Upgrade llama.cpp b11062 → b11069; drop patch 0011 - #443
Merged
Merged
Conversation
Seven upstream commits, 57 KiB, no project-source change. tools/server/, common/arg.*, src/llama-model.* and ggml/include are byte-identical across the range, so the server wire contract cannot have moved and eight of the nine local patches apply unchanged. Drop patches/0011 (lenient invalid UTF-8 in the PEG parser): upstream #29161 (first tagged b11063) fixed the same defect independently and more broadly. The until-parser now consumes every undecodable run and common_chat_peg_mapper emits one U+FFFD per run with the text after it kept, where the patch returned only the text up to the bad byte. The hunk the patch modified no longer exists, so the applier failed loud and the patch is dropped rather than refreshed, per the 0009/0013 precedent. Re-point the ContentOnlyParseUtf8 guard in test_utils.cpp at upstream's replacement contract (six cases; the new one pins the run boundary), so a future revert to FAIL still reds C++ Tests on every platform. Record the drop in CLAUDE.md, TODO.md, CHANGELOG.md, the verify-patches-applied.sh header and the breaking-changes history. Verified from a fresh configure: 8 patches stamped at 68d9053af, ctest 552/552, 40 Java_* exports / 0 mangled, NativeLibraryLoadSmokeTest 4/4. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TCpJSw2Qi3H5b51vquYrEH
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 10:45 — with 
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 10:45 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
startgate
September 21, 2026 10:45 — with
GitHub Actions
Error
GitHub Actions
Error
|
This branch had an error being deployed
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.
Summary
CMakeLists.txt,CLAUDE.md,README.md, and Java version constants.0011(lenient invalid-UTF-8 handling in PEG parser). Upstream llama.cpp #29161 (first tagged b11063) fixed the same issue independently and more broadly: instead of truncating at the bad byte, it now replaces each undecodable run with exactly one U+FFFD (Unicode "maximal subpart" rule) and preserves text after it.ContentOnlyParseUtf8C++ tests to pin upstream's replacement contract: malformed bytes are now replaced with U+FFFD rather than causing truncation. A trailing incomplete sequence at EOF is still withheld (not replaced), as before.Behavioral change
Before (patch 0011): A lone continuation byte
\x80in"Hello\x80World"would parse as"Hello"(truncated).After (upstream ggml-org/llama.cpp#29161): The same input now parses as
"Hello\uFFFDWorld"(replaced + text after preserved). Two adjacent invalid bytes\xFF\xFEbecome two separate U+FFFD replacements, not one.The fix is broader than the patch: it handles the invalid-UTF-8 case at the AST level (recording undecodable runs on nodes) rather than just at the parse-result level, so
common_chat_peg_mappercan emit sanitized text for every node.Test plan
ContentOnlyParseUtf8C++ tests updated and passing (4 tests covering lone continuation byte, truncated sequence, invalid lead byte, adjacent invalid bytes, and incomplete trailing sequence)Related issues / PRs
CLAUDE.mdpatch table for patch 0011 (now dropped, like 0009 and 0013 before it)Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_01TCpJSw2Qi3H5b51vquYrEH