Andrei Alexandrescu wrote:
On 6/13/11 9:12 AM, Robert Clipsham wrote:
On 13/06/2011 14:39, Andrei Alexandrescu wrote:
Walter is continuously working on improving speed of pull request
integration. The bottleneck right now is the compiler test suite, which
takes hours to run.
Andrei
Is there some way we could speed this up? Obviously it can't be run for
multiple pull requests at once, as they requests could affect each
other, so it'll need a third run if they're run at the same time. Is
there some way we could work around that?
Actually I suggested him to optimistically integrate several requests at
once and then run the test suite. Only if the suite fails, fall back on
integrating one at a time.
Well, to run a minimal test ( (a) no flags; (b) -O -inline -release )
takes about two minutes, and I've personally never seen a bug which was
caught by the full test, but not by the minimal test. (I believe that
such bugs do exist, but they're rarer than one in a thousand). OTOH it's
quite common to have a bug which passes the full test on one OS while
the minimal test fails on another.
I think it would be adequate to just run Phobos unittests + the minimal
test on both Windows-32 and Linux-64, (and maybe Mac as well) and then
rely on the auto-tester to catch the one-in-a-thousand bugs.
What hardware does it take hours to run on? Could a donation of some
faster hardware help? Perhaps something like Google does could be useful
to decrease the amount of time the suite takes (see my post about
continuous integration at Google not too long ago).
Walter?
Andrei
