[RFC] Add TIMEOUT environment variable support to all tests, usable via <test>-ENV= in Makefile.
Carlos O'Donell
carlos@baldric.uwo.ca
Thu Jan 8 20:59:00 GMT 2004
On Thu, Jan 08, 2004 at 12:07:28AM -0800, Ulrich Drepper wrote:
> That's not what I meant or want.
>
> Use an envvar TIMEOUTFACTOR or so. In the test skeleton code define a
> variable to 1 by default, otherwise to the integer value of the envvar
> (if > 0). Multiply the usual timeout with this number.
>
> And then add this variable in the environment when you compile. No
> Makefile changes. Otherwise, what's the point? One could just have
> changed the code.
I see timeouts serving two goals:
o Halting a test that is locked up.
- Case in point bug-iconv3, when the locking routines were
incorrect on hppa (fixed with malloc-machine.h)
o Alerting maintainers to a bug that has caused a *big* slowdown.
- Messing up lazy linking.
- arch<->generic code interaction.
- Kernel<->glibc interaction.
- tst-tls13 alerting that dlclose is *way* too slow as a result
of having to flush all aliased addresses on unmap.
As such I don't want a linear scaling factor to *all* my timeouts, just
one of them. Would you accept a patch that lets you override per-test
timetouts *and* an environment variable to pre-scale timeouts?
Cheers,
Carlos.
More information about the Libc-alpha
mailing list