Arch Linux

Oh man, this is a rough one... I feel for you, but I don't know a clean/elegant way to handle this.

Let me turn this question sideways a bit. If you had the ability to remove this stuff after install, would that fix things? That is, you'd still "waste" the download, but would automated removal be a happy medium?

Wasting the downloads would be ok (for me personally) in conjunction with some sort of comprehensive install-veto mechanism in pacman. Something I could feed, say, some regexen, and it would make me never see stuff described by them, be it through plain non-installation, or removal. If packed with ability to remove the vetoed objects from -Ql, that would be a reasonable solution.

Thing is, the install is getting a bit bloated lately, and it really shows on ultraportables. Rich locales in /usr/share/locale, a really fat lump in /usr/share/libgweather, the invasion of low-level docs (do you hack XCB or libX11 often enough to keep their apis and examples on your *other* computer..?), ruby index, etc, etc. It all comes down to too much developer documentation and incredible amounts of localization files.

I'm pruning that with cron atm. But i really think it would be worthwhile to investigate a better solution, no matter how unimportant this is on most installs, because other than that Arch is A++ God-given for small installs (too). ;)

Ok, so, what kind of opinions do we have on something like:

/etc/pacman.conf:
NoInstall /usr/share/info/*
NoInstall /usr/share/doc/*

and
$ pacman --noinstall '/etc/profile.d/*' -S foobar

I think functionality like this would solve a lot of our issues, and would give the user more choice and more control. I imagine we could do the matching using filesystem globbing (fnmatch), as that syntax makes more sense here.

This would enable me to, on a minimal system, do something like:
NoInstall *.a
NoInstall /usr/include/*
NoInstall /usr/share/{doc,info}/*

This might actually make sense on the ISOs too (though we want man and info pages)

Or we could just make NoExtract take "*"'s.

Or even get hooks implemented which would allow the user to define a rule like this.

I think (pcre) regexen are rather crucial. There are two good examples of directories you don't want anything installed into *except* some known items: /usr/share/doc and /usr/share/locale. I can't see how you can ever do

DontYouNever "/usr/share/locale/(?!klingon|lojban).+"
DontYouNever "/usr/share/doc/(?!ghc|nhc|uhc|ehc|yhc|hugs).+"

...with globs.

Also, if you take this approach, why not just augment NoExtract?

But yeah, local hooks would rock even more.

I agree, it is appealing on the ISO, it will be frustrating though for people who develop on their systems as they will have to reinstall each and every package in core in order to get the headers ... which again is no big deal if the the ISO is even a bit out of date since the it downloads a big deal of packages anyway at the first pacman -Syu.

That comes down to usability vs download size and bandwidth use on the servers.