On Mon, 2009-10-26 at 15:55 +0100, Frans Pop wrote: > Thanks for making an effort to look into this. > > On Monday 26 October 2009, Ian Campbell wrote: > > Thanks. I still cannot reproduce so I am somewhat clutching at straws. I > > compared the logs for a local (successful) i386 run of the daily-build > > script vs. the logs from > > http://people.debian.org/~joeyh/d-i/build-logs.html > > In the past I've always been able to reproduce mklibs errors. It does seem like the sort of thing which ought to fail fairly reliably if it fails. > > I've been concentrating on the netboot-gtk log and the differences are > > really rather small up until the failure point. The output from apt > > and/or dpkg differs a little -- I wonder if the build environment is not > > completely up to date and has a buggy (or just slightly differing in > > behaviour) version of one or the other? Seems unlikely to have effected > > so many build servers though (seemingly admin'd by several different > > people). FWIW I'm running apt 0.7.24 and dpkg 1.15.4.1. > > If you cannot reproduce it with an up-to-date sid build environment then > it's quite possible that the buildds are simply outdated. > apt and dpkg are not really relevant here. More relevant are versions of > mklibs itself and packages like binutils. I tried downgrading binutils, as you suggested in your next mail, to 2.19.91.20091006-1 from testing and still saw no issues. The issue seems to have started between http://people.debian.org/~joeyh/d-i/images/20090930-19:15 and http://people.debian.org/~joeyh/d-i/images/20091024-21:59/ which is unfortunately a rather large range. Apart from the version currently in testing binutils seems to have had one other version uploaded in that period (2.19.91.20091003-1) but in the absence of snapshot.debian.net I don't think I can get hold of it. mklibs doesn't seem to have changed since early August which seems to rule it out. > > The symbol _nss_files_parse_sgent is provided by /lib/libc.so.6 and > > required by /lib/libnss_files.so.2. Both libraries are part of libc. > > Does libc.so.6 provide the exact symbol looked for (I'd expect it does)? > In that case this looks like an mklibs/binutils issue. I believe it does: $ objdump -T tmp/netboot-gtk/tree/lib/libc.so.6 | grep nss_files_parse 00056a30 g DF .text 000002ed GLIBC_PRIVATE _nss_files_parse_pwent 0007c880 g DF .text 000003cf GLIBC_PRIVATE _nss_files_parse_sgent 00055900 g DF .text 0000030e GLIBC_PRIVATE _nss_files_parse_grent 0007c3f0 g DF .text 000003d3 GLIBC_PRIVATE _nss_files_parse_spent > > Comparing a manual "make build_netboot-gtk" with the daily-build logs > > (either my local ones or the build server ones), I see this which seems > > odd: > [...] > > That is indeed strange and should be looked into, but I doubt it is related > to the mklibs failure as neither libc6 nor libnss-files are "extraudebs". Originally I wondered if one of the extra udebs was introducing an extra dependency on the symbol which I wasn't finding because the extraudebs dir is deleted after it is used. This turned out not to be the case though. > > The current version of > > http://people.debian.org/~joeyh/d-i/build-logs.html seems to suggest > > even bigger issues though, every arch "failed to download summary log" > > Probably just a temporary network problem somewhere. Indeed, it seems to have resolved itself now. Thanks, Ian. -- Ian Campbell 'Scuse me, while I kiss the sky! -- Robert James Marshall (Jimi) Hendrix
Attachment:
signature.asc
Description: This is a digitally signed message part