Watching Arm failures

Information Source:

Arm buildd

This list of uninstallable packages can be useful for prioritisation too: http://ftp-master.debian.org/~aba/test-issues/current

Feel free to:

Significant issues

lost packages

Packages are sometimes built, but never uploaded. This is often due to lost mails:

java problems

We do now have some working java as jikes and sablevm-classlib have built. I'm not sure how much of the backlog that will enable to build. Or if it all actually works properly. Fixing other java tools remains a high priority.

This is our most important bug - please try to fix it ASAP

{{{Setting up gij-4.1 (4.1.1-13) ... /var/lib/dpkg/info/gij-4.1.postinst: line 20: 19362 Segmentation fault gcj-dbtool-4.1 -n /var/lib/gcj-4.1/classmap.db dpkg: error processing gij-4.1 (--configure):}}} This is preventing all sort of things building. More info can be found at http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=388505 and upstream: http://gcc.gnu.org/PR29206

This needs someone who understands something about java, gcc and arm to fix. This patch http://nevyn.them.org/~drow/fix-arm-gcj.patch (thanx to Dan jacobwitz) improves matters significantly, but it still doesn't seem to work or build properly.

A worked-around was put in gij-4.1 by simply not running gcj-dbtool, but then it pops up in ecj-bootstrap instead. (http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=394396). Another workaround is given in that bugreport, but apparently it just breaks the package.

{{{/usr/bin/ecj -source 1.4 -encoding UTF-8 -warn:-deprecation,serial,unused -proceedOnError -bootclasspath -classpath /build/buildd/kaffe-1.1.7/build-tree/kaffe-1.1.7/libraries/javalib/vmspecific:/build/buildd/kaffe-1.1.7/build-tree/kaffe-1.1.7/libraries/javalib/external/classpath:/build/buildd/kaffe-1.1.7/build-tree/kaffe-1.1.7/libraries/javalib/external/classpath/external/w3c_dom:/build/buildd/kaffe-1.1.7/build-tree/kaffe-1.1.7/libraries/javalib/external/classpath/external/sax:/build/buildd/kaffe-1.1.7/build-tree/kaffe-1.1.7/libraries/javalib/external/classpath/external/relaxngDatatype:.: -d . @classes Exception in thread "main" java.lang.NoClassDefFoundError: org.eclipse.jdt.core.compiler.IProblem

make[4]: *** [compile-classes] Error 1 make[4]: Leaving directory `/build/buildd/kaffe-1.1.7/build/jthreads/libraries/javalib/external/classpath/lib'}}}

Java dependencies

A lot of java stuff is broken - some due to gij not installing some due to kaffe not building. We need to unbung this logjam. gij4-1 segfaults on setup (See above), but it has been worked around

need gij4.1 fixing:

  • libgnujaxp-java - throws SAXException during build
  • ow-util-ant-tasks - waiting for kaffe-jthreads
  • jmagick - waiting for kaffe-jthreads
  • libglade-java - waiting for libgtk-jni
  • antlr- waiting for kaffe-jthreads
  • cairo-java - waiting for liglib-java
  • classlib
  • eclipse
  • libglade-java (libgtk-jni)
  • gcjdoc
  • libgconf-java (dep-wait)

mono

Mono can only build for v4 or later. see #394418. This seems fair enough. It seems that mono 1.2.1 or patches to earlier versions are needed to make it build.

  • galago-sharp

== uninstallable

Curently: (2006-11-06)

quodlibet 0.23.1-1: quodlibet-ext uninstallable - needs python-ctypes from ctypes packages which is not built for arm due to lack of ffi_closure support in gcc. see #397653

gcc4 ICEs

unclassified ICE's

These need proper bugreports generating and filing upstream. Comparing against cvs-snapshot on an unstable system if you can.

Normal issues

BinNMU needed

(packages which have built but didn't build properly for whatever reason (wrong versions of libraries, broken compilers)

builds timing out

These may be due to a genuine problem, or may simply be that the buildd really is taking a very long time (common with building large C++ files and linking a lot of files at once) and would actually complete eventually.

  • === timeout needs increasing ===
  • iceweasel/firefox - confirmed timeout === genuine hangs/problems === Listed below under 'bugs filed'

failed - bugs not yet filed

failed - bugs filed

  • php4-vpopmail - #309558
  • libooc-vo #355140. Broken on all arches.
  • kudzu #287244 - scheduled for complete removal
  • crystalspace #358545. last build killed due to inactivity.
  • open21xx - #379159 fails on arm, s390, powerpc
  • prc-tools - #348273
  • gnome-presence-applet - FTBFS on all arches. Not sure how it worked before. #389292
  • harbour - gives -- Invalid opcode 172 in hb_compPCodeEval() --- errors, similar #276962 on amd64
  • monotone #384565 - mtn command built then used by build hangs
  • tightvnc - #464797

depwait

currently (late nov 2006) 32 of these - mostly waiting on java, which needs attention.

Some others:

  • beagle waiting for libevolution2.0-cil

needs retrying

  • mixxx - libsndfile has now been binNMUed

The haskell compiler ghc6 wasn't present when all these tried to build. It is now in the archive, so this lot should work.

  • hpodder
  • muine_0.8.5-1.1

Wrong dep-wait:

  • libgconf-java (should be fixed)

Fucked build daemon (netwinder):

  • gnome-session_2.14.3-4

needs urgent research - regressions

  • vegastrike
  • pnet - #394764
  • advi - assembler problem
  • libprelude

needs research - never worked

  • basilisk2 - uh.. some gruesome pthread-related sigaction() problem
  • cduce - might be easy
  • atlc - just increase timeout @ buildd ?
  • pose
  • qemu
  • sope
  • mcvs
  • numactl (for arm?)
  • dfsbuild
  • mingw32
  • util-vserver (is the matching kernel patch/kernel image on arm? if not, this is not useful?)
  • imview
  • atlas3 - #243447 . Causes blitz++ to be uninstallable.
  • ctypes - #397653. Current not built for arm because ffi_closure support not present in libffi in gcc - needs adding. Causes quodlibet to be uninstallable.

buildd problems

  • libembperl-perl - #337351 fails test after building: Performing httpd syntax check 1, but builds OK in local pbuild. Fails when pbuilder user is not in /etc/group - which is true on some buildds. Now built OK on another buildd, but some buildds may this fixing?
  • perl Fixed by lowering optimization level, maintainer wants to fix properly, contact perl maintainer if you have an arm host where you can give him a host to log in. Due to some buildds having kernel configured without extended precision: 'CONFIG_FPE_NWFPE_XP'. buildds are having new kernels with this option installed.
  • Some packages need to be built on a machine with a lot of RAM (512MB or more):
    • kdelibs
    • octave

not-for-us

These packages are not for arch=arm so are not expected to build. They still appear on the buildd status pages as failures.

  • grub2
  • defrag
  • gnat-4.1
  • libflorist
  • xmms-openspc

ArmBuilddWatch (last edited 2009-03-16 03:33:03 by localhost)