[OE-core] [PATCH 1/1][RFC] devtool: Upgrade feature

Paul Eggleton paul.eggleton at linux.intel.com
Thu Aug 13 10:06:13 UTC 2015


Hi Leo,

On Tuesday 11 August 2015 14:04:14 leonardo.sandoval.gonzalez at linux.intel.com 
wrote:
> This is a new devtool's feature, which upgrades a recipe to a
> particular version number. Code was taken from [1] where some modifications
> were done (remove all email, buildhistory and statistics features,
> use devtool logger instead of AUH logger) to adapt the devtool framework.
> Once the upgrade is over, the new recipe will be located under the
> devtool's workspace. This is a first approach to this feature; pending
> tasks include:
> 
> 1. The AUH [1] is used to rename and update the recipe. AUH is not
> using the lib/oe/recipeutils library to do this work. AUH ported code should
> use this library which is the one being used by the rest of the devtool
> features.
>
> 2. Currently, when 'update-recipe' is executed, the recipe under workspace
> is updated with latest commits. The only task missing is to replace the new
> recipe with the old one, commonly located under the meta layer.
>
> 3. When patches fail to apply, follow the PATCHRESOLVE behavior instead of
> just failing.
> 
> 4. Patches most of the time do not apply correctly on the new recipe
> version, so include a command line option to indicate the system not to
> apply these, so it can be applied manually later on.

So, this is a good first implementation and gives an idea of how the command 
will work - and I'm quite keen for us to have this, I've had several people 
asking me recently about when we'll be adding it, so it's definitely something 
we want.

However, I'm concerned about the volume of code we're adding here, some of 
which duplicates what we already have in devtool or elsewhere in lib/oe. I 
know this is the easiest approach and you've noted it's todo item #1 above to 
work on - I'll be a lot happier if you can improve that before the final 
version.

In particular, I see the upgrade() function is duplicating a bunch of code 
from modify() - we should split the common parts out to separate function(s) 
that can be called from both places instead.

Some other comments:

* I can see it's actually making changes to the original recipe - I found this 
because I'd run a devtool upgrade on dropbear and specified the wrong version 
to --version Ctrl+C'd out during the fetch part, and the original dropbear.inc 
was modified instead of the one in the workspace. We don't want to touch the 
original recipe, not until you run update-recipe at least.

* As far as the source tree is concerned I think that this is doing things 
slightly backwards - it's using the AUH code to do the upgrade, and then it's 
extracting the code. What I'd like to see happen is it extract the code for 
the old version (including applying patches), use the AUH code to figure out 
how to fetch the new version, fetch it into a different branch (assuming it's 
not already there since it was fetched from git), then rebase the patches onto 
the new version - the user can then use git to fix things up if patches don't 
apply.

* I'm not sure if using an existing source tree is a reasonable default for 
this command - I'm not even sure it's something people will want to do at all 
in the context of an upgrade. I'm willing to be convinced otherwise though - 
any opinions (from anyone)?

* We need oe-selftest tests for this that test as much of the functionality as 
possible. This may require adding some simple "synthetic" recipes to meta-
selftest so that we have something that's always there to upgrade. Based upon 
my recent experience in devtool (with my own code), the sooner we add these 
the better.

* I'd like to see the implementation for this in its own file rather than in 
standard.py.

* Probably just a result of this being WIP, but I found that if I don't 
specify a version with --version then it says "NOTE: Upgrading to None" and 
then fails with "ERROR: cannot concatenate 'str' and 'NoneType' objects".

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre




More information about the Openembedded-core mailing list