meta-toolchain continues to woe

meta-toolchain continues to woe

[Holger Hans Peter Freyther]

Hi Guys,

I'm really getting frustrated by this crap. Every week I try to build meta-
toolchain and every week it is broken in another way. Not even talking about 
the runtime breakage I'm seeing and would like to fix...

So I either have a problem with the variable changes for the layout dir or in 
the opkg upgrade that happened.

any ideas?

	z.


The error:

You don't need to build everything, it will error out early..

bitbake -c populate_sdk -b recipes/meta/meta-toolchain.bb

./run.do_populate_sdk.27727 
Collected errors:
 * Could not obtain administrative lock for offline root (ERR: Bad file 
descriptor)  at /data/openembedded/build/arm/cortex-a8/tmp/work/x86_64-armv7a-
sdk-angstrom-linux-gnueabi/meta-toolchain-1.0-r0/sdk/image//usr/lib/opkg/lock

Re: meta-toolchain continues to woe

[Graham Gower]

2009/11/12 Holger Hans Peter Freyther <holger+oe@freyther.de>:
>
> ./run.do_populate_sdk.27727
> Collected errors:
>  * Could not obtain administrative lock for offline root (ERR: Bad file
> descriptor)  at /data/openembedded/build/arm/cortex-a8/tmp/work/x86_64-armv7a-
> sdk-angstrom-linux-gnueabi/meta-toolchain-1.0-r0/sdk/image//usr/lib/opkg/lock

Presumably, the lock file could not be creat()ed. Unfortunately, there
is no error checking on that creat() call in opkg and you don't get a
useful error message.

Do the paths leading up to the lock file exist?

-Graham

Re: meta-toolchain continues to woe

[Holger Hans Peter Freyther]

On Thursday 12 November 2009 05:34:20 Graham Gower wrote:

> Presumably, the lock file could not be creat()ed. Unfortunately, there
> is no error checking on that creat() call in opkg and you don't get a
> useful error message.
> 
> Do the paths leading up to the lock file exist?

They don't. Creating it will bring up the same problem with a different path. 
As in the previous mail... it is either a problem with the opkg upgrade (i was 
seeing svn updae..) or the layout changed...

The test case is really simple and I will repeat it here:

	bitbake -c populate_sdk -b meta-toolchain.bb


regards
	z.

PS: I'm seriously annoyed with the ongoing breakage...

Re: meta-toolchain continues to woe

[Holger Hans Peter Freyther]

On Thursday 12 November 2009 05:49:11 Holger Hans Peter Freyther wrote:

> The test case is really simple and I will repeat it here:
> 
> 	bitbake -c populate_sdk -b meta-toolchain.bb


bisect shows me:
	bea72c2fecde175add169bb55df1922b048030c8


IMHO the pace of the recent classes/ change was way too high, a branch and 
some builds would have been a lot better (no critics on the changes 
themselves). I'm about to write a simple meta-toolchain checker and plan to 
run it once a a week.... or if someone has a spare machine I would run this 
test there with every day..


z.

Re: meta-toolchain continues to woe

[Richard Purdie]

On Thu, 2009-11-12 at 06:11 +0100, Holger Hans Peter Freyther wrote:
> On Thursday 12 November 2009 05:49:11 Holger Hans Peter Freyther wrote:
> 
> > The test case is really simple and I will repeat it here:
> > 
> > 	bitbake -c populate_sdk -b meta-toolchain.bb
> 
> 
> bisect shows me:
> 	bea72c2fecde175add169bb55df1922b048030c8
> 
> 
> IMHO the pace of the recent classes/ change was way too high, a branch and 
> some builds would have been a lot better (no critics on the changes 
> themselves). I'm about to write a simple meta-toolchain checker and plan to 
> run it once a a week.... or if someone has a spare machine I would run this 
> test there with every day..

I think regardless of how we did this there would be some bumps. It was
on a branch for a week and it was obvious nobody was testing it expect
me. Getting on and doing it and dealing with problems quickly seemed
like the best approach and was what we agreed at OEDEM.

For this specific issue, I've pushed some fixes for meta-toolchain, let
me know if there are any further problems. My test build completed
locally (bitbake meta-toolchain).

Cheers,

Richard

Re: meta-toolchain continues to woe

[Holger Hans Peter Freyther]

On Thursday 12 November 2009 13:09:20 Richard Purdie wrote:

> 
> I think regardless of how we did this there would be some bumps. It was
> on a branch for a week and it was obvious nobody was testing it expect
> me. Getting on and doing it and dealing with problems quickly seemed
> like the best approach and was what we agreed at OEDEM.

You are right, I was simply not aware that you have posted a branch for 
review. So thank you for the changes and thank you for being stand bye
to address these. Who ever is going to be in the TSC I would greatly 
appreciate if we can formalize our process for updating the core 
infrastructure.

E.g.

1.) Do the changes in a branch
2.) Mail that the branch will be merged (top level posting, [MERGE] in the 
subject)
2.1) Get feedback from testers
3.) Merge core changes on a "core-merge" day...





> 
> For this specific issue, I've pushed some fixes for meta-toolchain, let
> me know if there are any further problems. My test build completed
> locally (bitbake meta-toolchain).

Thanks a lot! It is greatly appreciated.

Re: meta-toolchain continues to woe

[Chris Conroy]

On Thu, 2009-11-12 at 15:04 +1030, Graham Gower wrote:
> 2009/11/12 Holger Hans Peter Freyther <holger+oe@freyther.de>:
> >
> > ./run.do_populate_sdk.27727
> > Collected errors:
> >  * Could not obtain administrative lock for offline root (ERR: Bad file
> > descriptor)  at /data/openembedded/build/arm/cortex-a8/tmp/work/x86_64-armv7a-
> > sdk-angstrom-linux-gnueabi/meta-toolchain-1.0-r0/sdk/image//usr/lib/opkg/lock
> 
> Presumably, the lock file could not be creat()ed. Unfortunately, there
> is no error checking on that creat() call in opkg and you don't get a
> useful error message.
I submitted a patch to opkg back in September to fix this error message
to be a bit more useful. If you upgrade to opkg >=216 it will save you
some headaches whenever this error occurs.

http://code.google.com/p/opkg/issues/detail?id=22


--Chris Conroy

Re: meta-toolchain continues to woe

[Chris Conroy]

On Thu, 2009-11-12 at 10:14 -0500, Chris Conroy wrote:
> On Thu, 2009-11-12 at 15:04 +1030, Graham Gower wrote:
> > 2009/11/12 Holger Hans Peter Freyther <holger+oe@freyther.de>:
> > >
> > > ./run.do_populate_sdk.27727
> > > Collected errors:
> > >  * Could not obtain administrative lock for offline root (ERR: Bad file
> > > descriptor)  at /data/openembedded/build/arm/cortex-a8/tmp/work/x86_64-armv7a-
> > > sdk-angstrom-linux-gnueabi/meta-toolchain-1.0-r0/sdk/image//usr/lib/opkg/lock
> > 
> > Presumably, the lock file could not be creat()ed. Unfortunately, there
> > is no error checking on that creat() call in opkg and you don't get a
> > useful error message.
> I submitted a patch to opkg back in September to fix this error message
> to be a bit more useful. If you upgrade to opkg >=216 it will save you
> some headaches whenever this error occurs.

Apologies. I should have read your error message more closely since it
looks like you are getting the extra debug information from my patch.
Feel free to make a patch with more debug info for said error or just
let me know what you think would be appropriate and I can submit a new
one.

--Chris

> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

Re: meta-toolchain continues to woe

[Graham Gower]

Chris,
The patch you submitted prints out an error from the lockf() function
call. Which is fine, but its more common for the creat() call to fail
before hand, meaning that the lockf() function would have be passed -1
as the file descriptor.

I made a small change to hopefully improve this.
http://code.google.com/p/opkg/source/detail?r=288

-Graham

Re: meta-toolchain continues to woe

[Khem Raj]

On Thu, Nov 12, 2009 at 1:16 PM, Graham Gower <graham.gower@gmail.com> wrote:
> Chris,
> The patch you submitted prints out an error from the lockf() function
> call. Which is fine, but its more common for the creat() call to fail
> before hand, meaning that the lockf() function would have be passed -1
> as the file descriptor.
>
> I made a small change to hopefully improve this.
> http://code.google.com/p/opkg/source/detail?r=288

creat could be replaced by open() as SUS says creat is redundant
 if creat fails it will still report Could not obtain administrative lock .....
that would be misleading too.

-Khem

>
> -Graham
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: meta-toolchain continues to woe

[Holger Hans Peter Freyther]

On Thursday 12 November 2009 16:29:36 Chris Conroy wrote:
> > I submitted a patch to opkg back in September to fix this error message
> > to be a bit more useful. If you upgrade to opkg >=216 it will save you
> > some headaches whenever this error occurs.
> 
> Apologies. I should have read your error message more closely since it
> looks like you are getting the extra debug information from my patch.
> Feel free to make a patch with more debug info for said error or just
> let me know what you think would be appropriate and I can submit a new


Hi,

Koen recently upgraded opkg* to something above >= 216.

regards
	holger

Re: meta-toolchain continues to woe

[Chris Conroy]

On Fri, 2009-11-13 at 02:19 +0100, Holger Hans Peter Freyther wrote:
> On Thursday 12 November 2009 16:29:36 Chris Conroy wrote:
> > > I submitted a patch to opkg back in September to fix this error message
> > > to be a bit more useful. If you upgrade to opkg >=216 it will save you
> > > some headaches whenever this error occurs.
> > 
> > Apologies. I should have read your error message more closely since it
> > looks like you are getting the extra debug information from my patch.
> > Feel free to make a patch with more debug info for said error or just
> > let me know what you think would be appropriate and I can submit a new
> 
> 
> Hi,
> 
> Koen recently upgraded opkg* to something above >= 216.
> 
> regards
> 	holger

Hence my second message where I realized that and apologized for my
idiocy :P

--Chris