----- Original Message -----
From: "David Caro" <dcaroest(a)redhat.com>
To: "Alon Bar-Lev" <alonbl(a)redhat.com>
Cc: "Eyal Edri" <eedri(a)redhat.com>, "engine-devel"
<engine-devel(a)ovirt.org>, "infra" <infra(a)ovirt.org>
Sent: Friday, August 23, 2013 11:16:31 AM
Subject: Re: [Engine-devel] Proposal for new commit msg design for engine commits
On 07/20/2013 08:53 PM, Alon Bar-Lev wrote:
> ----- Original Message -----
>> From: "Eyal Edri" <eedri(a)redhat.com>
>> To: "Alon Bar-Lev" <alonbl(a)redhat.com>
>> Cc: "infra" <infra(a)ovirt.org>, "engine-devel"
>> "Fabian Deutsch" <fabiand(a)redhat.com>
>> Sent: Saturday, July 20, 2013 9:41:56 PM
>> Subject: Re: [Engine-devel] Proposal for new commit msg design for engine
>> This change to commit template has nothing to do with CI.
>> it's a change that should reflect updated components relevance to the
> This commit template is mostly invalid, as touching more than one
> 'subsystem' is possible, and has not enough granularity.
I suggest using a tag at the end with some extra syntax, like:
Components: core, storage, db
Components: all, !core, !db
> For example, database change should trigger what?
All the jobs that are tagged for that component (db upgrades I suppose).
And if the changes affect storage components then also storage, if the
changes affect others then those others too.
> Infra change should trigger what?
The same, all the jobs that are tagged as infra.
> A change of both user interface and command should trigger what?
All the jobs tagged by user interface and/or command.
> So you end up with:
> userportal: storage: core: db: some message
As I suggested before, I think it's better if you end with a commit
Components: userportal, storage, core, db
Actually it can be easily done with a tag in the gerrit comment instead
of the commit message.
> Just to make who happy?
The developer, the qe, the ci and the infra people. This mechanism is to
avoid running all the tests all the time. Of course there are some times
when all the tests should be run to make sure nothing else changed, but
most times you just need to run part of them to make sure you did not
break something obvious.
> And maybe there are 50 tests of network, and you need only 5 of them for
> the specific change, how do you mark it, so now a developer need to know
> any test? what if you add one tomorrow which is relevant to a similar
> change? how do you inform the developer that now he needs 6?
As I said before, what the developer specifies is not a list of tests,
but a list of components, that qe has to map to different sets of tests
that can change with time. So specifying webadmin will run all the tests
in that group, that might be only one, or 100, and might be
increasing/decreasing with time transparently for the developer. Adding
a new component is not common and there's no need to do it so frequently.
> Why should it be the developer responsibility and not the quality ensuring
> engineer responsibility to determine which tests should run and when?
Of course it's the responsibility of the qe engineer to determine when
and which tests should be run. But this is meant to be a new tool for
the developer not a substitute for the full qe tests, so the developer
can easily make sure that he's changes do not break anything obvious
before starting the real tests (that will take more time and resources).
The developer just adds some metadata so the qe engineer can decide
which tests to run per patch, so it's on qe's hand in the end to decide
if ignore or not the metadata and which tests to run.
> As far as this template was not actually used for anything but humans, it
> was not that important, but once you formalize it as an interface, I step
> forward and state that the subject line is not the right tool for the task
> at hand (or any for this matter).
I agree with that, I think that it should be a tag similar to Change-Id,
at the end of the commit message.
> The fact that you have in each commit are the sources that are modified,
> all the other data is just plain noise. From the sources that are modified
> you should be able to derive a test plan with high chance that this test
> program will be correct. Human intervention should be possible by ordering
> special tests that are outside of the standard policy, for cases in which
> the standard policy of deriving tests from sources is too narrow.
That's just not true. The sources are complicated enough to make two
changes in the same file to affect different components. Any reused code
is prone to affect multiple components, making it really hard to
determine by which changed files which tests to run. And if you go down
to the function/class level it's even harder to decide and to maintain.
And of course it's not human error free, as the metadata in the
files/directories is defined and maintained by a human. And in my
opinion is a lot harder to implement and maintain, and a lot less agile,
and does not get rid of the human factor.
1. Commit messages are not the place to specify metadata to interact with automation, it
is a record for future reviewer.
2. The metadata within sources are the mean to automate the list of tests related to a
specific source without human interaction on each commit.
3. If there is doubt from list of tests run them all, this is simple rule for automation.
4. The metadata within files are not the only way to order tests, one can do this manually
via jenkins or any other mean as one can now.
5. The metadata within files will help us to achieve other targets, such as automatic CC
maintainers, verify that +2/-2 are set by authorized maintainer of component etc...
6. Gerrit 2.6 supports labels which are actually what you are trying to achieve using
the commit message because of lack of other solutions.
Until we have metadata within source, and we don't as we void discuss this for long
time, and try to find manual workarounds and solutions.
Add label for each test, this will allow ordering tests via the gerrit web interface post
After people start to do this over and over and over and over they will appreciate the
need to add metadata into the source tree.
In future, if this works out we can help gerrit to improve by enhancing the labels into
free text/combo box etc...
Or maybe try to do this now if you like to help out gerrit to improve...