Files
searchi/tools/glib-regression/README.md
T

35 lines
1.5 KiB
Markdown

# Optimized GLib backport regression
Run on native Linux with the GLib development headers and the normal Rust
toolchain. This isolated harness selects searchi's exact vendored GLib crate.
Before building, verify all upstream bytes and the exact patch against the
official crate archive already available in the Cargo download cache:
```sh
python tools/glib-regression/verify_vendor.py /path/to/glib-0.18.5.crate
```
This reads the archive without extracting it or changing the Cargo cache and
rejects extra, missing or unexpectedly changed vendor files.
```sh
cargo test --manifest-path tools/glib-regression/Cargo.toml --release --locked --jobs 1
```
The release profile is essential: the upstream defect was sensitive to compiler
optimization. Tests exercise all five affected iterator operations, Unicode,
empty arrays, exhaustion and exact-size behavior. A compile-fail doctest verifies
borrowed strings cannot outlive their owning variant. No embeddings, media
decoders, network inference or GPU providers are used.
This harness does not replace a native desktop build/interaction check or an
advisory audit. Record actual execution and selected package identity before
calling the local backport validated.
Recorded execution, 2026-10-13: Ubuntu 24.04 WSL, Rust 1.99.0, GLib 2.80.0,
three optimized unit tests and one compile-fail doctest passed. Archive bytes
and exact patch were independently reverified. See
[dependency review](../../docs/DEPENDENCY-REVIEW.md) for hashes, compiler warnings
and remaining native desktop/security checks.