Reverted the dash symlink
Pointing /bin/sh at dash shaved a few milliseconds off script startup and broke enough AUR build scripts that it was not worth defending. Back to bash. The measurement was real; the tradeoff was bad.
CHANGELOG
Including the reversals. A configuration that only ever moved forward would be a configuration nobody actually ran.
Entries before v0.9.0 predate the current measurement method and are not directly comparable to the figures on the reference page.
Pointing /bin/sh at dash shaved a few milliseconds off script startup and broke enough AUR build scripts that it was not worth defending. Back to bash. The measurement was real; the tradeoff was bad.
Previously the install just skipped creating a swap partition. A linux-firmware update pulled in a unit that created a swap file anyway, which went unnoticed for around three weeks — the machine was quietly paging the entire time and the numbers on this site were wrong for that period.
Now swap.target is masked outright.
Chromium had been failing to start since the compositor was removed. The workaround (--disable-gpu) was in a personal dotfile for months before anyone thought to write it down, which is exactly the kind of thing this reference exists to prevent.
L2 hit rate is consistently above 99% L2 hit rate averages around 96% under the render loop. The original figure came from a microbenchmark that fit entirely in cache and did not represent real use.
Up to this point the framebuffer path was opt-in and the conventional stack was still installed alongside it. Removing the fallback made the configuration meaningfully harder to recover from, which is the point at which this stopped being a tweak and started being a distinct configuration.
Before this the configuration existed as a text file of instructions that were followed by hand, incorrectly, about half the time.