[Boost-users] Re: 1_32 compiling xml_grammer

From: Russell Hind (rh_gmane_at_[hidden])
Date: 2005-04-01 15:54:31

Jeremy Jurksztowicz wrote:
> Hi,
> 
> I am having problems compiling 1_32 on a G4 darwin system with GCC 3.x.
> 1_31 compiles fine, but when I the xml_grammer.o is being compiled in 1_32, it seems to take forever... as in forever. My computer seems to be doing things but I wonder if a modern processor should take over 4 hours to compile 1 lousy object file. That is how long I have left the compiler running for, 4 hours, still churning away at xml_grammer.o.
> Is it normal for xml_grammer to take a long time to compile? Nothing in xml_grammer.hpp (and xml_grammer.ipp) indicates that it should take so long. Perhaps it is my bjam build? I am using the bjam I installed with 1_31 to compile 1_32.
> 
> Thanks,
> Jeremy Jurksztowicz

I saw this when I briefly installed boost on my G4 laptop last week but 
didn't have much of a time to investigate.

It is xml_grammer.cpp from the serialization library.  It compiles fine 
in debug builds, but in release builds, the compiler must have a bug 
that sends it in to an endless loop.

Looking at 'top', the mem size is typically <100Mb for the compiler, but 
on this file, it got up to >1Gb after about 10 minutes, so I killed the 
process.

If someone has time to investigate it would be good as currently you 
can't build serialization release, so it means building all libs that 
build after serialization one by one.

And with the talk of just building the release libs by default, then 
this may become a bigger issue of OS X users.

(OSX 10.3.8, Apple's gcc 3.3)

Cheers

Russell