Re: [PATCH] gstreamer1.0*git: Update to build current AUTOREV from master branch.

[Peter Urbanec <openembedded-devel@urbanec.net>]

On 27/02/15 08:52, Burton, Ross wrote:
>
> On 26 February 2015 at 17:57, Peter Urbanec
> <openembedded-devel@urbanec.net <mailto:openembedded-devel@urbanec.net>>
> wrote:
>
>     -SRCREV = "127202d6f65584891dabf92be031f0d170b0e7f1"
>     +SRCREV ?= "${AUTOREV}"
>
>
> I'd say that the git packages should track the tarball releases *and be
> tested as such* so that they actually work out of the box and don't
> break over time as upstream changes.

I disagree. There's no point in having a git version of a package that 
is exactly the same as a release version. Doing so would also bring in 
all the patches specific to that release, which may not be appropriate 
for a different git revision. If you want a tested release, you would 
stick to something like gstreamer1.0*1.4.5.bb instead of choosing a git 
version.

> If someone wants to make a bleeding-edge build then it's a simple change
> to set the SRCREV to AUTOREV at the distro level.

Well, as it's shaping up, that may actually not be possible in oe-core 
packages. It certainly does not work with the current gstreamer1.0*git 
recipes, because as soon as you do that you end up with infinite 
recursion when expanding SRC_URI.

I think there are two main use cases for git builds. Case one is keeping 
up with the bleeding edge, which implies SRCREV="${AUTOREV}". Case two 
is freezing at a particular revision that suits you, which implies 
SRCREV="git-hash". In case one, you will probably want to run a fairly 
lean set of patches (or even none) to track the upstream. In case two, 
you could potentially have a comprehensive set of patches that will only 
apply to that revision. Both cases will be a lot easier to manage if the 
*git.bb recipe gives you a clean git repository without patches.