> It runs "date +%j" to get the Julian day of the year and checks to see if
> this is even or odd. I tried to categorize the tests so that a) the more
> basic versions of mcs51, hc08, z80, and ds390 (the four active ports that
> run these regression tests) would run every day, b) the variants of these
> would be split between the even/odd days, and c) roughly the same test
> load every day. So currently it is:
> Even: mcs51-small, hc08, ds390, mcs51-large, ucr3ka, ucgbz80, ucz80
> Odd: mcs51-small, hc08, ds390, mcs51-stack-auto, ucr2k, s08, ucz180
> At the moment, all of the DCF machines run the same tests for the same
> day. To make it machine specific (controlled by local/*.mk), we would need
> to change the order of some of the makefile includes. At least at the
> moment, I don't feel that I understand how all the pieces of the build
> system work well enough that there won't be some unintended side effect if
> I do this.
Would it be a good idea to use 'uname -n' and hash it in
some way into a 0 or 1 value and add that to 'date +%j'?
Hopefully this way all tests can be run every day with a
relative even distribution.