automake 1.5 - statistics invade
I wrote some scripts to try to build packages in unstable
with automake v1.4 and v1.5. There are sources of error:
if I failed to build a package, it might be because it
failed to download properly.
Summary:
227 source packages from sid were examined
97 packages that build-depend on automake built fine,
but did not use automake.
71 packages that built with version 1.4, failed with 1.5
20 of these built when I made automake 1.5 default to
--add-missing; perhaps only ./depcomp is required.
28 packages that build-depend on automake were able to
use 1.5 successfully, i.e. automake was run from within
the build directory.
20 packages failed to build for me on both; this may
be an artifact of a partly woody, partly sid system.
8 packages were ignored (it takes too long to build
gwydion-dylan, for example)
Caveat:
If you're responsible for, or especially interested in,
a specific package, please verify the build failure for
that package. Failure is defined as: 'sudo apt-get -y
build-dep package ; fakeroot apt-get --build source
package' returned status 100. This can happen for
reasons other than a package/automake bug.
Details:
The broken 71: (I may look into these further. shadow and
libnss-db are notable as they are required/important):
aime, alsa-lib, alsa-lib-0.5, alsa-utils, alsa-utils-0.4,
alsa-utils-0.5, aterm, bitcollider, black-box,
brahms, bumprace, crimson, cyrus-sasl-mit, dcethreads,
ecasound, ext2resize, fsviewer, g2s, gb, geomview, glide,
gnome-libs, intuitively, kaptain, kbear, kcpuload, kdbg,
kdebase-crypto, kdesdk, kdestudio, kdesupport, kdetoys,
kdevelop, keuklid, kimberlite, kleandisk, knetfilter,
knetload, koffice, konverse, koth, kprof, kweather,
lesstif1-1, libctk, libnss-db, libprintsys, libproplist,
licq, mangoquest, micq, modplugxmms, nail, newsx, odb,
openjade, orbitcpp, quanta, rpm, sabre, shadow, siag,
sppc, tcpstat, tilp, tk707, upx-ucl, xcircuit, xkbsel,
xmms-goom, yank,
The 20 that seem easily fixed:
alsa-lib-0.5, alsa-utils, alsa-utils-0.4, alsa-utils-0.5,
aterm, bitcollider, crimson, cyrus-sasl, dcethreads,
ecasound, g2s, kdevelop, keuklid, modplugxmms, openjade,
siag, sppc, sqlxx, tilp, yank,
The mysteriously successful 28:
coriander, cppopt, discover, dvgrab, gpr, hfsplus,
kdegraphics, kdelibs, kdemultimedia, kdenetwork,
kdepim, kdeutils, kfilereplace, leakbug, lib3ds,
libgnurdf, libppd, libsmi, liece, lletters, metamail,
opensp, orbit-mt, penguineyes, printerconf, snmpkit,
syslog-ng, wmaker,
The unnecessarily(?) build-depending on automake (though
don't take my word for it):
abuse-sdl, acct, allegro, alsa-lib-0.4, amcl, apcupsd,
blinkd, cdindex-client, codebreaker, curl, curl-ssl,
dact, dfm, extace, fbgetty, fuzz, fvwm, gfcc, gfslicer,
gidic, giram, gkdial, gltt, gmanedit, gql, grun, gsl,
gsmlib, gsumi, gtetrinet, gtkrecover , gtktiemu, hamlib,
ht, icqlib, id3lib, jags, jikes, kdeadmin, kdebindings,
kdoc, kmago, knocker, korelib, krecord, lbreakout,
libarr, libgphoto2port, libnss-ldap, libnss-lwres,
libpam-ldap, libpam-sfs, linneighborhood, marbles,
mtr, mysql, nano, nemesis, ntp, oaf, openafs, openal,
oregano, parted, penguin-command, pidentd, pidentd-des,
plib, plib-doc, plib1.3, plib1.3-doc, plib1.5,
plib1.5-doc, prboom, quantlib, rmc, run, sdl-net1.2,
ser2net, squidguard, stools, stringlist, strutilsxx,
tar, teg, textutils, tinc, traceroute, tree-puzzle,
wsoundserver, xcall, xconvers, xdialog, xlog, xmmsarts,
xmorph, xwrits,
Those I couldn't seem to build automatically using either
version:
alsaplayer, bibletime, detect, expat, glide3-alpha,
gnocatan, gnumeric, gpmudmon-applet, gtkmm, kde-i18n,
kdebase, kdelibs-crypto, libmng, libssrp, lirc,
plotutils, sfs, sqlxx, terraform, tipptrainer,
Gory Details:
I didn't recategorize cyrus-sasl, db2, and db3, which
worked fine in an earlier analysis.
I modified autoconf to print `pwd` and @ARGV to a logfile
to determine whether automake was invoked. If it was
called with --version (as configure calls to check if
it's there) that is not counted.
The systems were mostly identical, tracking woody until
told to update build-dependencies from sid. Hidden
version dependencies may be to blame for some failures.
Versions are not necessarily always the same; I uses a
proxy cache, not a mirror, for these tests.
Full logs are at:
http://www.cs.washington.edu/homes/nspring/debian/automake-1.4
http://www.cs.washington.edu/homes/nspring/debian/automake-1.5
Sorry 'bout the content encoding; you may have to
save to disk and zless.
The change to default to --add-missing is easy (around
line 340 of /usr/bin/automake), but didn't turn out to
be the panacea I thought it would be.
Issues (or, If You've Read This Far, You Must Really
Be Interested):
Bug reports seem appropriate for those 20 packages
that do not build from source. If you're ambitious,
please make sure it fails for you before doing so.
I'm not certain my systems are pristine.
Is it possible to make sure the 97 that didn't seem to
need automake would be fine without it? Does it matter
if the extra build-depends is there?
-neil
Reply to: