|
|
Blitz Support : |
From: Derrick Bass (derrick_at_[hidden])
Date: 2004-03-16 15:38:16
I use Blitz on Mac OS X all the time. This sounds like something I have
experienced before. There seems to be a bug that causes the Mac OS X
version of gcc to sometimes use extraordinary amounts of memory and
time when optimization is turned on (I once ran it for 3 days taking
1.5 GB of VM, but it still hadn't finished). It does not get triggered
by the blitz testsuite or example programs, but real-world programs
seem to have enough complexity to trigger it.
Older versions of gcc for all platforms would sometimes display this
behavior, but it was fixed long ago. I don't know if the Apple version
is suffering from the same problem or just one with the same symptoms;
it is based on pretty recent GCC sources, so in theory it shouldn't
display such behavior.
My first suggestion would be to remove all the CPU specific
optimizations. In fact, remove all the options you can except the -O
option (sometimes even innocuous seeming options interact in strange
ways with one another; if you find this helps, that is important
information for a bug report).
If that doesn't help, I suggest you download and compile a recent copy
of gcc (version 3.3.3 is the latest stable release) yourself
(http://gcc.gnu.org). There are a few limitations, compared with the
Apple supplied version of gcc: it does not support static members of
template classes, and I think there are some problems associated with
Cocoa; also you won't be able to use cool new Xcode features like
ZeroLink and distributed compiles.
Finally, you may want to create a precompiled version of your code (use
the -save-temps option and look for the .ii file) and send it to Apple
with a bug report (http://bugreporter.apple.com).
Derrick Bass