feat!: reuse freed memory only after a sweep and delete the analysis

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -1,5 +1,5 @@
 name: Bug report
-description: WeaveC crashed, rejected valid code, or accepted unsafe code
+description: weavec-cc crashed, trapped on a correct program, or let a memory bug through
 title: "[bug]: "
 labels: [bug, needs-triage]
 body:
@@ -8,10 +8,12 @@ body:
     attributes:
       label: Kind of bug
       options:
-        - False positive (safe code rejected)
-        - False negative (unsafe code accepted)
-        - Crash / assertion failure
-        - Wrong or unhelpful diagnostic
+        - False trap (a correct program stopped)
+        - Missed bug (an unsafe access ran)
+        - Crash / assertion failure in weavec-cc
+        - Wrong or unhelpful report
+        - Project fails to build with weavec-cc (drop-in)
+        - Performance (run time, build time, memory)
         - Build or installation problem
         - Other
     validations:
@@ -22,8 +24,9 @@ body:
     attributes:
       label: Minimal reproducer
       description: >
-        The smallest C program that shows the problem, plus the exact `weavec`
-        command line. Please avoid system headers where possible.
+        The smallest C program that shows the problem, plus the exact
+        `weavec-cc` command line and how you ran the program. Please avoid
+        system headers where possible.
       render: c
     validations:
       required: true
@@ -32,7 +35,10 @@ body:
     id: actual
     attributes:
       label: Actual behaviour
-      description: Full diagnostic output or crash backtrace.
+      description: >
+        The full report (`weavec: <kind> at ...`), compiler output or crash
+        backtrace. `WEAVEC_RT_BACKTRACE=1` adds the stack of the access that
+        stopped the program.
       render: text
     validations:
       required: true
@@ -47,8 +53,9 @@ body:
   - type: input
     id: version
     attributes:
-      label: weavec --version
-      placeholder: "weavec version 0.1.0-dev (abc123def456), built with LLVM 23.1.0"
+      label: weavec-cc --version
+      description: The `weavec-cc version` lines at the end of `weavec-cc --version`.
+      placeholder: "weavec-cc version 0.16.0 (abc123def456), built with LLVM 23.1.0"
     validations:
       required: true
 

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -1,5 +1,5 @@
 name: Feature request
-description: Propose an analysis, annotation, diagnostic, or tooling improvement
+description: Propose a guard, runtime, driver, or tooling improvement
 title: "[feature]: "
 labels: [enhancement, needs-triage]
 body:
@@ -16,8 +16,8 @@ body:
     attributes:
       label: Proposal
       description: >
-        How should WeaveC behave? Include example C code and the diagnostics or
-        inference results you would expect.
+        How should WeaveC behave? Include example C code and what you would
+        expect `weavec-cc` and the program it builds to do.
     validations:
       required: true
 
@@ -31,8 +31,8 @@ body:
     attributes:
       label: Affected areas
       options:
-        - label: Core ownership / borrow / lifetime model (`lib/Core`)
-        - label: Inference or checkers (`lib/Analysis`)
-        - label: Clang integration or driver (`lib/Frontend`, `tools/`)
-        - label: Annotations (`weavec.h`)
+        - label: Guard passes (`lib/Frontend/Guard*.cpp`)
+        - label: Library table (`lib/Core/LibrarySpec.txt`)
+        - label: Runtime (`runtime/`)
+        - label: Driver (`lib/Frontend/Driver.cpp`, `tools/weavec-cc`)
         - label: Build, CI, or documentation

.github/PULL_REQUEST_TEMPLATE.md

@@ -11,23 +11,23 @@ Small, focused PRs are much easier to review than large ones.
 
 <!--
 Anything a reviewer should know: alternatives considered, trade-offs,
-interactions with the ownership model, follow-up work. Changes to Core,
-checker rules, annotations or diagnostic ids must link the RFC in docs/rfcs/
-that specifies them.
+what it costs at run time or build time, follow-up work. Changes to the
+library table, the guard passes, the runtime, the annotation or the
+run-time report kinds must link the RFC in docs/rfcs/ that specifies them.
 Delete this section if not applicable.
 -->
 
 ## Test plan
 
-<!-- How was this verified? New lit/unit tests? Manual runs? -->
+<!-- How was this verified? New lit/unit tests, cases, corpus runs? -->
 
-- [ ] Unit tests added/updated (`unittests/`)
-- [ ] Integration tests added/updated (`test/`)
-- [ ] `ninja check-weavec` passes locally
+- [ ] Unit tests added/updated (`unittests/`, `runtime/test/rt_test.c`)
+- [ ] Integration tests added/updated (`test/Guards`, `test/Driver`, `test/cases`)
+- [ ] `ctest --preset dev` passes locally (a runtime change: on Darwin and Linux)
 
 ## Checklist
 
 - [ ] Code is formatted (`scripts/format.sh`) and passes clang-tidy
 - [ ] Public headers and behaviour changes are documented
 - [ ] Conventional Commit PR title and user-visible changes documented (changelog is generated)
-- [ ] No new dependencies on Clang/LLVM introduced into `lib/Core`
+- [ ] `scripts/check-hygiene.py` passes (no Clang/LLVM in `lib/Core`, runtime entry points declared in `weavec_rt.h`, line budgets)

.github/workflows/ci.yml

@@ -77,8 +77,8 @@ jobs:
           - name: Linux / Clang / Debug + ASan/UBSan
             os: ubuntu-24.04
             preset: ci-debug
-            # The test/cases suites analyse and build (guard passes and all)
-            # under ASan, without running the case programs.
+            # The test/cases suites build (guard passes and all) under ASan,
+            # without running the case programs.
             configure-args: -DWEAVEC_CASES_ARGS=--no-run
           - name: Linux / Clang / Release
             os: ubuntu-24.04
@@ -148,37 +148,27 @@ jobs:
           ulimit -c 0
           ctest --preset "${{ matrix.preset }}" -j "$(getconf _NPROCESSORS_ONLN)"
 
-      - name: Documentation examples
-        env:
-          ASAN_OPTIONS: detect_leaks=0:allow_user_poisoning=0
-        run: |
-          python3 docs/scripts/check-examples.py \
-            --weavec "build/${{ matrix.preset }}/bin/weavec" \
-            --weavec-cc "build/${{ matrix.preset }}/bin/weavec-cc"
-
-      # RFC 0035 gate H1: no checked-mode or old-engine residue, no library
-      # names compared outside the LibrarySpec table, no corpus project names
-      # in lib/, the engine seam, and the engine, library and runtime line
-      # budgets.
+      # RFC 0036 gate H1: no library names compared outside the LibrarySpec
+      # table, no corpus project names in lib/, the layering of lib/Core and
+      # the runtime, the runtime's entry points against the passes', and the
+      # library and runtime line budgets.
       - name: Repository hygiene
         if: runner.os == 'Linux' && matrix.preset == 'ci-release'
         run: |
           python3 scripts/test_check_hygiene.py
           python3 scripts/check-hygiene.py
 
       # RFC 0035 (section 11): the corpus gate's own tests, then --quick over
-      # pinned corpus configurations in test/corpus/: the weavec analysis
-      # against the triage in test/corpus/triage.json, and the enforcement
-      # ledgers of per-file weavec-cc compiles against the ratchet in
+      # pinned corpus configurations in test/corpus/: the enforcement ledgers
+      # of per-file weavec-cc compiles against the ratchet in
       # test/corpus/expected.json. The weekly Corpus workflow runs --full.
       - name: Pinned corpus
         if: matrix.preset == 'ci-release'
         run: |
           python3 scripts/test_corpus_gate.py
           python3 scripts/corpus-gate.py --quick \
-            --weavec "build/${{ matrix.preset }}/bin/weavec" \
             --weavec-cc "build/${{ matrix.preset }}/bin/weavec-cc" \
-            --only log.c --only cJSON-program --only linenoise-program \
+            --only log.c --only cJSON --only linenoise-program \
             --only jansson --json corpus-gate-quick.json
 
       - name: Upload pinned corpus results
@@ -192,7 +182,6 @@ jobs:
       - name: Smoke-test install
         run: |
           cmake --install "build/${{ matrix.preset }}" --prefix "$RUNNER_TEMP/weavec-install"
-          "$RUNNER_TEMP/weavec-install/bin/weavec" --version
           "$RUNNER_TEMP/weavec-install/bin/weavec-cc" --version
           printf 'int main(void) { return 0; }\n' > "$RUNNER_TEMP/hello.c"
           "$RUNNER_TEMP/weavec-install/bin/weavec-cc" "$RUNNER_TEMP/hello.c" -o "$RUNNER_TEMP/hello"

.github/workflows/corpus.yml

@@ -1,10 +1,10 @@
 name: Corpus
 
 # Runs the full corpus gate (RFC 0035, section 11) over the projects pinned
-# by SHA in test/corpus/manifest.json, but the sealed sets: the weavec
-# analysis and the per-file enforcement ledgers (checked against
-# test/corpus/triage.json and test/corpus/expected.json), each project's
-# build and test suite with CC=weavec-cc in trap mode (rerun in report mode,
+# by SHA in test/corpus/manifest.json, but the sealed sets: the per-file
+# enforcement ledgers (checked against test/corpus/expected.json), each
+# project's build and test suite with CC=weavec-cc in trap mode (rerun in
+# report mode, with the guard failures of test/corpus/triage.json allowed,
 # and built once more with the reference compiler for the build-time gate),
 # the injected bugs and the benchmarks (against the reference compiler and
 # ASan). The gate clones the projects into build/corpus. Pull requests run
@@ -61,7 +61,6 @@ jobs:
       - name: Run corpus gate
         run: |
           python3 scripts/corpus-gate.py --full \
-            --weavec build/ci-release/bin/weavec \
             --weavec-cc build/ci-release/bin/weavec-cc \
             --json corpus-gate-full.json