develooper Front page | perl.perl5.porters | Postings from June 2012

dude, where's my maint?

Thread Next
Ricardo Signes
June 11, 2012 05:54
dude, where's my maint?
Message ID:

A few months ago, I tried to build 5.14.3, and failed.  The issue was not one
of time, but of workingness.  That is:  I began to work through the approved
commits in cherrymaint and they would not apply cleanly.  Further, the ones
that did apply caused a number of smoke failures, and it was not clear to me
whether they were nearly caused by the cherry picks.  (At least one of them was
on George Greer's Win32 smoker, which can smoke black in the best of times.)

I think there are several ways that the maint process is failing.

A lot of it, I think, boils down to the idea that each cherry picked bugfix is
a trivial discrete entity, and that building maint is just cherry picking them
and shipping.

Or: "I saw the bug that annoys me there, and I +1'd it, and some other people
did, too, so very soon there will be a maint release with that fix."

Except the patch doesn't really work properly when applied, or conflicts with
something that was originally committed later on blead but cherry picked
earlier on maint.  Or something.

I think the simplest thing is for us to have topic branches or patch series to
include.  These can get done in a few ways:

  1. a committer can just make one; no sweat
  2. anyone can build a patch series (in the usual format)
  3. anyone can make a list of commits to cherry-pick, in order

In any case, the request to merge should go to perlbug, as usual.

In case 2 and 3, I (or anybody who beats me to it) will create a smoke-me

Teeny tiny single changes are as convenient to do through this as any others.
"Subject: apply 30a0eaf to maint-5.16, fixing perlglob" is an easy bug to file.

Once the branch has been smoked and has had a bit of time on-list to get
discussion, it may get merged.  I will decide what gets merged, based on input
on the list.  (If I have to squeeze input out of people on IRC or elsewhere, I
will report it on the list.)

I think the +3 rule is not working.  I think that cherrymaint is a good start,
but that its patch-oriented behavior may be getting in the way.

I would still like to get 5.14.3 out, yes.  I am not even opposed to 5.14.4.  I
wonder whether we ought to build a 5.14.3 soon with a few of the more important
outstanding changes, and then let 5.14.4 happen (or not) based on the number of
branches requested for it.  Specifically: whether there are zero, or more.

The expected date for 5.16.1 is coming up.  I expect to cherry pick just a few
things for it.  Offhand, I remember an improvement to the test that failed to
read Config after dropping permissions, and I remember an update to
Scalar-List-Util.  I'll look for more.  Otherwise, please file tickets.

As always, please speak up if you think this is (a) entirely wrongheaded (b)
not an improvement or (c) better, but easily improveable even more.


Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About