Bug triage: nine audit rulings fixed red-first, the temp-string AOT desync, and the gen1_to_gen2 rename

[borisbat]

Bug-triage follow-up to the docs arc (#3661): the audit surfaced 9 code-bug candidates, kept out of the docs PR by design. Every one is now ruled on and fixed here — each with a red-first regression test — plus the converter-binary rename that fell out of candidate 6.

Compiler

• operator []<- dispatch wired in ExprMove (f970b71) — the grammar accepted the operator but ExprMove never dispatched it, so a user-defined []<- was silently unreachable. Mirrors the existing []= promoter. Tests in tests/language/operators.das.
• require X as Y registers the alias for every require form (b04c0f2) — importName was only set for path-form requires (fs_file_info.cpp), so require daslib/strings_boost as sb was a silent no-op. Explicit as now registers unconditionally; importName guards only the implicit path-stem registration. Tests pin builtin/daslib/same-dir/path forms, with-module honor, additive alias, and the newly-reachable 20510 collision (require_as.das + fixture + failed_require_as.das).
• [nodiscard] results consumed by string interpolation are not discarded (7c3928e) — InferTypes never marked string-builder elements as consumed, so any [nodiscard] call inside "{...}" tripped a false error[30166], in call arguments (print("{f()}")) and let-inits alike. New preVisitStringBuilderElement override runs markNoDiscard on every element; a genuinely discarded statement-level call still errors (failed_nodiscard.das).

daslib

• each_ref crash fixed (2b247c6) — the ref-generator lowering builds a lambda whose parameter is a reference to a pointer; each_ref's signature didn't match that shape, and the mismatch went uncaught all the way to an AV in SimNode_ForWithIterator. Signature is now ? & with an ==& trap overload that turns the wrong shape into concept_assert 31400 at compile time. Tests: each_ref.das + failed_each_ref.das.
• qmatch captures are transactional (981e776 + fcd9154) — ALL 8 capture kinds ($e $v $i $t $c $f $b $a) leaked bindings when a match FAILED partway ( $v/$t were half-guarded, the rest not at all), so a later alternative saw stale captures. Captures now stage into temps and commit only on full success; dual-arm const-constructor forms dedup via a temps table with copy-init identity commits. Follow-up: array-capture temp buffers free deterministically (QmCaptures struct, finally cleanup). 12 new tests pin every kind — 10 were red before the fix; linq/flatten/sql consumers all green.
• linq _fold(chain) wrapping form defers until the chain type resolves (c785e2f) — LinqFold fired before instance resolution while select's dependent typedecl return type was still auto/alias, failing infer. One-line isAutoOrAlias defer. The standalone red/green test matters: the window needs a FRESH select instance, so an in-file test was masked by cache warmth.
• sql check_schema/schema_from see GENERATED columns (6a57d61) — PRAGMA table_info hides generated columns; switched to table_xinfo + the generated flag, with floor/ceiling count gates, a computed cross-check against @sql_computed, and schema_from skipping generated-column synthesis. TDD red→green; dasllama-server and the dictation bot verified equivalent-behavior.

builtin

• to_lower_in_place / to_upper_in_place removed from the das surface (b696734) — registered SideEffects::none while mutating their argument, and silently no-oping on literals. The C++ helpers stay (internal callers); das code uses the value-returning forms.

Tooling

• Converter binary renamed das-fmt → gen1_to_gen2 (e1818f3) — two different tools shared one name: the cmake das-fmt target (utils/dasFormatter, the gen1→gen2 syntax converter) and the source formatter utils/das-fmt/dasfmt.das (also compiled to bin/das-fmt.exe by CI). An SDK user typing das-fmt to format a file got their syntax converted instead. Renamed everywhere: cmake target + install, run_utils_tests, extended_checks build targets, bundle smoke EXE presence, shipped-skills exe regex, usage text, MCP convert_to_gen2 exe path, and the skills/CLAUDE.md name-trap notes. The formatter keeps the das-fmt name; the windows-ninja lane collision (both tools landing at ./bin/das-fmt) is gone as a side effect. Also drops three dead get_target_property(DAS_FMT_*) lines nothing consumed.

Found by preflight: temp-string AOT desync (fae23af)

The full AOT sweep for this branch turned up 162 × error[50101] across every interpolation-heavy daslib module (json, jsonrpc, clargs, logger, sql) — a pre-existing master bug shipped with the temp-string conversions (#3657), invisible to per-PR CI (which only builds test_aot_subset). The temp-string wrapper pass fired only when the driving program had persistent_heap on, and it mutates function bodies — shared-module ASTs included — so the same daslib function compiled to different trees (and AOT hashes) depending on who compiled it first: macro-context compiles (which run with macro_context_persistent_heap) wrapped shared daslib functions that AOT stub generation left bare.

Fix: the wrapper is now always inserted — a function's tree never depends on the driver — and the heap modes move to runtime where they belong: freeTempString already no-ops for interned heaps, linear-heap frees are safe bump-retreat no-ops, and disable_temp_string_reclaim becomes a runtime flag on the string heap (set at simulate from the entry program) instead of a hash-poisoning compile gate. New tests-cpp case pins the in-process invariant; the AOT sweep is the cross-process proof (162 → 0).

Also in the branch (e8dbd47): preflight-surfaced fixes — both-worlds LINT019 spellings on the qmatch template-line nolints, the each_ref handmade doc for its new signature hash (orphaned old-hash file removed), a das2rst group for the qm_tmp_* staging helpers, and a preflight exe-rail fix (the temp-path das-lint binary couldn't resolve libDaScriptDyn.dll; the daslang bin dir now rides the loader's environment).

Validation

• Full local preflight (format, lint, dasgen, ci-das compile sweep, docs incl. sphinx -W, ctest, interp/JIT/AOT suites, sequence smoke).
• Full tests/ sweep: 13300 passed / 0 failed; tests/language: 1548/1548.
• gen1_to_gen2 --tests green; MCP convert + tools tests green.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR is a bug-triage follow-up that fixes multiple compiler/daslib/runtime issues found during the docs arc, adds red-first regression coverage, resolves an AOT hash determinism bug tied to temp-string wrapping, and removes tooling ambiguity by renaming the gen1→gen2 converter binary.

Changes:

• Fixes compiler/operator/require-aliasing and [nodiscard] interpolation handling, with new language regression tests.
• Fixes daslib behaviors (qmatch transactional captures, linq _fold(chain) inference deferral) and SQLite schema introspection for GENERATED/@sql_computed columns, with targeted tests.
• Makes temp-string wrapping driver-independent (AOT hash determinism), and renames the converter das-fmt → gen1_to_gen2 across tooling/CI/docs/MCP.

Reviewed changes

Copilot reviewed 61 out of 61 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
utils/preflight/main.das	Prepends daslang bin dir to loader search path before running temp-path das-lint -exe.
utils/mcp/tools/convert_to_gen2.das	Updates MCP convert tool to call gen1_to_gen2 and renames helper.
utils/mcp/registry_das.das	Updates tool registry text and adds STYLE038 nolint to tool table.
utils/mcp/README.md	Documents convert_to_gen2 as using gen1_to_gen2.
utils/dasFormatter/Readme.md	Updates converter usage examples to gen1_to_gen2.
utils/dasFormatter/main.cpp	Updates converter help text to gen1_to_gen2.
utils/CMakeLists.txt	Updates utils tests to depend on/run gen1_to_gen2 --tests.
tests/strings/test_cpp_functions.das	Removes in-place case conversion from covered C++ string functions list; adds STYLE038 nolint.
tests/README.md	Updates tests index text for test_cpp_functions.das.
tests/linq/test_linq_fold_wrap_defer.das	Adds regression test for _fold(chain) deferral until type resolves.
tests/language/require_as.das	Adds coverage that require X as Y aliases work for all require forms.
tests/language/require_as_fixture.das	Adds same-dir fixture module for require-as tests.
tests/language/operators.das	Adds tests for []<- move-store operator dispatch (free-form + method + ref-return).
tests/language/nodiscard.das	Adds regression test that [nodiscard] results inside interpolation are “consumed”.
tests/language/failed_require_as.das	Negative test for alias collision (20510).
tests/language/failed_nodiscard.das	Negative test for genuinely-discarded [nodiscard] call.
tests/language/failed_each_ref.das	Negative test that wrong each_ref lambda shape is rejected with 31400.
tests/language/each_ref.das	Positive tests for corrected each_ref signature and generator lowering.
tests/dasSQLITE/test_check_schema_computed.das	Adds SQLite tests covering GENERATED columns and struct/DB computed-ness rules.
tests/ast_match/test_qmatch_no_bind_on_fail.das	Adds extensive regression coverage for transactional qmatch captures across all tags.
tests-cpp/small/test_temp_wrap_hash_persistent.das	Persistent-heap entry program for hash determinism test.
tests-cpp/small/test_temp_wrap_hash_default.das	Default-heap entry program for hash determinism test.
tests-cpp/small/test_temp_wrap_fixture.das	Fixture module whose function hash must be driver-independent.
tests-cpp/small/test_temp_string_wrap_determinism.cpp	New doctest ensuring function hash is driver-independent across driver options.
src/parser/parser_impl.cpp	Fixes require ... as ... alias registration to apply to all require forms.
src/builtin/module_builtin_string.cpp	Removes to_lower_in_place / to_upper_in_place from the das surface.
src/ast/ast_simulate.cpp	Moves disable_temp_string_reclaim handling to runtime heap configuration.
src/ast/ast_infer_type.cpp	Marks string-builder interpolation elements as “consumed” for [nodiscard].
src/ast/ast_infer_type_op.cpp	Adds []<- operator dispatch path to ExprMove on ExprAt LHS.
src/ast/ast_allocate_stack.cpp	Makes temp-string wrapping pass unconditional to avoid AOT hash desync.
skills/strings.md	Updates guidance to drop in-place case conversion API.
skills/sql.md	Updates schema checking guidance to include GENERATED/@sql_computed behavior.
skills/preflight.md	Updates “name trap” documentation for converter vs formatter.
skills/mcp_tools.md	Updates MCP tool docs to reference gen1_to_gen2.
skills/make_pr.md	Updates converter vs formatter naming note to gen1_to_gen2.
skills/daslang/references/macros.md	Documents transactional qmatch capture semantics.
skills/das_macros.md	Updates qmatch capture semantics (transactional on failure).
skills/das_formatting.md	Updates converter vs formatter warning to reflect gen1_to_gen2.
modules/dasSQLITE/PROVIDER_CONTRACT.md	Updates provider contract to specify table_xinfo behavior.
modules/dasSQLITE/daslib/sqlite_provider.das	Switches schema introspection to PRAGMA table_xinfo, tracks GENERATED columns.
modules/dasSQLITE/daslib/sqlite_boost.das	Updates try_check_schema to handle/generated columns and computed-ness mismatches.
include/daScript/simulate/simulate.h	Makes freeTempString respect runtime reclaim-disabled flag.
include/daScript/simulate/heap.h	Adds reclaim-disabled flag plumbing to string heap.
include/daScript/ast/ast_infer_type.h	Declares preVisitStringBuilderElement override in infer visitor.
doc/source/stdlib/handmade/function-builtin-each_ref-0xd5c96473551e8fec.rst	Adds updated handmade doc for new each_ref signature hash.
doc/source/stdlib/handmade/function-builtin-each_ref-0x518c9960a2242c9.rst	Removes old handmade doc for prior each_ref signature hash.
doc/source/reference/utils/mcp.rst	Updates MCP docs to reference gen1_to_gen2.
doc/source/reference/tutorials/sql_02_insert_data.rst	Updates tutorial text for computed-vs-generated schema checks.
doc/source/reference/language/functions.rst	Documents []<- semantics and provides an example.
doc/reflections/das2rst.das	Updates reflection grouping to drop removed in-place case funcs; adds qmatch tmp staging group.
daslib/sql.das	Adds is_computed field to column info.
daslib/sql_provider.das	Adds generated field to SchemaFromCol.
daslib/sql_boost.das	Validates computed-ness vs GENERATED in schema_from; avoids synthesizing fields for GENERATED cols; carries is_computed.
daslib/linq_fold.das	Defers _fold(chain) macro until chain type is fully inferred.
daslib/builtin.das	Fixes each_ref signature; adds trap overload for wrong lambda shape; tweaks error text punctuation.
daslib/ast_match.das	Implements transactional qmatch captures (staging temps + commit-on-success) and adds cleanup discipline.
CMakeLists.txt	Renames converter target to gen1_to_gen2 and removes unused target-property queries.
CLAUDE.md	Updates formatter reminder text to reference gen1_to_gen2.
ci/smoke_test_bundle.sh	Updates bundle exe presence checks from das-fmt to gen1_to_gen2.
ci/check_shipped_skills.py	Extends shipped-skill exe regex to include gen1_to_gen2.exe.
.github/workflows/extended_checks.yml	Updates extended_checks build targets/comments to use gen1_to_gen2.

Suppressed comments (1)

utils/mcp/tools/convert_to_gen2.das:34

• do_convert_to_gen2 builds a shell command string that embeds file and executes it via popen(cmd). This is unsafe for paths containing spaces/quotes and can become command-injection if a caller passes a crafted filename (the MCP tool is externally callable). Prefer popen_argv with an argv array so the converter path and the file path are passed as literal arguments.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 61 out of 61 changed files in this pull request and generated no new comments.

Suppressed comments (1)

utils/mcp/tools/convert_to_gen2.das:34

• do_convert_to_gen2 builds a shell command string and runs it via popen(cmd). That makes conversion fail for file paths with spaces, and it also opens the door to shell injection if an MCP client passes a path containing shell metacharacters. Prefer popen_argv with an argv array (and use path_join + a Windows .exe suffix) so the executable and file path are passed verbatim.