Abuild GLib archive

This repository preserves the historical GLib source revisions formerly selected by Abuild. The authoritative project is maintained in the GNOME GLib repository; the original documentation and license remain part of every preserved source revision.

The long-lived branches have deliberately separate roles:

• master mirrors the official upstream main branch;
• abuild is the final source revision selected by Abuild;
• archive/abuild-pin-c95e2ae9 preserves the other revision selected by historical Abuild commits; and
• abuild-gh adds only this maintenance README and files below .github/ to abuild.

The final abuild branch is exactly the official 2.39.92 tag at 599c669b. Abuild initially selected the official 2.61.0 tag at c95e2ae9 and replaced it with 2.39.92 one day before removing the GLib and QEMU submodules together.

No Abuild build rule or source dependency consumed this GLib submodule. Its short-lived inclusion accompanied the former vendored QEMU experiment, so the branches are retained for provenance and reproducibility rather than as a supported Abuild runtime dependency or a recommended GLib version.

Continuous integration

Run the same Trixie check locally with Docker:

.github/ci/run .github/ci/check

The check regenerates the historical Autotools files, disables the release's obsolete warning-as-error policy for the modern compiler, builds the GLib 2.39.92 core library, runs all of its Trixie-compatible upstream tests, stages the core installation, and verifies its header, static library, and source version. The gdatetime test binary is explicitly excluded because this old GLib release cannot parse current Trixie TZif data. The check also deliberately does not build the unused historical GIO tools: their code generator requires Python 2.5, which is not available in Trixie. Everything executes as the image's nobody identity in a read-only container without network access or Linux capabilities.

The old test suite contains randomized property tests. The archive harness passes a fixed GLib test seed so identical source and toolchains produce a repeatable result. This retains the gvariant coverage: an observed random seed reproducibly crashed its historical serialiser test, while the committed seed was verified repeatedly before it was selected.

The weekly upstream monitor compares master with the authoritative upstream main branch and checks all mirrored tags whose names are plain three-part version numbers. It reports drift but never updates branches automatically.