PyCForge 1.3.86 / PCFC11 0.2.33 / PyCForge SDK 0.6.4 unchanged
November 4, 2026 — unpublished development handoff

Completed increment

PCFC11's shared integer control-flow graph now admits statement switch dispatch
and fallthrough. The selector is evaluated once in its original position. Case
entry, default/no-match exits, nested switches, break/continue and direct-label
paths are represented. Constant, copy and known-bit analyses use these edges.
All switch arms remain possible to the analysis, including for a constant
selector. Case-value refinement, reaching definitions, SSA, shared backward
liveness, general alias analysis and broader allocation are not introduced.

The scan validates case ownership and dispatch metadata before publishing a
verified graph. Cases inside statement expressions and computed control retain
conservative fallback, as do resource or allocation failures. Existing bounds
were not increased. No new C syntax or Python/import contract is advertised.
See the pinned source's docs/CFG_SWITCH.md for the precise contract.

README maintenance

The v1.3.78 Classes & Functions/navigation row was removed from the Conversion
bug tracker. The current behavior is now described under editor features, while
its repair history remains in the original engineering records. The version
heading is 1.3.86. The approved four images and single SDK PyPI link are unchanged.
The SDK README already has purposeful sections and matches the unchanged 0.6.4
wheel metadata byte for byte; its product text and version were left unchanged.
Both governing README guidelines now require correct categorization and only
applicable version updates when no product-information correction is needed.

Full application suite

The complete unfiltered run passed 3,908 cases, skipped 39 and failed zero,
accounting for all 3,947 collected cases. It used four workers, the pinned
PCFC11_PATH and a 180-second outer watchdog; it completed in 640.25 seconds.
Twenty skips require unavailable Clang, 18 require sealed historical artifacts,
and one requires an environment without PySide6. That last test passed separately
in the SDK environment without Qt. Two existing QDragEnterEvent deprecation
warnings remain. No test was deselected or assertion weakened.

The terminal counts agree with all 3,947 unique JUnit testcase records. The
reporter's aggregate tests attribute is inconsistent (8,943); the original XML
is retained, and full-accounting.json derives counts from individual outcomes.
This reporting discrepancy is not counted as additional tested cases.

Compiler qualification

- The 91-function switch corpus checks 64 inputs and 17,472 independently
  calculated result/effect values per execution profile. GCC O0/O2 and UBSan,
  PCFC11 native O0/O1 and Windows instruction-model O0/O1 all pass. Two O1
  assembly assertions per target show that proven facts are actually used.
- Direct builder/verifier/dataflow checks pass 3,397 assertions per GCC O0/O2,
  PCFC11 O0/O1 and ASan/UBSan configuration. The suite covers malformed ownership,
  unreachable preludes, fallthrough, effect barriers, allocation/budget refusal
  and preserved output after rejected source. The unchanged 0.2.32 compiler
  passes the same semantic corpus.
- All 42 standard Linux feedback phases pass. The inherited O0/O1 runs,
  self-compilation and 32 suites through the self-compiled compiler pass:
  35 recorded steps. The sanitizer build and all 32 suites pass after the
  test-layout corrections below. Eight additional reference/toolchain gates pass.
- Independent Windows-target C assembly preflight passes 326 checks on Linux.
  This is not native Windows execution. No broad performance claim is made.

Fresh applications, memory and installed product

All 53 freshly converted source bundles pass their 265 native execution variants
and eight production BuildService profiles. The four complete application
oracles and failure/stale/cancel preservation checks retain their established
contracts. Thirty-nine PYC4699 Runtime-preview warnings remain explicitly recorded.
The installed managed compiler emits Windows assembly for all 53 fresh bundles
at O0/O1 (106 checks), with per-case source and assembly hashes retained.

The regenerated 40-case GC kit passes 80/80 checks with PCFC11 and 80/80 with
GCC on Linux. Installed wheel checks compare every shipped product byte, build
the pinned compiler through ManagedCompiler, verify its recorded identity and
execute converted Python plus a direct switch-entry program at O0 and O1.
The installed Qt version check passes application/startup/About and every Help
footer. All eight source gates pass: Ting validation, promotion checks, conversion
documentation, README rendering, SDK catalogue, bytecode, Ruff and dependencies.

SDK compatibility

SDK 0.6.4 was installed separately. All seven starters complete with both PCFC11
and GCC: 119 native fixtures and 82 expected rejections per compiler. These
workflows verify readiness, identical repeated packages, explicit registration
and byte-preserving rejection after an author edit. The independent vector and
record namespaces participate. GCC uses the shipping wheel in a fresh engine
environment; a separate starter also passes with that wheel and its managed
PCFC11 compiler. No SDK interface or client change is required.

Platform feedback and limits

CC reports similar successful Windows results for the preceding PyCForge/SDK
baseline and no Linux run on the Windows-only laptop. That user-reported feedback
is preserved separately. Exact logs, tested case IDs and toolchain/display details
were not supplied in this turn. These Linux checks do not establish native
Windows execution of the new 1.3.86 / 0.2.33 increment or an unaided author trial.
The SDK's declared native verification profile remains Linux. Finite tests do
not establish correctness for arbitrary C/Python programs or every environment.

Retained setup and preliminary attempts

The first application run was stopped after discovering that PCFC11_PATH was
unset, so native array tests were skipping. Its incomplete log is retained and
is not a complete-suite pass. The final unfiltered run explicitly selects the
pinned compiler. No product or test assertion changed for that restart.

The first new assembly witness used compound assignment, which the existing
frontend lowers through an address-taken temporary. The final corpus retains
that semantic witness and adds an ordinary local assignment for the intended
known-bit assertion. No expected result was changed. A preliminary label run
hit this host's LeakSanitizer process-inspection restriction. Address/undefined
checks run with detect_leaks=0; leak detection remains unqualified.

A relocated sanitizer executable initially lacked its adjacent include directory.
The two affected gates were rerun with the correct header layout. An additional
toolchain gate was initially given an unsupported --compiler argument; it passed
with its documented interface. Original failures and corrected reports remain
linked. No compiler implementation change was needed for these setup corrections.
The first Windows assembly collector recorded duplicate generated/harnessed
sources under reused output names; the final installed-compiler collector retains
unique per-case assembly and hashes and supplies the reported 106-check result.

Evidence and next work

All records are under evidence/pcfc11_switch_1_3_86, with EVIDENCE_INDEX.txt as
the entry point. Historical predecessor evidence remains unchanged. The roadmap
now proceeds to reaching definitions/use chains on the shared topology, while
keeping measured large-function lowering and native-platform work in its queue.
The source distribution retains maintained tests, tools, instructions and the
current pinned compiler; bulk evidence and older compiler archives stay in the
complete development ZIP. Final package parity and SHA-256 values are recorded
in the accompanying delivery report. CC alone decides and performs publication.
