[dmd-internals] [D-Programming-Language/dmd] 10640a: dang, forgot that one, too
Jonathan M Davis
jmdavisProg at gmx.com
Wed Nov 7 04:07:34 PST 2012
On Wednesday, November 07, 2012 12:38:49 Leandro Lucarella wrote:
> Walter Bright, el 6 de November a las 22:12 me escribiste:
> > On 11/6/2012 9:04 PM, Brad Roberts wrote:
> > >On 11/6/2012 7:44 PM, Walter Bright wrote:
> > >>I do run the test suite locally before committing.
> > >
> > >On all platforms? On the code you _actually_ committed not vs what you
> > >have in some other client?
> > >
> > >Based on the frequency of omitting a file and/or seeing a broken build,
> > >the answer to both of those is all too often 'no'. Both of which would
> > >be true if you followed the pull + merge model that everyone else
> > >follows.>
> > I run it on all the platforms but FreeBSD64 and then copy the files
> > to the git repository on Linux.
>
> That clearly doesn't detect errors when you forgot to commit stuff. Is NOT
> the same as going through the autotester which compiles everything from a
> fresh checkout.
Well, there are really 2 problems here.
1. Copying files across rather than using git on Windows.
2. Committing directly instead of creating pull requests and letting the pull
tester test them.
If Walter were to just create pull requests (even if he's the one that merges
them rather than having them reviewed), then problems caused by #1 would be
caught by the pull tester. And even if he were to commit from Windows, that
could still get screwed up (e.g. forgetting to add a file).
So, much as most of the rest of us think that it would be better if he just
used git on Windows, if he'd just do pull requests, then he can copy the files
to Linux and use git there if he wants to without causing problems, because
the pull tester would catch them. So really, the problem is that he's not
doing pull requests, which is understandable, since he's never had to operate
that way before, but if we want to be appropriately organized and robust in
our process, then he really should be creating pull requests like the rest of
us do rather than committing directly (at this point, pretty much no one with
commit privileges other than him commits directly except for really minor
stuff). And I think that he's said previously that he agreed that he should
move to that in the future, but he obviously hasn't gotten around to making
that change yet (just like he hasn't gotten around to using branches either).
- Jonathan M Davis
More information about the dmd-internals
mailing list