Repository navigation
Regenerate the benchmark report - #421
Conversation
|
Important Review skippedReview was skipped due to path filters ⛔ Files ignored due to path filters: 49
CodeRabbit blocks several paths by default. You can override this behavior by explicitly including those paths in the path filters. For example, including
|
[Low impact] The reviewed documentation changes appear safe to merge.
|
|
An automated preview of the documentation is available at https://421.corosio.prtest3.cppalliance.org/index.html If more commits are pushed to the pull request, the docs will rebuild at the same URL. 2026-11-16 18:08:06 UTC |
|
GCOVR code coverage report https://421.corosio.prtest3.cppalliance.org/gcovr/index.html Build time: 2026-11-16 18:24:16 UTC |
0f039af to
ffca653
Compare
Regenerates the published benchmark pages from a full run of
benchmark-reportondevelopat 13bcf93. That run builds against capy 25299eb4, which includes the recycling frame allocator's free list from cppalliance/capy#485. The last refresh was on 2026-11-06 (#413). Only the generated pages and charts change.Source data: run 43810023573, 7 iterations of 2 s on the Linux, Windows and macOS machines.
What moved
Page summaries, corosio against asio callbacks on the same reactor:
The allocator change shows up on the wide
fan_outrows, where many frames of one size are live at once. Comparing each configuration's own medians with the previous run:nested/16concurrent_parents/16fork_join/64Over the same two runs, asio callbacks moved by no more than 1.5% on any
fan_outrow except the macOS 16-parent rows (about 20% variation between iterations), and the narrow rows (fork_join/1,/4) are flat on Linux and macOS. The epoll median moves by less than its noise:fan_outis 18 of the 113 rows, and corosio's own epoll medians are unchanged across the other categories.Testing
The pages were generated with
report_page.pyas described inbench/README.md, and a second generation produced identical output. The local doc lint reports no findings on the benchmark pages, and the include-tag check passes. The Antora site was not built locally; the Documentation workflow on this PR builds it.