[m-rev.] diff: alternative fix for bug 264 test case

Zoltan Somogyi zoltan.somogyi at runbox.com
Mon Nov 23 03:54:48 AEDT 2015


On Mon, 23 Nov 2015 01:09:59 +1100, Mark Brown <mark at mercurylang.org> wrote:
> What is a good example to measure on?

I have half a dozen stress tests for the compiler, but while
I *think* they are freely redistributable, for most of them
I am not *sure. However, maybe you guys can help with that,
since they are from projects/companies that you worked for.

The rcpsp_cpx_bug, zm_enums test cases are from the g12
project. They each contain a big module that is the stress test,
and a bunch of supporting modules. In each case, the big module
is auto-generated (from Zinc), so I don't think they contain
any deep secrets but the supporting modules aren't, so they
may be problematic.

The mas_objects test case from Mission Critical may have
the same problem: non-interesting auto-generated large file,
hand-written supporting code.

In all three cases, it should be possible to cut down the supporting
code to just what is needed for the stress test, but this has
not been done (yet).

The mday_nasty test case is from Michael Day of YesLogic.
It is also auto-generated, and cut down from the original, real
code, so I think it is likely to be ok to distribute, but again,
I am not sure. It is standalone.

The zm_bug344 test case is similarly standalone. It is auto-generated
by the Zinc compiler.

The last test case I am sure is freely distributable,
so it is attached. While this originally WAS a stress test
for the mode checker, I think I fixed all the bottlenecks
that affected it years ago. Before I started those fixes,
the compiler would thrash the machine to death when invoked
on this module; now it compiles in five seconds, no problem.
I have no idea whether your diff touches the parts that are
still performance critical, but there is just one way to find out:
trying it.

Zoltan.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: training_cars_full.m
Type: text/x-objcsrc
Size: 1150985 bytes
Desc: not available
URL: <http://lists.mercurylang.org/archives/reviews/attachments/20151123/23056356/attachment.bin>


More information about the reviews mailing list