On Thu 16 Aug 2001 14:20, Nicholas Clark <[EMAIL PROTECTED]> wrote:
> On Thu, Aug 16, 2001 at 01:50:06PM +0200, H.Merijn Brand wrote:
> > It's delicious to see the OK reports flowing in. A recent change in the API
> > defenition has made the '-Dusereentrant' part of the configurations
> > superfluent. Could you please change the mktest scripts to drop it?
>
> I was wondering about hacking the scripts a bit to make the configurations
> tested less intertwined with the code, as I'm testing different configurations
> on FreeBSD and Debian, and it's getting a pain to keep having to copy the
> configuration bits back into the new version of the test program.
Good thought. Someone already suggested a www available base conf for the
non-thinking, but personally I don't like this.
My thoughts were towards a *perl* script that was required from mktest,
defining a multilevel LoL in which the structure defines the number of loops.
I was thinking towards a perl script rather than some kind of .ini or .cfg
file to be able to build in user deps and system deps and maybe even speed
checks. (If the first test takes longer than 1 hour, skip tests x, y and z),
which could be specified for each test:
my @conf = (
[ "Threading",
[ "", "-Dusethreads -Duseithreads" ]
],
[ "64bitting",
[ "", "-Duse64bitint", "-Duselongdouble", "-Dusemorebits",
"-Duse64bitall", "-Duse64bitall -Duselongdouble" ]
],
[ "Debugging",
[ "-DDEBUGGING", "" ]
],
[ "Optimizing",
[ "", # Default
"-O", "-g" ]
],
);
or extended with timing levels
my @conf = (
[ "Threading",
[ [ 0, "" ], # do always
[ 2, "-Dusethreads -Duseithreads" ], # Do only if faster than cat 2
],
[ ...
for whatever meaning we tel the category level to be, all reletive to the
first configuration.
> Also I suspect that I may be wanting to test a 4d grid of permutations rather
> than the current 3, I might like to add another test that changes cflags
I like that, but please be considerate to Win32 users.
> (so make the -DDEBUGGING slightly more general), copy the perlio/stdio test
> when -Uperlio (rather than running 2 tests) and experiment with tree cloning
Optimizing is done in both mktest and mkovz.pl, look for hpux. AFAICR, I
didn't optimize the perlio for -Uuseperlio away, cause it really hurts when
core dumps occur with -Uuseperlio and $PERLIO=perlio (Don't know if that still
holds, but it *did* in the early days of perlio.
> using hard links rather than make distclean)
Should be optional, and probably in some configuration script (Makefile.PL?)
--
H.Merijn Brand Amsterdam Perl Mongers (http://www.amsterdam.pm.org/)
using perl-5.6.1, 5.7.1 & 623 on HP-UX 10.20 & 11.00, AIX 4.2, AIX 4.3,
WinNT 4, Win2K pro & WinCE 2.11 often with Tk800.022 &/| DBD-Unify
ftp://ftp.funet.fi/pub/languages/perl/CPAN/authors/id/H/HM/HMBRAND/
