On Jan 12, 2007, at 12:44 AM, Vladimir Ivanov wrote:
On 1/12/07, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote:
On Jan 11, 2007, at 5:55 AM, Alexey Varlamov wrote:
> I guess this is about adding one more mode to HUT CC runs?
> +1 for that, too.
So, let's define the required configurations for CC.
Now CC configuration (in buildtest module) includes:
1) build of classlib
2) build of drlvm
3) run drlvm tests
4) run of classlib tests over drlvm
All configuration settings are default: debug build of drlvm by gcc
or msvc
compilers.
What should be added?
beats me :) I haven't looked at it for a while.
How it should be organized to have a reasonable time for 1 CC cycle?
A ha! (oh, I've been waiting and waiting for this...)
Here's what we need to do - we need to define multiple CC
configurations, and do it in a reasonably modular way so that it
makes it easy for people to create new ones (like a perf module that
runs a SPEC or DaCapo benchmark...)
First is the "fail fast", to show us when we've broken something
obvious. That would probably be what we have today. We want them
running in as many places as possible, as often as possible.
Next are "medium" configurations, things that take longer, something
you might want to run once a day on each platform. There could be
many of these, and probably will include all sorts of different things.
Finally are the "long/heavy" configurations, like the JCK, which you
might run continuously, but are long running, and may have more
complicated results. These also may run not on SVN HEAD, but on a
revision for which the "fast" and "medium" runs have completed, so we
know that the heavyweight testing / long running testing isn't being
on code that we have zero knowledge about.
So I think that for this case, it's a "medium" configuration. I have
no clue how these might work. Ideally, it would be nice if all of
these configurations were distinct and kept in SVN, and any user
would check them out, and then maybe add them to a list of configs
for their local CC to run.
geir
