IRC channel logs

2021-03-19.log

back to list of logs

<pder>stikonas: for tar, I attempted to build tar 1.22 which is the first version to support the J option for xv. It wants autoconf 2.60
<pder>*xz
<fossy>stikonas[m]: is there a reason we haven't been using autoreconf before now?
<stikonas[m]>fossy: non-perl versions need find
<fossy>stikonas[m]: ahh
<stikonas>fossy: I think autoreconf-2.53 also didn't wor for me
<stikonas>hence I started using it from 2.54
<stikonas>anyway, I've adjusted it to run with -f flag
<stikonas>(and also pushed one commit to add MAKEINFO=true earlier in toe bootstrap)
<stikonas>I already had to add it to autoconf 2.12 and 2.13...
<stikonas>but looks like automake 1.4 also needs it
***jelle is now known as Guest93583
***jelly1 is now known as jelle
<stikonas>pder: I think I might be a be to go a bit further with autoconf without updating libtool... not sure if I can reach 2.65 though
<stikonas>but 2.61 should be possible
<pder>stikonas: 2.61 or later of autoconf would be helpful for building tar 1.22. I was checking out earlier versions of xz, and v5.0.0 wants autoconf 2.64 and libtool 2.2
<stikonas>libtool is tricky
<stikonas>even 1.5 wants cpp
<pder>could tcc -E be used for cpp?
<stikonas>possibly...
<stikonas>I haven't tried...
<stikonas>feel free to try :)
<stikonas>1.4.3 build easily, but not sure if it gains us much
<stikonas>and 1.5 already needs it
<stikonas>also if you feel up to it try building texinfo...
<stikonas>(which has makeinfo binary)
<stikonas>would remove the need for MAKEINFO=true
<pder>I can take a look. What are the missing dependencies for gcc at this point?
<stikonas>I tried autoreconf 2.59, it was missing some macros
<stikonas>maybe newer autoconf
<stikonas>other than that we can try building gcc
<pder>Did you ever try findutils?
<stikonas>hmm, I tried briefly
<pder>Also do we need to rebuild gawk? You said that newer autoconf fixed the bash || true workaround.
<stikonas>I think some problems with autoreconf, maybe wants gettext
<stikonas>possibly patchable in configure.in/ac
<stikonas>pder: rebuilding gawk would still be better
<stikonas>otherwise it can't print proper error messages
<stikonas>in configure scripts
<pder>Do we need a newer gawk? 3.0.4 builds with CC=tcc ./configure --build=i386-unknown-lin
<pder>ux
<stikonas>I guess le'ts just rebuild current one for now
<pder>I am not sure what all needs to be regenerated though
<stikonas>just try running autoreconf-2.xx -f
<stikonas>should work quite well now
<stikonas>well you need my PR...
<stikonas>for non-broken autoreconf
<stikonas>autoreconf-2.52 depends on findutils, so doesn't work
<pder>I am on your autotools branch. autoreconf-2.59 -f seemed to work followed by configure and make
<stikonas>I'll probably push 2.61 later... but it will be easy to rebase...
<stikonas>and if not, you can always stay on 2.61
<stikonas>2.59
<stikonas>we might also replace bash 3.2.57 with something newer...
<stikonas>maybe even 5.1 (which is the very latest)
<pder>That would be nice- bash 4 I know has some nice additions
<stikonas>I think bash has fairly low requirements on autotools...
<stikonas>I think even latest git needs just 2.61
<stikonas>hmm, autoconf 2.63 (or 2.62) both need newer automake, I can try later 1.9.x...
<stikonas>pder: so I've build autoconf 2.61 now
<stikonas>had to build help2man first
<stikonas>and help2man is a tiny bit broken
<stikonas>(missing shebang)
<stikonas>can only run it with perl /after/bin/help2man
<stikonas>instead of just "help2man"
<stikonas>pder: I've just pushed autoconf 2.61 (with fixed help2man too) if you want to play with newer bash...
<stikonas>pder: fossy: so automake 1.9 seems to be another case of circular dependency...
<stikonas>it depends on automake 1.8a (first alpha of 1.9), we don't have it though
<stikonas>I foudn 1.8b on alpha.gnu.org, but that already seems to depend on 1.8a...
<stikonas>maybe should try to patch configure.ac
<stikonas>hmm, yeah, not too hard to patch. Although, it's just another case of let's depend on ourselves for no good reason...
<Hagfish>i almost think that the person responsible for that engineering decision should be named and shamed, but i'll settle for this example being included in a "what not to do" list taught in future computer science undergraduate courses
<Hagfish>let it go down in history like the Therac-25
<stikonas>sell, the fix was sed -i 's/1.8a/1.8.5/; s/ filename-length-max=99//' configure.ac
<stikonas>(removing that filename-length-max, not too sure what exactly it does, I guess limits some filenames somewhere)
<fossy>autotools loves those circular deps lol
<stikonas>anyway, I'm testing 1.9.6 now
<stikonas>autoconf might actually be getting better later
<stikonas>just checked 2.65 only wants 2.60 (although we need to get automake 1.10 first)
<stikonas>2.65 would be nice to have as it is needed for xz
<Hagfish>what's the earliest piece of software where upstream sources require xz?
<Hagfish>(in terms of the bootstrap path, although i'd be interested to know the release date of it too)
<stikonas>Hagfish: almost nothing really requires xz
<stikonas>you can download gzip tarballs
<stikonas>but xz has better compression ration
<stikonas>ratio
<Hagfish>oh, cool
<stikonas>anyway, for older software that we used to build (something around 2004) .tar.xz was not an option, it only becomes an option once we move to newer stuff...
<stikonas>like from 2010
***Noisytoot is now known as NOISYTOOT
<stikonas>(one can probably build xz from older version, but there are no tarballs available anymore)
<stikonas>5.0.0 is the oldest I could find
<Hagfish>yeah, i see it's the oldest version in the NEWS file too https://git.tukaani.org/?p=xz.git;a=blob;f=NEWS;hb=HEAD
<stikonas>oh, maybe they just started releasing from version 5...
<stikonas>anyway, we'll soon have new enough autotools, I'm almost done with them (except for libtool)
***NOISYTOOT is now known as VERYNOISYTOOT
***VERYNOISYTOOT is now known as Noisytoot
***fitzsim` is now known as fitzsim