ci/spec22-gems: incident 13 round-8 — FFI raw LoadLibraryEx matrix + library_aliases manifest - #93
Merged
Merged
Conversation
…ses manifest (incident 13) Round 7 pinned the failure to ffi's win32 dl_open: LoadLibraryExA with LOAD_WITH_ALTERED_SEARCH_PATH on the covered VFS path resolves the main module but not its dependency search at require time (error 126 on lib/sassc/libsass.so with every byte verified and every solo-load green). probe.rb: - The fiddle flag matrix dies with ruby 4.0 (fiddle left the default gems; LoadError is ScriptError) and is replaced by sassc_raw_matrix — kernel32 LoadLibraryExA/FreeLibrary/GetLastError bound through FFI itself: libsass.so vfs/host x flags 0x0/0x8/0x1100 plus the sibling runtimes, each freed after success. - sassc_ffi_load_legs reordered (libsass first, siblings, libsass:again), LoadError rescued alongside StandardError, the nonexistent FFI::DynamicLibrary#free call dropped. - New ARGV mode 'sassc-matrix' (windows-gated; posix prints skip). run-msys.sh: - stamp_library_aliases() finds the sassc gem's *.dll closure at press time and appends library_aliases: entries to BOTH probe manifests (aliases are orthogonal to the materialize: negative oracle), with post-press greps asserting the stamp landed. - Fifth dogfood leg 'sassc-matrix' with st_matrix gating; the canary cross-check counts it. Rides tebako v0.1.9 (ccecb275): the driver's alias.rs boot-materializes declared aliases and PREPENDS their dirs to the process PATH — the raw-surface answer, interception-free by design. rspec 110/110 green; ruby -c / bash -n clean.
…elling (incident 13)
Round 8's verdict: the library_aliases fix WORKS — require sassc +
sassc-main green on the alias-materialized PATH; the unmaterialized
oracle red exactly as pinned; sinatra 200 x2. The one red leg,
sassc-partial, is a probe bug: TEBAKO_EXEC_CACHE arrives in the host
spelling (backslashes) and Dir.glob treats backslash as an escape on
windows, so the materialization check globbed a degenerate pattern and
reported hits=0 while the driver log shows the styles tree materialized
fine (resources/4afd6aab980ce2ab/probe/styles). The same trap kept the
raw matrix's host legs 'skipped (no spelling)'.
Both Dir.glob sites now normalize with tr('\\', '/') — a no-op on
posix, where this leg was already green. rspec 110/110.
ronaldtse
marked this pull request as ready for review
August 19, 2026 08:17
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
2 participants
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.
Incident 13, round 8 — mechanism matrix + the alias fix experiment
Round 7 (run 32211646613) narrowed the sassc-on-tebako-windows failure to one
mechanism: ffi 1.17.4's win32
dl_opencallsLoadLibraryExA(name, NULL, LOAD_WITH_ALTERED_SEARCH_PATH)on the full VFSpath. The covered route resolves the main module (every solo-load green, every
byte sha256-verified) but the dependency search of that call fails at
requiretime — error 126 onlib/sassc/libsass.so. The sibling runtime DLLs(
libwinpthread-1.dll,libgcc_s_seh-1.dll) live co-dir in the image andmaterialize fine.
This round changes the probe in two directions at once:
1.
sassc-matrix— the raw mechanism recordThe fiddle flag matrix died with ruby 4.0 (fiddle left the default gems;
LoadErrorisScriptError, notStandardError). It is replaced bysassc_raw_matrix: kernel32LoadLibraryExA/FreeLibrary/GetLastErrorbound through FFI itself, run against:
lib/sassc/libsass.so— vfs spelling vs host spelling × flags0x0/0x8(ALTERED) /0x1100(ALTERED|LIMITED)0x8Each successful handle is freed. This is the diagnostic that decides the fix
surface if the alias route fails:
vfs:alteredfailing means the coveredroute cannot serve the main module at all (driver surface change); succeeding
while sibling raw-loads fail means the search order is still broken.
2.
library_aliases:— the designed fix, exercised end to endrun-msys.shnow stampslibrary_aliases:into BOTH probe manifests at presstime (the sassc gem's
*.dllclosure, deduped by name) and greps that thestamp landed. This rides tebako v0.1.9 (
ccecb275): the driver'salias.rsboot-materializes declared aliases and PREPENDS their dirs to the process
PATH — the raw-surface answer, interception-free by design. Aliases are
orthogonal to the
materialize:negative oracle, which stays.Acceptance (unchanged)
proof-sassc.logmust carryPROBE gem-loaded no+PROBE sassc-main ok+PROBE sassc-partial ok+ the unmaterialized leg's pinned expected-fail +SPEC22-GEMS-MSYS-ACCEPTANCE-OK.proof-sassc-matrix.logis the mechanismrecord.
Also fixed:
sassc_ffi_load_legsreordered (libsass first, siblings,libsass:again),
LoadErrorrescued alongsideStandardError, thenonexistent
FFI::DynamicLibrary#freecall dropped, and module paths resolvevia
Gem::Specification.find_by_name.Local gates:
rspec110/110,ruby -c/bash -nclean.Draft — do not merge. The harness runs UNMERGED via the factory's new
harness_refdispatch input; merging is a C3 close-out decision, not aprerequisite for the probe.