Re: PATCH to optimize two Channels methods

[Richard Earnshaw <rearnsha at arm dot com>]

> Tom Tromey wrote:
> 
> > As far as I know, libgcj falls under the same rules as the rest of the
> > compiler, see http://gcc.gnu.org/develop.html.  In particular, 48
> > hours after a regression is introduced, you may back out the patch.
> 
> Only under certain conditions, including that it's an "important"
> target, and that "two people wih write priviliges" determine that
> "the best course of action is to revert the patch."
> 
> But quibbling over rules is counter-productive.  Backing out a patch
> without the patcher's agreement is very seldom done.  Suggesting a
> patch be backed-out is touchy, too.  And you need to distinguish
> between a patch that is wrong (it's the wrong fix or does more harm
> than good), and a patch that is incomplete (it makes a real improvement
> or fixes real problems in a reasonsable manner yet causes regressions
> that appear to be fixable).  Prematurely talking about reverting the
> latter is annoying.  First try to engage in a dialogue about what the
> problem might be, how easy it might be to fix, and only if it determined
> that the fix is too difficult or remote then it is time to discuss
> if we're better off with or without the patch.
> 
> "No regressions" is a guideline, not an absolute.  We even break
> platforms when they're obsolete and no-one is willing to maintain them.
> (Of course embedded arm is not in this category.)
> 
> I apologize for not responding to Richard's problem, but it's inevitable
> that sometimes I will drop an email.  Talking about backing out my
> patches because I haven't responded to one message makes me defensive
> and annoyed.

OK, let's all agree that there was an unfortunate breakdown in 
communication here... In retrospect I should have phrased my message 
differently.

I think the timeline went something like this:

Thursday evening I noticed that the build was failing on a run I started 
at home.

Checking again the following morning at work showed that the problem was 
persisting and I sent my original mail.

I noticed that you didn't send any mails to java-patches at all on Friday, 
so we were then into the weekend.

Finally on Monday I noticed that you were communicating again, but still 
hadn't responded to my message.

By this point I was also feeling somewhat touchy (5 days since the 
breakage occurred and no response), hence my email.  I don't want to start 
bombarding folks with emails, but at the same time, with no response it's 
hard to know what is happening.

I'd really like to find a way to eliminate long periods like this when 
development is being inhibited by fairly trivial breakages.  As Tom points 
out, this is by no means the first time that this has happened; it's a 
pity that the main automatic regression checker doesn't seem to be 
checking libjava on a cross build (in fact, it doesn't seem to be checking 
a cross build at all these days).

R.