Bugzilla Bug 7467

Can't emerge abiword-1.0.2-r1.ebuild, on two different boxes (P3 laptop and
Athlon XP desktop), both with Gnome 2 and gcc 2.95.3.  All the prereqs install
fine.  I've tried turning off all custom USE settings and cranking down
optimizations to the minimum, to no avail.

The one red flag in the error messages is what looks like an absolute path being
mistakenly appended to the end of a relative path.  Says to ignore that
non-fatal error, but then a fatal error with the same ac-helpers dir immediately
follows.  The only reference to ac-helpers in the ebuild is to clear invalid
symlinks.

Full dump of emerge output follows:

# emerge abiword
Calculating dependencies ...done!
>>> emerge app-office/abiword-1.0.2-r1 to /
>>> md5 ;-) abiword-1.0.2.tar.gz
>>> Unpacking source...
>>> Unpacking abiword-1.0.2.tar.gz
perl
>>> Source unpacked.
perl
spell
xml2
jpeg
automake (GNU automake) 1.5
cp: cannot stat
`ac-helpers/fallback//var/tmp/portage/abiword-1.0.2-r1/work/abiword-1.0.2/abi/ac-helpers/fallback':
No such file or directory

*************************************************
* Ignore above ERROR as it does not cause build *
* to fail.                                      *
*************************************************

configure: error: cannot find install-sh or install.sh in ac-helpers ./ac-helpers

!!! ERROR: The ebuild did not complete successfully.
!!! Function src_compile, Line -123, Exitcode 1
!!! (no error message)

!!! emerge aborting on  /usr/portage/app-office/abiword/abiword-1.0.2-r1.ebuild .

is this still a problem ?

Still a problem.

al dripton # emerge app-office/abiword
Calculating dependencies ...done!
>>> emerge app-office/abiword-1.0.2-r1 to /
>>> md5 ;-) abiword-1.0.2.tar.gz
>>> Unpacking source...
>>> Unpacking abiword-1.0.2.tar.gz
perl
>>> Source unpacked.
gnome
perl
spell
xml2
jpeg
automake (GNU automake) 1.5
cp: cannot stat
`ac-helpers/fallback//var/tmp/portage/abiword-1.0.2-r1/work/abiword-1.0.2/abi/ac-helpers/fallback':
No such file or directory

*************************************************
* Ignore above ERROR as it does not cause build *
* to fail.                                      *
*************************************************

configure: error: cannot find install-sh or install.sh in ac-helpers ./ac-helpers

!!! ERROR: The ebuild did not complete successfully.
!!! Function econf, Line 9, Exitcode 1
!!! econf failed

Well i can see what goes wrong, but i'm not sure as to why. The line after
after
'cp: cannot stat' should read just ac-helpers/fallback/, this suggests a
problem
with automake. Can you re-emerge automake and try again ?

Re-emerging automake did not help.  Same error.

Can you clean out the temporary build directory (usually /var/tmp/portage ) and
try again ?

Clearing out /var/tmp/portage didn't help.  "emerge abiword" still fails with
the same error.

what shell are you using, its very hard to figure stuff out from here. but you
can see for yourself where it goes wrong, take the abiword source file
(abiword-1.0.2.tar.gz) and unpack it somewhere. then go into the unpacked dir
abiword-1.0.2/abi/ . See the autogen.sh file, the for loop from line 25 to 33
doesn't work for you for some reason. I have honestly no clue why. The $script
var gets wrongly initialized at some point. 

This looks really a problem with your machine settings somewhere, a badly
installed package or something. I hope with this handle you can figure things
out a bit for yourself...

$script should anyway never be
'/var/tmp/portage/abiword-1.0.2-r1/work/abiword-1.0.2/abi/ac-helpers/fallback' .

The problem was programmable bash completion.  Took the bash completion setup
line out of root's .bashrc, and now abiword emerges successfully.

I have the default gentoo bash-completion setup, currently version 20021013. 
(But note that this problem also happened with previous versions.)

bash-completion has been blocked for the new abiword build (1.0.3). Thanks for
your help.

Created an attachment (id=6881) [edit]
Better way to fix the bash-completion crash

The current fix is not very good.  Many people can have bash-completion
installed on the system, while not running it as root (they probably
shouldn't).

This patch checks if bash-completion has been run and aborts compilation if so.
 The ideal patch would check for the exact bash completion setting causing the
problem I guess.

The patch is against 1.0.4_pre1-r2 btw.

I guess I'm not allowed to change the status of the bug? 

Eivind, I think your patch is reversed: it adds the block on the package and
remove the test on BASH_COMPLETION.

So is this bug resolved by blocking bash-completion?

yep

you can reinstall it afterwards if you like, it's only a possible compile-time problem.

I've noticed the bash-completion-blocks-abiword thing several times and always
wondered why.  Having read this, I guess I know.  But surely the solution is to
fix bash-completion so that it doesn't erroneously complete things to cause
these errors (surely it should only complete on tabs, anyway, and not during a
script?).  It seems odd that one cannot have both bash-completion enabled, and
compile abiword.  With the new "userpriv" feature, this will become more of a
problem, as normal users can and often will have bash-completion enabled, and
want to compile abiword.

On another note, I have also noticed that using "emerge -u abiword" does not
complain about bash-completion at all.  That is, "emerge abiword" is bleak
about bash-completion, but "emerge -u abiword" isn't.  Surely a "-u" option to
emerge should not change the blocking?

blocking bash-completion is an insane fix, and very confusing to users, i see
people asking how something like bash-completion can block a gtk+ word
processor on irc regularly. 

Better fix might be, change the pkg_postinst() messages from

pkg_postinst() {
        einfo "Add the following line to your ~/.bashrc to"
        einfo "activate completion support in your bash:"
        einfo "[ -f /etc/profile.d/bash-completion ] && source
/etc/profile.d/ba
sh-completion"
}

to 
...
einfo "[ -f /etc/profile.d/bash-completion -a -t 1 ] && source
/etc/profile.d/ba
sh-completion"
...

bash completion will then only be sourced if STDIN is connected to a terminal.
problem solved.

ack, s/STDIN/STDOUT/

i don't know bash completion all that well and i'm not about to figure it out
for some silly totally weird bugs. If that works then it is still not our
problem to fix. Actually i don't think the bash completion block is a big deal,
it's all in the ChangeLog.