Repository navigation
Check reuse and report untracked reads inside rustc; findings 6 and 7 - #18
Merged
Merged
Conversation
zmaril
changed the title
zmaril
changed the title
zmaril
added this pull request to stack #19
October 7, 2026 20:17
…uild docs/hunt/verify-reuse.patch: under RUSTC_VERIFY_REUSE, reused metadata is encoded again and compared, and at the end of the session every green cached query value is computed again and compared by stable hash, Debug text and encoding, plus which allocations values share. With each of the three fixes reverted it reports that bug; with all three it is quiet. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
… embedded source Found by a short fuzz run at -Copt-level=2: binaries differed from clean builds only in ThinLTO's .llvm.<hash> suffixes, because a reused codegen unit's DIFile has the previous MD5 of the edited file. With -Zembed-source the object embeds the previous file. Report drafted, reproduction added to repro.sh. The fuzzer now names allocation-sharing findings by their queries. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
sourceContentRead flags reads of a source file's contents or hash. Over the compiler it finds finding 6, the same reads in rustc_codegen_cranelift (confirmed: with -Zembed-source its rebuilt object embeds the old file), and the metadata site of finding 4. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
verify-reuse.patch now also keeps each generated codegen unit's unoptimized IR and, when a later session reuses the unit, generates it again and compares (srcloc cookies left out). It flags finding 6 under -Zembed-source and is quiet on sink. Looking at its srcloc noise led to finding 7: warnings LLVM reports for inline assembly are not shown again when the unit is reused (1.60 through the nightly). Report, repro and a run-make test that fails on the pinned compiler. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
…) from P6 The threads fixture also loses its async fns (rust-lang/rust#162202). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
docs/hunt/report-untracked.patch: under RUSTC_REPORT_UNTRACKED, a read of an [UNTRACKED] option, of source file contents, or of the crate store's crate-wide state inside a task whose result may be reused is reported, unless the task declared it. On one sink build it names finding 5's three options and finding 6, and two new ones: -Zno-leak-check (an incremental rebuild accepts a program a clean build rejects) and -C extra-filename (reused metadata keeps the old value, since 1.90). fuzz.py and replay.py collect the lines in untracked.txt. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
… option Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
… crates Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
…erated all recomputes values computed this session too: none of 575,658 in a clean regex build differs, nor in the other crates or a threaded build. eval_static_initializer is no longer recomputed (it can make a definition). report-untracked.patch had been generated against the pinned commit and so contained everything; it is now relative to the fixes and the check, and the stack reproduces the compiler tree exactly. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
… from text Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
…things Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
…allocation twice Found by fuzzing with -Zmir-opt-level=4, named by the reuse check's sharing report (now printing the allocation), reduced automatically to three lines. On 1.60.0 through the nightly. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
…en-patches.sh Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh
zmaril
force-pushed
the
mirth/verify-reuse
branch
from
October 7, 2026 20:17
6d85c70 to
cf7b264
Compare
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
1 participant
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.
Stacked on #17.
The check.
docs/hunt/verify-reuse.patch(against the pinned rustc), on underRUSTC_VERIFY_REUSE, checks what an incremental session reused:Debugtext and standalone cache encoding, plus which allocations values share;With each of the three earlier fixes reverted it reports that bug; with all three it is quiet apart from two benign patterns (
docs/shadow-mode.md).rustc/fuzz.pyandrustc/replay.pyset it on every build.Untracked reads.
docs/hunt/report-untracked.patch, on underRUSTC_REPORT_UNTRACKED, reports every read of an[UNTRACKED]option, of source file contents or of the crate store's crate-wide state inside a task whose result may be reused, unless the task declared it (docs/untracked-reads.md). On one ordinary sink build it names finding 5's three options and finding 6, plus three more untracked options, folded into finding 5:-Zno-leak-check(an incremental rebuild accepts a program a clean build rejects),-C extra-filename(reused metadata keeps the old value, since 1.90) and-Zfuture-incompat-test.New bugs, with reports, reproductions in
docs/hunt/repro.shand run-make tests that fail on the pinned compiler (no fixes: each fix is a maintainer's trade-off):-Zembed-sourcethe old file; with optimizations the binary's ThinLTO symbol names differ from a clean build's. Since 1.44; Cranelift too.Harness.
fuzz.py --check(cargo check),--p5-builds(a second clean build tells nondeterminism from P6),replay.py --rustflags; runs under other flags, threads (withfixtures/sink-threads.patchworking around #162202) and real crates as fixtures; an Ur query for finding 6's pattern.🤖 Generated with Claude Code
https://claude.ai/code/session_018Mnrg9JXj9X1ht6Qkz2ybh