[oe] SRCREV defined too late -how to fix?

[Mike Westerhof]

Martin Jansa wrote:
> On Wed, Apr 07, 2010 at 10:07:12PM -0500, Mike Westerhof wrote:
>> Chris Larson wrote:
>>> On Tue, Apr 6, 2010 at 6:50 AM, Mike Westerhof <mike at mwester.net> wrote:
>>>
>>>> After a recent commit that rearranged where (and how) SRCREVs are defined,
>>>> building results in:
>>>>
>>>> NOTE: preferred version 2.6.27.8+svnr1 of linux-ixp4xx not available (for
>>>> item kernel)
>>>> NOTE: preferred version 2.6.27.8+svnr1 of linux-ixp4xx not available (for
>>>> item kernel-module-ext2)
>>>> NOTE: preferred version 2.6.27.8+svnr1 of linux-ixp4xx not available (for
>>>> item kernel-module-jbd)
>>>> (etc.)
>>>>
>>>> Note the bogus "svnr1" on the end -- I think this has happened because
>>>> someone changed OE to define SRCREVs in each package.  So now, deep in
>>>> recipes/linux/linux-ixp4xx.inc, we find:
> 
> Hi, that was me, sorry for inconvenience, but we should be able to
> resolve it..

I hope so.  Cleaning stuff up and making it all pretty is great, but it
really sucks when it not only breaks things, but also removes
functionality we previously enjoyed. :(

>>>> SRCREV = "1089"
>>>>
>>>> I think that's wrong.
>>>>
>>>> It *does* (sort of) work -- it actually results in that SRCREV (1089) being
>>>> built, despite the messages (above) telling you that it can't find the
>>>> PREFERRED_VERSION with SRCREV == 1.
>>>>
>>>> But what that line does is not really what we (the ixp4xx kernel
>>>> maintainers) had in mind.
>>>>
>>>> The idea is that the SRCREV, along with the kernel version, would be
>>>> defined as a preferred version, like this line which is in
>>>> machine/include/ixp4xx.inc:
>>>>
>>>> PREFERRED_VERSION_linux-ixp4xx ?= "2.6.24.7+svnr${SRCREV}"
>>>>
>>>> Of course, SlugOS overrides that because that distro prefers a more recent
>>>> kerrnel:
>>>>
>>>> PREFERRED_VERSION_linux-ixp4xx = "2.6.27.8+svnr${SRCREV}"
> 
> This is exact line from your local.conf? Where did you define that newer
> SRCREV for this PV?. I've grepped whole tree for
> defined SRCREV_pn-something and changed that (see OPKG_SRCREV also
> defined in SlugOS conf).

This is the exact line from
openembedded/conf/distro/include/preferred-slugos-versions.inc, actually.

It doesn't matter if I remove the similar line from my local.conf; it
still fails because SRCREV is not defined when the line is evaluated.
Any user building any IXP4xx kernel in OE will be greeted with the
aforementioned messages, even if they have an empty local.conf.

> Have you tried
> 
> SRCREV_pn-linux-ixp4xx = "4833"
> PREFERRED_VERSION_linux-ixp4xx = "2.6.27.8+svnr${SRCREV_pn-linux-ixp4xx}"
> 
> in your local.conf?

So basically, you are suggesting that we leave the SRCREV in the
package, but I should define it along with the preferred kernel in one
of the SlugOS distro config files -- in other words, it's reverting your
changes specifically for SlugOS and the ixp4xx kernel.

I'm fine with that; I like the old behavior far better than the
(mis)behavior I observe with the new model.  I'll try that and if it
works, I can commit that change.

>>>> That's not really a "recent" kernel, of course -- my local.conf file has a
>>>> more recent one that I've not yet committed.  The point is that the way it
>>>> *USED* to work, one could use the normal means to set both PREFERRED_VERSION
>>>> and SRCREV.  With the recent commit, we can't do that anymore.
>>>>
>>>> Apparently SRCREV isn't defined soon enough with this new structure, so we
>>>> get the messages about svnr1 not being available.  And, of course, one
>>>> cannot override the preferred version anymore due to the use of "=" instead
>>>> of "=?" for the assignment.  (At the very least, when all the SRCREVs were
>>>> moved into the recipes, shouldn't the weak assignments have been preserved?)
> 
> Yes you can override it with _pn-something in local.conf (as ie
> fso-autorev.conf, shr-autorev.conf), SRCREV_pn-abc = "123" is preferred
> over SRCREV = "12" in abc_svn.bb. Doesn't matter if there was weak or
> normal assignment in recipe.
> 
> The point for not-weak assignement is because some time ago, RP said,
> that would be better to define SRCREV = None ie in bitbake.conf to get
> better error message than those "svnr1". Using weak SRCREV in recipes
> wouldn't be possible.
> 
>>>> At any rate, I wish to remove the bogus messages about svnr1, and I want to
>>>> restore the behavior that would allow me to provide the SRCREV in my
>>>> local.conf.  What is the correct way to do this with the new "world order"
>>>> as it relates to SRCREVs?   To what config file do I move that
>>>> SRCREV="1089"?  Who would I anger if I just put it back to the way it was?
> 
> Please try solution suggested above first.

But if understand your proposed solution, it is just putting it back to
the way it was for SlugOS.  (It'll still be broken for Angstrom and
other distros that build IXP4xx kernels, though).

>>> >From what RP was saying on IRC the other day, you can utilize "%" in version
>>> preferences for versions that include srcrev, as a marker.  2.6.27.8+svnr%.
>>>  I haven't looked at that code, though, so I may be completely out in left
>>> field here :)
>> NOTE: preferred version 2.6.27.8+svnr% of linux-ixp4xx not available
>> (for item kernel)
>> NOTE: preferred version 2.6.27.8+svnr% of linux-ixp4xx not available
>> (for item kernel-module-ext2)
>> NOTE: preferred version 2.6.27.8+svnr% of linux-ixp4xx not available
>> (for item kernel-module-jbd)
>>
>> :(  Does this %-sign feature perhaps require a new bitbake version?
> 
> Yes this needs bitbake master.

That doesn't help me, then.  I can probably upgrade, but I'd like to get
this kernel issue resolved before I upgrade bitbake and do all the
testing necessary to make sure it all still works.

> Regards,



-Mike (mwester)