Re: Why not gnat Ada in gcc?

[Mike Stump <mrs at windriver dot com>]

> To: dewar@gnat.com, law@redhat.com
> Cc: gcc@gcc.gnu.org, kenner@vlsi1.ultra.nyu.edu, rms@gnu.org, rth@cygnus.com
> Date: Wed,  1 Nov 2000 22:58:41 -0500 (EST)
> From: dewar@gnat.com (Robert Dewar)

> But it is VERY rare for regressions to occur on nightly runs, I
> would say that 90% of the time the nightly builds are completely
> clean and run the entire regression suites completely. I cannot
> believe that this is true of the current gcc open tree (unless the
> test suites are very weak indeed).

Instead of believing, or guessing, why not measure.  We have a
database of runs.  Tell us what the numbers are.  I know what I am
used to.  I am used to the numbers being fairly stable.  My biggest
gripe is that they are not exactly zero.  I run the libio libstdc++
objc fortran and g++ testsuites for a couple of years and guess what,
nearly always zero.  g++ hovered around 25 for a while, then dipped
into the 9 range, where it has been stable for a long time.  I've been
watching the testresults for the past 8 years.  For gcc, it seems to
hover around 26.  Has for a while, will for a long while from now.

My experience is, once you measure, and then provide that measurement
back to the developers in a timely fashion, in a sane format, they
will actually learn to keep the numbers at zero, once they get them
down to zero.  Currently the testing feedback to developers kinda
sucks right now at the FSF, it is better inside of Cygnus, ACT.

If you provided the feedback for the public sources, in a sane manner,
I suspect you could get the developers to stop breaking things as
often as you think they do.