Re: [new-regalloc-branch]: No patch

[Daniel Berlin <dan at cgsoftware dot com>]

Also note that while i'm not one to get territorial at all , please 
remember I did create this branch (at the end of january).   The
branch is and was, effectively, 
the mainline plus the new register allocator, because the new ra can't
go on the mainline.  This is why merges are performed quite often.
Is this a stability headache? Sometimes.  We have no real way to determine
good points on the mainline to perform merges from, it's a black art.
Does this mean you have to make an effort to minimize non-ra specific
changes? Yup.
You'll also notice the main purpose of the very large change was to
fix bugs (the extra stuff it added is specifically commented out in
the committed version, since i hadn't finished completely testing it
to my satisfaction yet).

It's frustrating to see someone rant about the fact that there were
obvious bugs in committed code, most of which, you'll note, i
specifically mentioned in emails to you and Denis, yet hadn't had
time to commit a fix for (remat problems, for instance) , or that
they spent time fixing the branch, only to see a large commit get
committed, when they had said they wouldn't have time to be working on
it.    If someone says to me that "I won't be working on this for at least a
week, maybe a lot more", and there are only 3 people working on it,
and I know what the third person is doing, I have no qualms about
doing large amounts of work and committing them until that someone
comes back to life.  In fact, you've been too busy for about a month,
by the commit logs.  I also have no qualms about leaving the branch in
a broken state when i know it won't affect the 1 other person still
working actively, and i'll have committed fixes by the time the one
taking a break gets back.  Notice i made no rant when i wasn't
committing stuff for about 4 months, and all your changes repeatedly
broke the stuff i was working on. This is because i said i wouldn't be
working on it continuously for a while, 
which left you as the only one continuously working on it, and thus,
free to do whatever you want.  I also made no rant pointing out the
obvious errors in the code, i just fixed them as i found them, once i
started working continuously on them again.  

I'm not suggesting you create a new branch, that would be quite silly,
just communicate more. It'll cause less frustration.
However, if you want something you can just tweak for months on end,
and are happy with in general, then you might be better off taking a
known-good revision of ra.c, and keeping a local copy. I'll happily do
the merge work for your code (There's no point in creating a branch
just to do that, although you could if you want).

Anyway, back to work.

--Dan


--
"When I go, I'm flying Air Bizarre.  It's a good airline.  You
buy a one way round trip ticket.  You leave any Monday, and they
bring you back the previous Friday...  That way you still have
the weekend.
"-Steven Wright