Patchwork [RFC,0/1] packagehistory improvement

login
register
mail settings
Submitter Paul Eggleton
Date Nov. 22, 2011, 2:52 p.m.
Message ID <cover.1321972798.git.paul.eggleton@linux.intel.com>
Download mbox
Permalink /patch/15245/
State Accepted
Headers show

Pull-request

git://git.openembedded.org/openembedded-core-contrib paule/packagehistory

Comments

Paul Eggleton - Nov. 22, 2011, 2:52 p.m.
This is the initial implementation of the improvement for
packagehistory.bbclass. This bbclass is intended to be used to help
track changes to package output over time and hopefully catch unusual
situations such as a package doubling in size. You can try it out by
adding INHERIT += "packagehistory" to your local.conf, building
something that performs do_package and then looking in TMPDIR/pkghistory
to view the output.

I've tried to keep it fairly simple - i.e. no new storage backend, just
plain files, although I anticipate most people using it will want to
store the output in a git repository. There's potential for recording
more per-recipe / per-package information, if there are suggestions I'd
be keen to hear them.


Please review the following changes for suitability for inclusion. If you have
any objections or suggestions for improvement, please respond to the patches. If
you agree with the changes, please provide your Acked-by.

The following changes since commit 5f1dc1991f97037692193572298eb7f7a5640760:

  staging.bbclass: Add BB_SETSCENE_VERIFY_FUNCTION function (2011-11-22 13:04:31 +0000)

are available in the git repository at:
  git://git.openembedded.org/openembedded-core-contrib paule/packagehistory
  http://cgit.openembedded.org/cgit.cgi/openembedded-core-contrib/log/?h=paule/packagehistory

Paul Eggleton (1):
  classes/packagehistory: fix and extend

 meta/classes/packagehistory.bbclass |  226 +++++++++++++++++++++++++++++------
 1 files changed, 189 insertions(+), 37 deletions(-)
Koen Kooi - Nov. 22, 2011, 5:25 p.m.
Op 22 nov. 2011, om 15:52 heeft Paul Eggleton het volgende geschreven:

> This is the initial implementation of the improvement for
> packagehistory.bbclass. This bbclass is intended to be used to help
> track changes to package output over time and hopefully catch unusual
> situations such as a package doubling in size. You can try it out by
> adding INHERIT += "packagehistory" to your local.conf, building
> something that performs do_package and then looking in TMPDIR/pkghistory
> to view the output.
> 
> I've tried to keep it fairly simple - i.e. no new storage backend, just
> plain files, although I anticipate most people using it will want to
> store the output in a git repository. There's potential for recording
> more per-recipe / per-package information, if there are suggestions I'd
> be keen to hear them.
> 
> 
> Please review the following changes for suitability for inclusion. If you have
> any objections or suggestions for improvement, please respond to the patches.

I have a question instead:

Does it still suffer from the bug where it will break if you change machine and something goes backward in PV (e.g. gcc 4.6 is used on arm, but 4.5 on ppc)?

regards,

Koen
Paul Eggleton - Nov. 22, 2011, 5:32 p.m.
On Tuesday 22 November 2011 18:25:25 Koen Kooi wrote:
> Does it still suffer from the bug where it will break if you change machine
> and something goes backward in PV (e.g. gcc 4.6 is used on arm, but 4.5 on
> ppc)?

If the package in question is not tied to the machine or the arch that was 
changed, yes it will still cause do_package to fail. I'd be happy to address 
this, but my question would be, how do we tell the difference between a 
legitimate version pinning and when something has gone wrong?

Cheers,
Paul
Paul Eggleton - Nov. 23, 2011, 1:24 p.m.
On Tuesday 22 November 2011 18:25:25 Koen Kooi wrote:
> Does it still suffer from the bug where it will break if you change machine
> and something goes backward in PV (e.g. gcc 4.6 is used on arm, but 4.5 on
> ppc)?

So unless I'm mistaken the fix that you yourself applied to OE in 2009 (and 
which was included in Poky, and hence OE-core) already works around this 
specific situation, since the package history is always stored under a 
BASEPKG_TARGET_SYS directory (or MULTIMACH_TARGET_SYS now with my patch since 
the former is no longer available). Am I missing something?

Cheers,
Paul
Koen Kooi - Nov. 23, 2011, 1:31 p.m.
Op 23 nov. 2011, om 14:24 heeft Paul Eggleton het volgende geschreven:

> On Tuesday 22 November 2011 18:25:25 Koen Kooi wrote:
>> Does it still suffer from the bug where it will break if you change machine
>> and something goes backward in PV (e.g. gcc 4.6 is used on arm, but 4.5 on
>> ppc)?
> 
> So unless I'm mistaken the fix that you yourself applied to OE in 2009 (and 
> which was included in Poky, and hence OE-core) already works around this 
> specific situation, since the package history is always stored under a 
> BASEPKG_TARGET_SYS directory (or MULTIMACH_TARGET_SYS now with my patch since 
> the former is no longer available). Am I missing something?

I'm trying hard to remember why I disabled it last year, I was convinced it was the TARGET_SYS thing, but apparently not. Good to know that I fixed it already :) I'll ask my coworker when he gets back, but I vaguely remember something about it acting funny with nostamp making consecutive runs of 'bitbake some-image' really slow.

regards,

Koen