Re: [xml] too many mallocs?

[Daniel Veillard <veillard redhat com>]

On Mon, May 20, 2002 at 09:48:58PM +0200, Peter Jacobi wrote:
> Daniel, All,
>
> I didn't want to be negative, or criticize for
> its own sake.
>
> I'm just wondering about the stats I get. And asking,
> if you're aware of the situation. 

  yes, that's something I'm aware of for a long time.

> It may be a niche case to see malloc dominate, it's
> most pronounced on a document without attributes but
> with many tags, most of them empty, it's about 330k
>
> On a rather mediocre WNT system, my xmllint needs 
> 260ms per iteration (Igor's MSFT version needs 410ms,
> at least I can outperform this!). Of these 260ms, over
> 100ms are spent in malloc and free, and another large
> part in related activities (_stos amd xmlStrndup).

  That would be around 1MByte/s, no that's not good.
On my Duron 1.2MHz parsing building a tree is at 6MB/s
(not validating).

> Parsing the libxml2 API document (490k) is slower, showing
> that attributes are more expensive than elements, using 400ms
> (690ms) with about the same absolute time spent in malloc
> and friends.

  Attribute normalization and such introduce more copies in
attributes, yes.

> I'll try to understand better, what's going on, and maybe
> there is a solution.

  Well some people suggested dropping xmlMalloc/xmlFree
to use alloca() in some context where the result is discarded 
anyway, but I'm not too fond of this.

Daniel

-- 
Daniel Veillard      | Red Hat Network https://rhn.redhat.com/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/