Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 13 Mar 2006 12:52:31 -0800
From:      Jason Evans <jasone@FreeBSD.org>
To:        SANETO Takanori <sanewo@ba2.so-net.ne.jp>
Cc:        freebsd-current@FreeBSD.org
Subject:   Re: malloc options incompatibility between phkmalloc and jemalloc
Message-ID:  <4415DB8F.6060409@FreeBSD.org>
In-Reply-To: <4413C5D8.70208@ba2.so-net.ne.jp>
References:  <200601161150.k0GBoamk010563@locutus.newmillennium.net.au>	<91B40C65-A11B-427E-B352-8B6EF8A55864@freebsd.org>	<43CECA2C.6030400@ba2.so-net.ne.jp>	<6A234191-C4DA-4E1B-BB4D-1F3F8043DDA9@freebsd.org> <4413C5D8.70208@ba2.so-net.ne.jp>

next in thread | previous in thread | raw e-mail | index | archive | help
SANETO Takanori wrote:
> I don't think this is serious problem, but it's annoying.
> 
> When I put 'k' in /etc/malloc.conf, some commands (cvsup and
> vmware-checkvm) show warning about unknown character. These commands
> seem to have phkmalloc implementation linked statically.
> 
> On the other hand, when I put '>' in /etc/malloc.conf, many commands
> will show warnings.
> 
> For second case, isn't it reasonable for jemalloc to just ignore
> phkmalloc specific characters?
> For first case, how about introducing new option source (i.e.
> /etc/jemalloc.conf, JEMALLOC_OPTIONS or something like that)?
> 
> What do you think?

This is a sticky issue.  I'd rather leave things as they are, partly 
because I expect the "fix" would actually be quite messy.  I view malloc 
options as pretty much exclusively a devel/debug aid, rather than a 
stable API.  In order to truly fix this API, I think we'd have to go 
further than what you suggest.  The _malloc_options global variable is 
harder to deal with than MALLOC_OPTIONS or /etc/malloc.conf when trying 
to design a fix.

The problem for _malloc_options is that we can't reliably use a 
function-based API, since malloc may be called before the application 
has a chance to configure the allocator.  As such, there is no direct 
way for an app to respond to malloc option incompatibilities.

It would be possible to construct a more complex interface for malloc 
configuration that would mostly solve these issues, but I don't think 
it's worth the added complexity for a devel/debug facility.

Jason



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4415DB8F.6060409>