Re: GCC 3.3 release criteria?

[Matt Austern <austern at apple dot com>]

On Tuesday, January 7, 2003, at 09:02 AM, Craig Rodrigues wrote:

> Hi,
>
> What are the release criteria for GCC 3.3?
> This web page lists them for GCC 3.1, but I can't find anything for 
> GCC 3.3:
> http://gcc.gnu.org/gcc-3.1/criteria.html
>
> Also, on the following web page, how was it decided what packages
> to list for GCC testing purposes:
> http://gcc.gnu.org/testing/index.html
>
> The list of applications on that web page seems to be
> a subset of the applications listed on the GCC 3.1 criteria web page.
I notice that there is a section on compile-time performance in the
3.1 release criteria, but that it's rather vague.

I suggest that for 3.3 compile-time release criteria we choose one
or two large files.  For C++, something from KDE or Mozilla might
be a good choices---I hear people complaining all the time about
how long they take to build.  I also suggest that the flags we care
most about for compile-time performance should be -O0, or perhaps
-O0 -g.  (Rationale: people expect that optimization takes time, but
they also expect that there should be *some* compilation mode
that's fast.)

Finally, while the 3.1 release criteria (up to a 15% regression from
2.95 was acceptable) were reasonable because 2.95 -> 3.x was such
a huge change, I hope we're not going to say, for each release, that a
15% regression from the previous release is OK.  I'd suggest that the
quantitative criterion form 3.3 should be literally the same as the one
for 3.1: no more than a 15% regression from 2.95, i.e. parity with 3.1.

			--Matt