Re: patches and regtest checking

[Graham Percival]

On Sun, Jul 24, 2011 at 06:16:24PM +0200, address@hidden wrote:
> I completely second this and would like to apologize for wasting anyone's time
> with regtests.

oops, my initial email was a bit too harsh.  I should have written
"in the future, if James finds problems in your patch, you should
be ashamed".

We're still sorting out how to organize development, both
officially (via GOP) and unofficially (via emails like this).  In
the past, we just sent patches "whenever", and people looked at
them "whenever".  The whole idea of "James should not find
problems" is a *change* of policy/tradition, not a statement about
existing policy/tradition.


> Anytime someone runs regtests for me, it's very helpful (I am
> having trouble rebuilding lily from a blank tarball, so I can't get a separate
> regtest branch up and running).

Hmm, that sounds like something we should take a look at -- but of
course there's a lot of things to juggle, include real work.  :)

> I used to belabor under the assumption that people ran regtests if they saw
> nothing obviously wrong but wanted to test some things out: I now see that
> people run them as a courtesy without the intention of playing with the patch.

Clarification: James is running regtests as a "courtesy" to
reviewers.  The intent is that nobody should review a "serious"
patch unless we have evidence that the patch does not introduce
obvious regressions.

In the future we could toy with ways of making a clearer
distinction between "serious" patches and "proof of concept"
patches.

>  As this is the case, I will adopt Graham's suggestion for all future work.
>  I'd even go one step further in requesting that other people run regtests 
> only
>  if they are asked for, but I don't know if this is problematic for other
> individuals.

I would rather keep the blanket rule that "James tests all patches
unless explicitly stated otherwise"; most patches on the list are
"serious" patches intended for pushing.  I'd rather have you add
"this is a proof of concept; do not test" to your patches by
default.

Cheers,
- Graham