[EMAIL PROTECTED] (H.Merijn Brand) writes: > On Fri 05 Sep 2003 14:51, Abe Timmerman <[EMAIL PROTECTED]> wrote: > > [ Most of the Dutch is (loosly) translated ]
OK, thanks and sorry for my rambling in Dutch... > > <translation density="high"> > > C-compiler info is only retrieved from the last build. No successfull > > Configure means no c-compiler info. > > </translation> It seems switching compilers inside the smoke matrix isn't a very good idea. It tends to confuse the tooling that tries to interpret the results. For example, on AIX one might invoke the C compiler as cc, cc_r, xlc, xlc128, and many other names. It's the same compiler every time, but you get a different set of options. To make things worse, ./Configure (correctly) selects cc_r upon seeing -Duseithreads, and because this usually appears on the final line in the default smoke matrix, smoke reports from AIX often say "... using cc_r version ...", even if it's just IBM C. > > > Heb je suggesties voor een betere benadering? En hoe zou je omgaan met > > > > <translation>Any suggestions?</translation> > > > > > == > > > > > > -Dcc=gcc > > > == > > > > My personal view on this is: "Create a separate smoke config for each > > c-compiler and send multiple (smaller) reports from the same box.". I think That's almost what I'm doing. I have smokes for IBM C and GCC running in parallel on separate machines. Alternatively you'd have to set up separate smoke trees for each compiler, and then you can run them in parallel on one machine without them interfering with each other (except from taking CPU cycles). > But are *much* harder to start from cron, since you have to write wrapper > scripts that start the separated smokes in sequence (we don't want them > running side by side unless you have more than 2 CPU's), and then the smoke > guards (+HH:MM, HH:MM) will not work anymore. It seems that the underlying problem is, the smoke matrix doesn't allow you to specify all the independent dimensions. Perhaps it would help to make 'choice of compiler' an explicit variable in the matrix, but that only works if all compilers are compatible with all the other options selected. The other variable that you currently have to setup manually is the choice of what source tree to smoke. For example, at the moment you have to configure Test::Smoke separately for perl-58x and perl-current, even if most of the configuration could in be shared. In fact, I tend to be smoking both in parallel using one smoke.cfg, giving -58x higher priority than -current. > > that will reveal more detailed information about failures and the environment > > in which they occur. > > True, but still leaves the HP-UX problem where we have different compilers for > gcc in 32 and 64 bit mode, and splitting them up in two separate smokes will > only aggravate the problem I just mentioned. > > > The smoke_db has a decent interface to give you the information you may want > > for specific platforms or even cross platform/specific compiler version. That > > is why it is there (thanks Alain)! Indeed. -- $_ = "Campo Weijerman [rfc822://nl.ibm.com/]" and tr-[:]/-<@>-d and print;
