Repository navigation
Conversation
|
Full upstream regression testing completed on the patched PARI 2.19.0 source trees retained from the local Homebrew builds.
I also ran the additional targets that PARI explicitly excludes from
The extra I/O mismatch was reproduced with the unpatched Homebrew/core 2.19.0 bottle, with an identical diff. The macOS parallel test also exceeded a 30-second timeout with the unpatched bottle; the patched test originally stalled and was subsequently rechecked with the same timeout. These macOS runs were performed in the local command sandbox. The kernel check declares Thus the complete standard upstream |
Add a fork of Homebrew/core PARI 2.19.0 with the candidate complex Gaussian pivot fix reported in PARI bug #2708, investigated in cypari2 #234. Reject complex pivots whose squared norm is numerically zero, allowing the existing numerical recovery path to handle them.
Retain the mainline build settings and data-directory layout. Add regression tests for eigenvectors at every working precision from 15 through 193 bits and the shared matker/matrank/matimage/matindexrank failure at 53 bits. This is a candidate patch awaiting upstream review.
Validation: source builds, formula tests, strict audit, and style checks passed locally on macOS ARM64 (Apple Clang) and Linux ARM64 (brewed GCC 16). Installed Sage with Git cypari2 passes 48 precision checks on each of Python 3.13 and 3.14. Sage's PARI doctests have the same 130 failures with the original bottle and patched library; this patch does not address the separate precision/default-output differences.
The existing tap CI tests macOS Intel/ARM and Linux x86_64/ARM64. AI-assisted implementation and local validation by Codex.
On the local Tart Linux VM, the default GCC 13 source build crashed at GP startup with SIGILL; rebuilding with HOMEBREW_CC=gcc-16 passed. CI will check the default bottle build settings on the GitHub runners.