| Summary: | Rootfs creation broken when using package_deb as primary PACKAGE_CLASSES with non Debian-OS build system | ||||||
|---|---|---|---|---|---|---|---|
| Product: | [Build System, Metadata & Runtime] Meta-yocto | Reporter: | Joshua Lock - Disabled <josh> | ||||
| Component: | meta-yocto | Assignee: | Paul Eggleton <bluelightning> | ||||
| Status: | RESOLVED FIXED | QA Contact: | |||||
| Severity: | normal | ||||||
| Priority: | Medium | CC: | jessica.zhang, poky.bs.watcher, poky.watcher, richard.purdie, scott.a.garman, sgw | ||||
| Version: | 1.2 | ||||||
| Target Milestone: | 1.2.1 | ||||||
| Hardware: | x86 | ||||||
| OS: | Multiple | ||||||
| URL: | [1.2] | ||||||
| Whiteboard: | (P2)(1.2) Fix posted and queued in denzil-next | ||||||
| OS type for building Yocto: | --- | Type of Regression: | --- | ||||
| Verified: | Documentation change: | --- | |||||
| Bug Depends on: | |||||||
| Bug Blocks: | 1881 | ||||||
| Attachments: |
|
||||||
|
Description
Joshua Lock - Disabled
2011-12-21 22:03:12 UTC
I've created a patch to correctly install the apt.conf.sample for use by apt-native, there are outstanding issues with rootfs creation still. I'm working through these. The package_deb rootfs construction works fine on an Ubuntu 11.10 test machine, but suffers all sorts of failures on a Fedora 16 system. Work is ongoing. Fix merged to master in revision 91a77020eb175000847e493ec30196aba9fb11f2. Created attachment 487 [details]
log.do_rootfs containing dpkg errors
This is still broken for me on Fedora 16 x86_64, see the attached log.
From the above log: package architecture (i386) does not match system () Seems there were multiple problems here; for whatever reason I did not get this on the older Fedora system I was testing on. The above warning is not significant; it appears on a working system as well. The heart of the problem is that zlib is failing to decompress the data.tar.gz archive within the deb; however it doesn't seem to be a problem with the way either zlib or dpkg-deb are built - I copied both of these over from a known working system and they still failed. Nor does it appear to be a problem with the construction of the deb - a deb copied from the "broken" system works fine on a known working system, and additionally I was able to extract the deb by hand (using ar and tar/gunzip) without trouble on the "broken" system. I tracked this down to an issue where fflush() was being used on an input stream, the behaviour of which is not defined by POSIX and seems to have changed in a recent version of glibc. Luckily once identified it was just a matter of backporting a fix from upstream. I've sent a patch for this which is in the denzil-next branch already: http://lists.linuxtogo.org/pipermail/openembedded-core/2012-April/021442.html I suspect we should update dpkg to 1.6.x soon but that's something for 1.3. |