Every so often, I’ll be working on updating a package for Gentoo, and suddenly I’ll:
* QA Notice: One or more compressed files were found in docompress-ed
* directories. Please fix the ebuild not to install compressed files
* (manpages, documentation) when automatic compression is used:
*
* /usr/share/man/man6/warzone2100.6.gz
“What’s the problem?", upstreams cry! They are trying to help us packagers out – it’s one less thing to worry about!
There’s two issues:
1a. You don’t know my system!
Portage (Gentoo’s primary package manager) allows me to choose my own compression
command using the variable PORTAGE_COMPRESS in make.conf(5). As it happens,
right now, it defaults to bzip2.
The vast majority of man pages on my system are therefore compressed with it. This is because most packages don’t bother compressing their man pages - they accept the consensus that it’s for distributions/a user preference. It’s handled by my package manager at the point of installation.
But it doesn’t stop with choice of algorithm. What about compression levels? What
if I choose to use say, pbzip2 instead (notably the same algorithm)?
i.e. Even if an upstream correctly guesses the right tool I’ve been using to be consistent with my other man pages (and I love consistency), they may end up doing it wrong anyway.
1b. You don’t know your users' systems!
-
RPM (Fedora, RHEL, …) recompresses based on preferences (defined in the spec file).
-
Arch Linux (re)compresses all man pages, but allows customisation too.
(Aside: Arch Linux’s wiki actually has a great set of examples of the kind of tweaks you might want to make to allow parallelism.)
-
Debian policy mandates
gzip -9. -
Some tooling has unexpected side effects. gzip, for example, may embed a timestamp within its header which can hamper efforts to create reproducible builds.
-
The Meson build system dropped support for automatic compression, accepting that it’s to be handled on the packaging side, not that of the build system.
2. It’s a micro-optimisation.
Uncompressed, this file is 4.3KB. The compressed (bzip2) variant is 1.8KB, and the compressed (gzip) version is 1.7KB.
I really don’t mind consuming a few extra KB if I’m installing several 100MBs of assets for a game. But even if I did, I wish they’d leave it to me.
If nothing else, my message is this: I can promise upstreams don’t want PRs for their project to
switch from gzip to zstd or whatever the next hot compression algorithm
is! The only winning move is not to play.