Glenn Fowler wrote:
> > Yeah, I like the automated regression tests, too. I'm not sure that
> > "make all" is the right phase to run the tests, but it's not obvious to
> > me that "install" or "check" is any better. So, anyway, this would be a
> > last resort.
>
> for what its worth
> the ast distribution separates building, installation, and testing
> we run the tests ("package test" or "nmake test" to test subtrees)
> before packaging official releases
> but because the tests themselves can have bugs or "acceptible" errors
> we don't make it a default part of building or installation but
> do make it simple ("package test" or "nmake test") for users to invoke
The idea behind making the tests part of the "default" build procedure
in Solaris was to catch/solve/handle multiple issues:
- Someone breaks libc/libnsl/libsocket/etc. and ksh93 goes mad elsewhere
(and we get the blame).
- Bite the matching engineer back immediately instead of doing a full
iteration of "hit bug; file bug report ; assign bug report to engineer ;
search for the bug ; fix bug ; commit fix back" when he/she breaks
something
- Catch errors/mistakes during code updates, e.g. when I update the
ksh93 sources and forgot one piece up update.
- Catch errors in other locales (with the main focus on multibyte
locales). Currently the tests are run once per locale - which currently
include "C" and "en_US.UTF-8" - but I'd like to include at least
"ja_JP.UTF-8", "ja_JP.PCK" and "zh_CN.GB18030" in this list (once the
|strtod| issue has been fixed)
I know the tests aren't perfect... but so far they are the best
(opensource (AFAIK the Unix98/03/etc. test suites aren't opensource))
instrument we have for these tasks and currently they work very well...
:-)
----
Bye,
Roland
--
__ . . __
(o.\ \/ /.o) roland.mainz at nrubsig.org
\__\/\/__/ MPEG specialist, C&&JAVA&&Sun&&Unix programmer
/O /==\ O\ TEL +49 641 7950090
(;O/ \/ \O;)
