Re: [Boost-build] [Boost][build][v2 head] Memory footprint

From: K. Noel Belcourt (kbelco_at_[hidden])
Date: 2006-09-07 15:08:16

On Sep 7, 2006, at 12:57 PM, Rene Rivera wrote:

> Rene Rivera wrote:
>> Rene Rivera wrote:
>>> Anyone have an idea how to compile the GC code without using their
>>> makefiles? There doesn't seem to be much docs on the
>>> building/integration aspects.
>>
>> Well never mind... Figured out a simple way to build it. So if one:
>>
>> 1. Downloads the Boehm GC allocator and places it in a "boehm_gc"  
>> subdir
>> of jam/src.
>>
>> 2. Adds "--gc" option when building bjam.
>>
>> It will compile in the GC code, and use it. Right now there's not
>> advantage as the algo Dave suggested is not implemented. But hey it's
>> easier for others to play with it this way. But if you do hack  
>> around,
>> use the macros defined in the "mem.h" header.
>
> I've reached the limit of how much time I can spend on this for now...
> But I implemented the string GC as Dave outlined. But, it doesn't  
> work.
> Or I should say, it does work, but errors during one of the GC cycles
> with assertion failures in the GC code. Since this happens reliably  
> for
> me inside the var expansion code, I suspect there's a buffer  
> overrun in
> the variable expansion code :-(
>
> So if anyone has the time, it would be nice to hook in, the same  
> the GC
> allocator is, a debug allocator like dmalloc.

Thanks, I appreciate all the comments and your work.  When I get some  
time, I'll try to run purify or valgrind on bjam head.  I know the  
buffer overruns are reliably reproducible when building one of our  
applications with the vacpp toolset.

-- Noel Belcourt