| Age | Commit message (Collapse) | Author | Files |
|---|
|
- SOURCE_MIRROR_URL is new variable which point to source mirror which will be
used before fetching from original SRC_URI location.
|
|
of using comments
The limit of comments is too low. Use attachments to work around this
issue. Also start using HTTP post to avoid really long URLs. They lead
to a pipe error.
To post we use a MultiPartPoster which is licensed LGPL and can be freely
downloaded. You need to put it into the PYTHONPATH
Add a new variable for creating attachments and document it.
|
|
rsplit was introduced in python 2.4 and is like split but starts to
split from the right side. The only difference is obviously if you
have maxsplit set (which we don't). So it is fine to remove the 'r'
|
|
rewrite in 1.9)
|
|
This class can find the broken bug, reopen it, or leave it open.
There is some issue with sending the logfile to the server, the
url itself is correct, when opened in a browser the right thing
happens but not from within python *sigh* need a working tcpdump
to look into this (which means I will continue tomorrow)
|
|
|
|
It is time to try it
|
|
Posting bugs should work (besides typos), now only reopening bugs
is missing.
|
|
and 'f02f7da26cf1279907a41d7957e054f42575adbd'
|
|
want to log in at all.
|
|
bugzilla directly
Start adding seppuku. It listens to TaskFailed and NoProvider failures
logs into a bugzilla and calls empty stubs to find a bug report.
We use the urllib2 and cookielib from python2.4 to handle the session
cookie for us automatically. This library is sadly not present in python2.3
but it eases my life.
The old BugZilla installed at OpenMoko has no support for listing the
bug report as csv or xml file, this will force me to parse the HTML which
I will do for now. This assumes that only one bug report with the exact
name can exist.
|
|
on populate staging for do_rootfs (both from poky)
|
|
integer/floating point values for <=.
|
|
* Fixes insane.bbclass QA errors.
|
|
them exist
Do some cleanup on the supported arches
|
|
"inherit package_ipk package_tar" and seems to interfere with do_split_packages
|
|
and '922cf0f28b50dfb43c18e0654db44c32b7d09bc4'
|
|
* Closes #1909.
|
|
|
|
and '826367bbd68c19eb406efdb0f5a9c5276fbf3ca8'
|
|
sed out /usr/include references
|
|
|
|
add bb.fatal
udev has a link to host binaries, do not check the architecture
of these binaries. We will skip all links for now for the two
reasons:
-We either point to a file in our package and then we will
check it anyway
-It is a file for a another package, which will be checked
anyway
Make architecture mismatch a fatal/insane error and return False.
This should fix the issue koen has seen when he added a bb.fatal
the make_fatal_error method returns true if the error should be
made fatal, in this case our checks need to set sane to False.
The return not makes sure this is happening.
|
|
arguments for format string by removing a spurious '%s'
|
|
Added extra entry into dict to mark 32bit/64bit and use it for manipulating
ELF headers.
Some architectures would need checking (not OE ones).
|
|
|
|
|
|
|
|
The idea of insane.bbclass is to report all issues found at
once. This is why the checks return True if no error was found
and False if at least one was found.
At the end of all check runs we will see if at least one check
failed and call bb.fatal
|
|
so do bb.fatal instead if bb.error
symptom:
ERROR: Architecture did not match (20 to 3) on /work/ppc603e-angstrom-linux/udev-092-r15/install/udev/sbin/scsi_id
This means OE detected type 3 (i386) while it expected type 20 (powerpc), most likely caused by some fool hardcoding 'gcc' as compiler
|
|
package-index use it
|
|
per architecture subdirs of deploy/ipk
see: http://lists.linuxtogo.org/pipermail/openembedded-devel/2007-February/001462.html
|
|
|
|
|
|
native ones:
.la files should have installed=yes (at least I believe
this). And somehow some believe this as well, some don't though...
some like fakeroot love to install two .la files...
non-native:
.la should have installed=no
both:
should not point to tmp/work. libgettext and libltdl(libtool)
fails this...
happy fixing and arguing
|
|
r1258,r1195)
The .so check now looks for symlinks ending with .so. These are a typical
trait of -dev packages and kills a lot of false positives.
This was authored by ross and was taken from the poky svn repository.
|
|
|
|
|
|
* change "," to "%" so python doesn't die if an error occurs
* spell architecture correctly
* add arm/eabi tupple that works for stripped binaries, but fails for -dbg binaries (abiversion 0 vs 97)
|
|
Debian packages
|
|
x86_64 can not be detected properly with this code - needed change:
- ELFFile.my_assert(self.data[ELFFile.EI_CLASS], chr(ELFFile.ELFCLASS32)) # only 32 bits
+ ELFFile.my_assert(self.data[ELFFile.EI_CLASS], chr(ELFFile.ELFCLASS64))
|
|
|
|
Add a small ELF parser and read OSABI, ABIVERSION, ENDIAN and MACHINE. This
code compares TARGET_OS and TARGET_ARCH to a list of known versions. Currently
we have values for linux and linux-uclibc for arm, armeb and powerpc. mips,
i386, amd64, ia64 are missing. ABI wise ARM EABI is missing as well, but koen
is likely to enter the values into the documented dict
|
|
Add a method to implement koen's skipping of
errors of type .so in non dev package. This
will turn into a list of known and acked issues which
we decided to ignore. I do not yet know how to maintain
such a list properly though
Split the .so in non-dev package and the .debug in non-dbg
package into two different packages
|
|
-Print all errors of the built packages and if we have printed
any errors fail
-Log the errors to a log file, so we can see them afterwards
-Print the path (- the TMPDIR) of the package in question
|
|
errors
-Inspired by doku's work on mpd's buildsystem grep through
the config.log and find broken autotools tests
-Make errors fatal so we will notice these things more quickly.
|
|
some packages
|
|
and 'bf8b52c7c7ffdfe7cb88423cc23d4eeb9c573c2f'
|
|
Welte
|
|
nslu2, nas100d and dsmg600 image building out into classes so they can be reused with multiple distros (like SlugOS and Angstrom)
|
