Possibly a regression bug in g++ (Updated)
yesterday I did a more test and I bounded the problem a bit more. In the
compiler options we have defined:
I have tested with -03 only with the collapse result.
I have tested with -ffast-math only with the collapse result.
I have tested with -02 only with the collapse result.
The file that fails is one that has a include file with a lot of templates and
with some size (1.5Mb).
When I compile with 4.2 the compiler has some difficult, but in some minutes
could compile and link. The colapse scenario I'm referring is to need more
than 20 minutes to compile one file with 90% of cpu 65% of memory and I have
to kill gcc. I have an amd64 with 1Gb of Ram, a bit old, but I can work with
> first of all I'm not on the debian-gcc list, so please CC to me. I'm
> helping in the development of a project dlraw.sf.net. It's a program to
> process files (digital photo) in raw .
> The main developer, Jos De Laender work with g++-4.1 and can compile the
> program without any problem. I'm working with g++-4.3, (lenny and unstable)
> and g++-snapshot (4.4) and g++ arrives to one file and begin to eat memory,
> cpu, etc, and I must kill the process because my box begin to collapse.
> I have tested it with g++-4.2 and it works. So, there's have been some
> change betwend 4.2 and 4.3 that make the problem. So, please, could some of
> you confirm that gcc fails and it's not bug in the code?
> PS the application is a qt4 program. To compile the sources you need:
> a trunk, qmake-qt4
> and wait....