Issue with git fetcher

Issue with git fetcher

[Gary Thomas]

I like to run Poky/Yocto with a pre-populated source cache,
built up over time, and rely on BB_NO_NETWORK to tell me
when stuff actually needs to come from the network.  This
works quite well - only when there is a new/changed recipe
do I have to relax BB_NO_NETWORK and once I've built the
offending recipe, I save the sources (I also use BB_GENERATE_MIRROR_TARBALLS="1")
back to my source cache.

I ran across an issue with this with a recipe I imported from
meta-openembedded/meta-oe.  It contained these lines:
   SRCREV = "2fb4be9a99a2c2a9435339830e3d940171cc0d9b"
   SRC_URI = "git://git.libav.org/libav.git;protocol=git;branch=release/0.8"

So, the first time I built it without BB_NO_NETWORK, then I saved
the resulting tar file.  However, when I tried to build it again,
this time in a new build tree with BB_NO_NETWORK set, it failed!
I went looking through the tar file (just a bare clone of the git tree)
and found that the revision in question was not there.

The only thing I could see that's unique about this recipe is the
'branch=release/0.8'.  To see if this was the issue, I removed it
and replayed the whole process (rebuilding the tar file) & now fetch
works even with BB_NO_NETWORK :-)

Is this expected, or is it a bug?

n.b. I hammer on this functionality as I have to deliver fixed source
packages to customers (some of whom may not have easy network access,
some who just want to know what they are getting/using), so I rely
on being able to build with BB_NO_NETWORK set.

Thanks

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: Issue with git fetcher

[McClintock Matthew-B29882]

On Thu, Oct 18, 2012 at 11:37 AM, Gary Thomas <gary@mlbassoc.com> wrote:
> I like to run Poky/Yocto with a pre-populated source cache,
> built up over time, and rely on BB_NO_NETWORK to tell me
> when stuff actually needs to come from the network.  This
> works quite well - only when there is a new/changed recipe
> do I have to relax BB_NO_NETWORK and once I've built the
> offending recipe, I save the sources (I also use
> BB_GENERATE_MIRROR_TARBALLS="1")
> back to my source cache.
>
> I ran across an issue with this with a recipe I imported from
> meta-openembedded/meta-oe.  It contained these lines:
>   SRCREV = "2fb4be9a99a2c2a9435339830e3d940171cc0d9b"
>   SRC_URI = "git://git.libav.org/libav.git;protocol=git;branch=release/0.8"
>
> So, the first time I built it without BB_NO_NETWORK, then I saved
> the resulting tar file.  However, when I tried to build it again,
> this time in a new build tree with BB_NO_NETWORK set, it failed!
> I went looking through the tar file (just a bare clone of the git tree)
> and found that the revision in question was not there.
>
> The only thing I could see that's unique about this recipe is the
> 'branch=release/0.8'.  To see if this was the issue, I removed it
> and replayed the whole process (rebuilding the tar file) & now fetch
> works even with BB_NO_NETWORK :-)
>
> Is this expected, or is it a bug?
>
> n.b. I hammer on this functionality as I have to deliver fixed source
> packages to customers (some of whom may not have easy network access,
> some who just want to know what they are getting/using), so I rely
> on being able to build with BB_NO_NETWORK set.

Was there some recent discussion on oe-core about this?

-M

Re: Issue with git fetcher

[Burton, Ross]

On 18 October 2012 17:37, Gary Thomas <gary@mlbassoc.com> wrote:
> I ran across an issue with this with a recipe I imported from
> meta-openembedded/meta-oe.  It contained these lines:
>   SRCREV = "2fb4be9a99a2c2a9435339830e3d940171cc0d9b"
>   SRC_URI = "git://git.libav.org/libav.git;protocol=git;branch=release/0.8"

Specifying a revision *and* specifying a branch doesn't make sense
with git.  Presumably the SRCREV is the correct reference and the
branch parameter is just causing network traffic for no reason.

Removing the branch parameter sounds like exactly the right thing to do.

Ross

Re: Issue with git fetcher

[Gary Thomas]

On 2012-10-18 10:41, Burton, Ross wrote:
> On 18 October 2012 17:37, Gary Thomas <gary@mlbassoc.com> wrote:
>> I ran across an issue with this with a recipe I imported from
>> meta-openembedded/meta-oe.  It contained these lines:
>>    SRCREV = "2fb4be9a99a2c2a9435339830e3d940171cc0d9b"
>>    SRC_URI = "git://git.libav.org/libav.git;protocol=git;branch=release/0.8"
>
> Specifying a revision *and* specifying a branch doesn't make sense
> with git.  Presumably the SRCREV is the correct reference and the
> branch parameter is just causing network traffic for no reason.
>
> Removing the branch parameter sounds like exactly the right thing to do.

That's pretty much what I thought.

Maybe a warning or some such would be useful here.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: Issue with git fetcher

[Martin Jansa]

[-- Attachment #1: Type: text/plain, Size: 743 bytes --]

On Thu, Oct 18, 2012 at 05:41:13PM +0100, Burton, Ross wrote:
> On 18 October 2012 17:37, Gary Thomas <gary@mlbassoc.com> wrote:
> > I ran across an issue with this with a recipe I imported from
> > meta-openembedded/meta-oe.  It contained these lines:
> >   SRCREV = "2fb4be9a99a2c2a9435339830e3d940171cc0d9b"
> >   SRC_URI = "git://git.libav.org/libav.git;protocol=git;branch=release/0.8"
> 
> Specifying a revision *and* specifying a branch doesn't make sense
> with git.  Presumably the SRCREV is the correct reference and the
> branch parameter is just causing network traffic for no reason.

It does make sense.

e.g. AUTOREV set in some .inc file.

Cheers,

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 205 bytes --]

Re: Issue with git fetcher

[Ross Burton]

On Thursday, 18 October 2012 at 18:43, Martin Jansa wrote:
> It does make sense.
> 
> e.g. AUTOREV set in some .inc file.
Fair enough.  If anyone is actually doing that with libav then this can remain, but clearly bitbake's fetcher has a bug too then - it shouldn't hit the network to resolve a branch name if it has an explicit srcrev.

Ross