _build_var_type_map, _find_all, _text, _CsIndex, _DESCRIBE_NODE_TYPES, | ||
| _q_all_call_site_infos, | ||
| ) |
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
No description provided.