> > While this makes sense for a single driver, I don't recommend this for > > complete systems like teTeX. People might be suprised if files get > > installed into their old system even though a new prefix was chosen. > > But wouldn't that only be for those directories where --datadir > isn't passed down (as you explain in the paragraph quoted next), > and where no `make install' is done anyway?
No. Assume that someone just downloads teTeX-src and not teTeX-beta tarball. Then, he unpacks teTeX-src and configures for a new prefix. Your kpsewhich "trick" would activate his already existing texmf tree as datadir. I guess that source package should not behave like this. If the user specifies a prefix, then nothing else should go outside (at least if no other options are given that just specify this). > Hmm, without this knowledge, that warning would make me pretty > nervous as well (and I don't find it too reassuring either that > the default location is `NONE/share/texmf' - quite the contrary, > I would suspect there's an error in my configuration.) Yes, I agree. Something like an intruductional text as you suggested or better a way to avoid these warnings should be implemented. > My remark was also about what happens when --prefix is not > specified at all. Shouldn't the default configure be able to > find an existing texmf tree in this case? With RH 7.0/7.2 and I tend to say "no". I prefer the "safety" of not writing into an existing tree unless requested. > Slackware 8.0, where the texmf tree is located by default in > `/usr/share/texmf', I always need to specify --prefix or > --datadir explicitly, otherwise none of the configures (in You can have a config.site file where you can specify system-wide defaults, e.g. datadir=/usr/share or prefix=/usr. Just set CONFIG_SITE or have config.site file in $prefix/share or $prefix/etc. I think that this config.site "feature" of GNU configure scripts is unknown to a large number of people... Thomas
