Re: The direction of malloc?

[Siddhesh Poyarekar <siddhesh at redhat dot com>]

On Tue, Dec 10, 2013 at 07:59:51AM +0100, Andreas Jaeger wrote:
> This was done in the past to sync with Doug Lea's malloc implementation.
> 
> Our malloc seems to be based on version 2.7 but 2.8.6 is current (see
> http://gee.cs.oswego.edu/) - and Wolfram had a ptmalloc v3 on his
> homepage (http://www.malloc.de/en/) that seems not to have been
> integrated with glibc.
> 
> So, going to ptmalloc3 is an option to consider as well...
> 
> Since Wolfram seems not anymore interested with glibc (I've copied him
> on this email so that he could chime in), we can just go forward.
> 
> Before changing the code massively, I suggest to discuss whether we stay
> with ptmalloc2 or move to ptmalloc3. If we do the later, it might be
> better to do the overhaul after the update.
> 
> Summing up: Yes, we own it and should adopt it - but let's not make our
> life harder if possible,

>From a quick look through ptmalloc3, I don't think merging ptmalloc3
changes is going to be any easier before making any changes to our
malloc.  Our code base seems to have diverged enough (with multiple
arenas, per-thread arenas, dynamic thresholds, hooks, probes, etc.) to
make it very different from ptmalloc.  I definitely see merit in
looking for stuff in ptmalloc3 that we can use, but given the current
complexity of that task, I don't think it makes sense to have it hold
up cleanups.

Siddhesh