Landing: 37a9529bcc8b

← All landings

Project / Subsystem

gcc / libstdc++

Date

2026-04-16

Author

Jonathan Wakely

Commit

37a9529bcc8b2e550b39fff2f7b55ed578fb7e19

Source

github

Impact

4/10

Perf win

Breaking

View on GitHub →

All attributes

project: gcc
subsystem: libstdc++
patch_id: —
commit_hash: 37a9529bcc8b2e550b39fff2f7b55ed578fb7e19
source_type: github
headline: Libstdc++: Skip std::generate_canonical tests for IBM long double.
tldr: Skips tests for `std::generate_canonical` with long double type on POWER architectures using IBM's long double format.
why_it_matters: Avoids test failures on platforms where the long double implementation does not conform to IEEE semantics, ensuring more accurate test results.
author: Jonathan Wakely
impact_score: 4
impact_rationale: Fixes test failures on specific platforms, but impact is limited to testsuite results.
outcome: committed
performance_win: false
breaking_change: false
series_id: —
series_parts: []
tags: • libstdc++
• testing
• long double
• IBM
• POWER
discussion_id_link: —
bugzilla_pr: —
url: https://github.com/gcc-mirror/gcc/commit/37a9529bcc8b2e550b39fff2f7b55ed578fb7e19
date: 2026-04-16T00:00:00.000Z

The std::generate_canonical tests assume IEEE floating-point semantics, which are not met by IBM’s long double format used on POWER architectures. This commit skips the long double portion of the std::generate_canonical tests when compiling for such platforms to prevent spurious test failures.