This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: 'make check' does nothing?


On Mon, May 21, 2001 at 05:03:38PM +0200, Gabriel Dos Reis wrote:
> Phil Edwards <pedwards@disaster.jaj.com> writes:
> | Okay, next silly dejagnu question:  does this code from dg.exp
> | 
> |     ## Set ulimits.
> |     ## This should normally be handled on a per-test basis through @xxx@-keywords.
> |     ## The following limit is expressed in kilobytes.  For history, have a
> |     ## look at http://gcc.gnu.org/ml/libstdc++/2000-10/msg00029.html
> |     set maximum-memory-usage 16384
> |     set shell-ulimit-command ulimit
> |     remote_exec host ${shell-ulimit-command} "-d ${maximum-memory-usage}"
> |     remote_exec host ${shell-ulimit-command} "-v ${maximum-memory-usage}"
> | 
> |     # Main loop.
> |     dg-runtest [lsort [glob -nocomplain $srcdir/*/*.cc]] \
> |             "" $DEFAULT_CXXFLAGS
> | 
> | really have an effect?
> 
> They should, according to the master of DejaGnu, Rob Savoye :-)
> If you find they don't then please let me know as soon as possible.

Well, I don't think I can /prove/ they are not working.  But I strongly
suspect they aren't, based on 1) the extraordinary thrashing that my
machine performs during the test runs (it hasn't done this since we
added the memory limits under the old mkcheck), and 2) from my limited
understanding of DejaGnu, and reading the libstdc++-v3.log file, it looks
like the ulimits are simply being called by themselves before anything else
runs, in separate subshells of their own.  That won't effect later subshells.

Also, the memory limits can't be applied to /all/ subshells, because then
the compiler would die after using up its 16 allotted MB of memory.
(Another reason to suspect it's not working -- the compiler isn't dying.  :-)
We need to only run the ulimit calls in the same subshell as when we run
a test executable.

I'm just reiterating this for the record and for those who come after.
Also I have no idea how to fix it, so I'll stop complaining now.  :-)


Phil

-- 
pedwards at disaster dot jaj dot com  |  pme at sources dot redhat dot com
devphil at several other less interesting addresses in various dot domains
The gods do not protect fools.  Fools are protected by more capable fools.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]