Re: Last minute changes - Review Request

David Stewart

From: yocto-bounces@... [mailto:yocto-
bounces@...] On Behalf Of Richard Purdie
Sent: Friday, October 22, 2010 7:24 AM

Coming up to release there are a few things that the extended testing
has shown up which we have fixes for and which we should consider
including in the release. I also finally got around to doing the final
sstate stress testing and found several problematic issues. Given that
sstate and checksums are a significant feature of this release, I'd
really like them to work as well as we can make them. Prior to this I
had stress tested the backend up not the use of the packages. These
changes don't change any sstate packages themselves, just the use of

Since we already have the release images prepared and tested and these
are not going to change, the criteria for potential changes:

a) We can unit test the changes and be confident they don't
break/regress things.
b) They fix important bugs that the user can easily run into
or that make the project look bad.
c) The changes are small, well documented and are obviously correct
looking at the code/patch.
d) The don't change the generated images.

I'm proposing the following, for each I've provided a rationale:

Printing "fatal:" to new users is worrying

[Reported by Dirk]

Users can easily hit the sanity warning about missing 32 bit libs. We
don't need this functionality at this time so we might as well turn it
off by default

[Reported by Dirk]

Doesn't change the build output but fixes a build issue people can
easily run into in from scratch builds due to a missing dependency.

[Reported by Dexuan]

Removes the "WARNING: Function do_build doesn't exist" message which
could worry users and looks bad

[Reported by Richard]

Several complaints have been received about the inability to easily
delete sstate and DL_DIR contents to recover from failures. This adds a
cleanall task which does this. Downside is that this is undocumented.

[Reported by Darren and others, run into by Dirk]

Fixes an sstate bug where the do_package sstate packages don't install
correctly. A user could hit this if cleaning and rebuilding a package
under certain circumstances.

[Reported by Richard]

Fixes an issue with sstate where rebuilding a package using partially
prebuilt state would break rpm generation and silently remove all the
rpm packages.

[Reported by Richard]

Fixes a typo spotted in the debian packaging backend. We don't use this
by default but it should be fixed.

[Reported by Richard]

User testing keeps showing up issues with the pseudo directory creation
failing to happen. This patch solves the problem in a brute force
fashion, once and forall.

[Reported by Mark and others]

Fixes an sstate issue seen on the Poky mailing list where file:// sstate
mirrors wouldn't work.

[Reported by Gary Thomas (poky list)]

Just changes a log level to make the output from sstate slightly more

[Reported by Richard]

I'm not happy about being in this position and I know Dave will be very
nervous about these late changes. To mitigate this I'd like to propose
that a selection of people (Josh, Mark, Saul?) review these changes and
report back on whether they feel these are appropriate and also give the
build some testing with these applied.
I'm so predictable... :-) Yes, I'm nervous. I looked at all of the patches and with the exception of one or two, they mostly seem like good ones. I will accept these if Josh/Mark/Saul give us a +1 on their review & testing.



yocto mailing list

Join to automatically receive all group messages.