[pull] master from GaijinEntertainment:master

daslib/perf_lint.das

@@ -10,7 +10,7 @@ module perf_lint shared private
 //!
 //! Detects common performance anti-patterns in daslang code at compile time.
 //! When this module is required, a lint pass runs after compilation and reports
-//! warnings as ``CompilationError::performance_lint`` (error code 40217).
+//! warnings as ``CompilationError::runtime_macro_performance`` (error code 31208).
 //!
 //! Rules:
 //!   PERF001 — string += in loop (O(n²))

daslib/style_lint.das

@@ -10,7 +10,7 @@ module style_lint shared private
 //!
 //! Detects non-idiomatic patterns in daslang code at compile time.
 //! When this module is required, a lint pass runs after compilation and reports
-//! warnings as ``CompilationError::style_lint`` (error code 40218).
+//! warnings as ``CompilationError::runtime_macro_style`` (error code 31209).
 //!
 //! Rules:
 //!   STYLE001 — unnecessary <| pipe before block argument

doc/internal/error_audit_design.md

@@ -0,0 +1,190 @@
+# daslang error-code refactor — audit synthesis & design
+
+This document is the design output of Phase B of the error-reporting refactor.
+It is **a working document**, not user-facing documentation. Source of truth
+during implementation; can be moved to a permanent design-docs location once
+the work lands, or removed if not useful.
+
+## Source artifacts
+
+All produced by the audit pipeline at `D:/daslang-error-audit/`:
+
+| Artifact | Role |
+|---|---|
+| `D:/daslang-error-audit/taxonomy.md` | The shared facet/subject/modifier vocabulary used by audit agents |
+| `D:/daslang-error-audit/synth_design.py` | Reads `error_audit.csv`, produces the design artifacts below |
+| `D:/daslang-error-audit/design/new_enum_draft.h` | Drafted replacement for `compilation_errors.h` (561 entries) |
+| `D:/daslang-error-audit/design/uniques.tsv` | Each new enum name + its site count + the old cerrs that mapped into it |
+| `D:/daslang-error-audit/design/legacy_map.tsv` | Per-site: file, line, old cerr, new name, new num — input to retag/sweep |
+| `D:/daslang-error-audit/design/verbose_gate.tsv` | 89 sites that should be wrapped in `if (verbose)` |
+| `D:/daslang-error-audit/design/recategorize.tsv` | 62 sites whose cerr semantics change (not just rename) |
+| `D:/daslang-error-audit/design/wording.tsv` | Wording fixes (typos, missing spaces, vague phrasing) |
+| `doc/internal/error_audit.csv` | The consolidated audit (1116 rows, 1 per error site) — checked in |
+
+## What was audited
+
+Every `error(...)` call site in:
+- `src/ast/*.cpp` (semantic, infer, lint, validate, simulate, derive_alias, export, const_folding, annotations, typedecl)
+- `src/parser/*.ypp`, `*.lpp`, `parser_impl.cpp` (lexer/parser sources only — generated `.cpp` skipped)
+- `src/builtin/*.cpp`, `src/simulate/*.cpp`
+
+**Total: 1116 sites across 32 source files.**
+
+Six parallel Explore-style agents performed the inventory; each wrote a CSV
+slice; slices consolidated into `doc/internal/error_audit.csv` with columns:
+`file, line, current_cerr, message_template, context_function, classification,
+tag_composition, proposed_enum_name, wording_change`.
+
+## Decisions locked with Boris
+
+| Topic | Decision |
+|---|---|
+| Storage | Flat `enum class CompilationError` — unchanged shape |
+| Naming | Tag-composition discipline: `<facet>_<subject>[_<modifier>]` |
+| Migration scope | Bundle everything in one PR — retag + wording + recategorize + verbose-plumbing + dedup helper + 111-test sweep + fix the 5 pre-existing under-specified `expect` lines |
+| Renaming policy | Rename `*_not_found` → `lookup_*` for prefix consistency. Other agent-proposed names kept as-is (after mild collapse of trailing modifier suffixes). |
+| Granularity | Mild collapse: 1116 sites → 561 unique codes (was 632 before collapse). Modifier suffixes like `_void`, `_ref`, `_const`, `_value` are dropped from cerr names; the info stays in the message text. |
+| Follow-on / internal granularity | Fine-grained per subject. `not_resolved_yet_<subject>` for each subject (38 codes); `internal_<subject>_<facet>` for each (42 codes). |
+| Numbering | Loose ranges: 1xxxx lex / 2xxxx parse / 3xxxx semantic / 4xxxx logic / 5xxxx integration. Inside a stage, cluster by leading facet (within-stage offset = `(facet_idx % 90 + 1) * 100`). New codes go next-available within the cluster. |
+| Verbose flag | Reuse existing `InferTypes::verbose` ([include/daScript/ast/ast_infer_type.h:30](../../include/daScript/ast/ast_infer_type.h#L30)) — no new user-facing flag. Plumb through 89 emit sites identified by audit. |
+| Taxonomy expansion | Accept 11 new tags surfaced by the audit: 8 subjects (`string, comment, indentation, runtime, iterator, pointer, buffer, context`) + 3 facets (`runtime, performance_lint, style_lint`). Reject narrow parser-only tags (`aka, pipe, keyword, dispatcher`). |
+
+## The naming discipline
+
+```
+<facet>_<subject>[_<subject2>][_<modifier>]
+```
+
+| Slot | Source | Examples |
+|---|---|---|
+| facet | what kind of fault | `invalid`, `lookup`, `cant`, `missing`, `mismatching`, `exceeds`, `ambiguous`, `already_declared`, `not_resolved_yet`, `internal`, `unsafe`, `recursion`, `runtime` |
+| subject | what construct | `type`, `function`, `argument`, `result`, `structure`, `field`, `variable`, `array`, `table`, `tuple`, `variant`, `enumeration`, `enumerator`, `expression`, `module`, `annotation`, ... (full list in `taxonomy.md`) |
+| modifier | rare; only when needed | (mostly stripped during normalization — they live in message text, not the cerr) |
+
+Composition order: facet first, then subject(s) general → specific. So
+`function | argument | type | invalid` becomes `invalid_function_argument_type`.
+Multi-word facets (`not_resolved_yet`, `already_declared`) are matched as a
+single facet token.
+
+**Future contributors:** when adding a new error site, derive the cerr from the
+tag composition. If the composed name doesn't yet exist in
+`compilation_errors.h`, add it.
+
+## The new enum — facet sizes
+
+| Stage | Facet | Count | Notes |
+|---|---|---|---|
+| 1xxxx (lex) | invalid | 2 | line, string |
+| | mismatching | 3 | curly_bracers, parens, module_name |
+| | exceeds | 2 | constant ranges, file boundary |
+| 2xxxx (parse) | invalid | ~30 | grammar-rule rejections |
+| | missing | 1 | module_name |
+| | exceeds | 1 | bitfield |
+| | already_declared | ~10 | parser-detected name collisions |
+| 3xxxx (semantic) | invalid | 213 | the bulk — type/argument/field/etc faults |
+| | cant | 62 | `cant_<verb>` family — operations forbidden in context |
+| | lookup | 46 | `*_not_found` renamed to `lookup_*` |
+| | unsafe | 37 | unsafe-required operations |
+| | not_resolved_yet | 37 | follow-on cascade family (verbose-gated) |
+| | missing | 22 | required-but-absent (return paths, init, etc.) |
+| | already_declared | 12 | name collisions |
+| | exceeds | 16 | counts/sizes/ranges |
+| | ambiguous | 17 | multi-candidate resolution |
+| | mismatching | 15 | two things expected to match don't |
+| | recursion | 7 | unbounded/forbidden recursion |
+| 4xxxx (logic) | (lint) | small | performance_lint, style_lint families |
+| 5xxxx (integration) | internal | 42 | validate-AST stubs, DAS_VERIFY, "should not happen" sites |
+| | runtime | 14 | DataWalker/hash/serializer compile-time-eval errors |
+
+**Total: 561 distinct codes** + 1 `unspecified = 0`.
+
+Full enum draft: `D:/daslang-error-audit/design/new_enum_draft.h`. To copy into
+`include/daScript/ast/compilation_errors.h` in Phase C.
+
+## Transition by phase
+
+### Phase C — Apply enum + emit-site retags + wording fixes
+
+1. Replace `include/daScript/ast/compilation_errors.h` body with `new_enum_draft.h` content (preserve `#pragma once` + namespace + `unspecified = 0` line).
+2. For each site in `legacy_map.tsv`, edit the `error(...)` call to pass the new cerr value. The TSV is per-site `(file, line, new_name)` — straight mechanical retag.
+3. For each site in `wording.tsv` (~22 sites), apply the wording fix inline. Examples:
+   - `subexpresion → subexpression` (×2: ast_infer_type_helper.cpp:966, 969)
+   - `strcut → struct` (ast_infer_type_function.cpp:1162)
+   - `cant't → can't` (ast_infer_type.cpp:4126)
+   - `intializer → initializer` (parser_impl.cpp:791)
+   - `simluateNullCoalescing → simulateNullCoalescing` (ast_simulate.cpp:1352)
+   - missing space at ast_infer_type.cpp:312 (`<name>can't` → `<name> can't`)
+   - duplicate emit at ast_infer_type.cpp:1617/1620 — drop the second
+   - inverted logic at ast_simulate.cpp:3874 (`stt.size()==1` triggers "unknown structure" — should be `>1`)
+   - 7 sites in `ds_lexer.lpp` saying "curly braces"/"square braces" but tagged `mismatching_parens` — retag accordingly
+4. Update consumer files (only matters if they reference enum names by-symbol; none currently observed but check during edit):
+   - `utils/daScript/main.cpp`
+   - `utils/daslang-live/main.cpp`
+   - `utils/dasFormatter/`
+   - `include/daScript/simulate/runtime_string.h` (the `reportError` formatter — unchanged behavior)
+5. `include/daScript/ast/ast_infer_type.h:5` declares CompilationError values used in inference; verify still resolve.
+6. Regenerate parsers (CMake handles `.ypp`/`.lpp` automatically when sources touched).
+
+Worker plan: split Phase C across 4-6 subagents by file group, mirroring the inventory-agent split. Each agent reads `legacy_map.tsv`, filters by its files, applies edits via Edit tool. **No hand-tuning** — the TSV is the source of truth.
+
+### Phase D — Plumb `verbose` through 89 emit sites
+
+`verbose_gate.tsv` lists every site classified `mid_inference` or `follow_on`.
+89 rows. For each:
+
+```cpp
+if (verbose) {
+    error(...);
+}
+```
+
+`verbose` is `false` on intermediate infer passes (no work, no emit) and
+`true` on the final failure-explanation pass via `reportingInferErrors`. Match
+the style of the ~25 existing `if (verbose)` sites in `ast_infer_type.cpp`
+(e.g. lines 1011, 3712, 4135).
+
+### Phase E — `Program::deduplicateErrors()`
+
+See plan file `C:/Users/Boris/.claude/plans/look-at-wip-das-and-synchronous-quill.md` "Phase E" — implementation per the four-rule algorithm. The `not_resolved_yet_*` family is identified by the facet prefix; Rule 4 drops all of them when ANY non-not_resolved_yet, non-unspecified code exists.
+
+### Phase F.1 — Test sweep
+
+`D:/daslang-error-audit/regen_expect.py` already validated baseline (106/111 match, 5 pre-existing under-specified `expect` lines logged at `D:/daslang-error-audit/expect_report.tsv`).
+
+After Phase C-E land:
+1. Re-run `regen_expect.py` — every test will now mismatch.
+2. Run with `--write` to apply the new emitted codes in-place.
+3. Manual review of sample (~20 files) confirms correctness.
+4. Fix the 5 pre-existing under-specifications (`cant_get_field.das`, `failed_aliasing.das`, `failed_call_depth.das`, `failed_condition_must_be_bool.das`, `failed_const_ref.das`) — full multiset, not subset.
+
+### Phase F.2 — Final verify
+
+1. `examples/wip.das` — exactly one error.
+2. Full test suite green: `daslang.exe dastest/dastest.das -- --test tests`.
+3. AOT suite green: `test_aot.exe -use-aot dastest/dastest.das -- --use-aot --test tests`.
+4. MCP smoke: `compile_check` on `daslib/`, `tutorials/`, `D:\DASPKG\*\*.das`.
+5. Negative test: a `.das` emitting only `not_resolved_yet_*` errors prints them all (Rule 4 doesn't fire when no primary).
+6. Performance: time `dastest -- --test tests` before/after; expect small (<5%) speedup on tests that fail compile (verbose-plumbing avoids the work on intermediate passes).
+
+## Open notes / followups
+
+1. **Recategorization scope.** The 62 `recategorize` sites change cerr semantics, not just names. Most consequential clusters in the audit:
+   - 12 shadowing sites in ast_infer_type.cpp (L3298…L4901) marked `variable_not_found` → `already_declared_variable`
+   - 4 "returning … requires unsafe" sites carrying `invalid_return_type` → `unsafe_function_result`
+   - 3 call-depth sites carrying `too_many_infer_passes` → `exceeds_call_depth`
+   - 8 lookup-style codes reused for ambiguous-multi-match → `ambiguous_<subject>` family
+   - Per-site detail in `recategorize.tsv`.
+
+2. **Duplicate emit at ast_infer_type.cpp:1617/1620.** Audit flagged identical message emitted twice in adjacent lines. Drop the second during Phase C wording sweep.
+
+3. **`mismatching_parens` mislabel cluster** (7 sites in ds_lexer.lpp at L989, 1005, 1009, 1026, 1030, 1047, 1064) — message says "curly braces" or "square braces" but cerr is `mismatching_parens`. Retag during Phase C using `mismatching_curly_bracers` (existing) or new `mismatching_square_brackets`.
+
+4. **Inverted logic at ast_simulate.cpp:3874.** Audit flagged: `stt.size()==1` triggers "unknown structure" but should be `>1`. This is a *bug fix*, not an audit change. Apply alongside wording sweep but call it out in the PR description as such.
+
+5. **Sites with empty `current_cerr` (default unspecified).** ~120 sites total. After Phase C every site has a cerr; the `unspecified` code becomes a fallback for runtime-emitted errors only.
+
+6. **Tests with `expect` headers under-specified pre-refactor.** `dastest` accepts subset matches today, hiding the under-spec. Phase F.1 explicitly tightens these 5 to full multiset matches.
+
+7. **Parser-side `expect` directive.** Stays numeric (`expect 30305:3`). No syntax change. The 111-test sweep handles renumbering.
+
+8. **Backwards compatibility for code that reads `CompilationError::xxx` by-symbol.** All four consumer files (`utils/daScript/main.cpp`, `utils/daslang-live/main.cpp`, `utils/dasFormatter/`, `runtime_string.h`) currently flow numeric codes through; no symbol references observed. Re-verify during Phase C.