Re: [xml] too many mallocs?

[Daniel Veillard <veillard redhat com>]

On Mon, May 20, 2002 at 12:30:28PM +0200, Peter Jacobi wrote:
> Daniel, All,
>
> In the process of better understanding the inner
> workings of libxml2, and noting that simple parsing
> is bound by malloc in my tests, I came across
> following factoid:
>
> Going from the input file
>
> <foo><bar><xxx>XXXXXX</xxx><yyy>YYYYY</yyy></bar></foo>
>
> to
>
> <foo><bar><xxx>XXXXXX</xxx><yyy>YYYYY</yyy></bar><bar><xxx>XXXXXX</xxx><yyy>YYYYY</yyy></bar></foo>
>
> The number of mallocs used by libxml2 raises by 44 (from 120 to 164).
> This seems a bit wasted for two new start tags, two new end tags and
> two new char nodes.

  okay, I take patches as long as they preserve:
    1/ the portability
    2/ the conformance to the XML spec and the results of the
       various regression tests
       

> And there are 37 mallocs of size 3944 in case 1, compared with
> 62 mallocs of size 3900 in case 2, most likely from xmlInputBufferGrow.

  I think Garry Pennington looked at this recently.

> Is something broken with my libxml2? I compiled 2.4.21 with 
> all features off. (Using Watcom C 11.0b under WNT4.0).
>
> Is gcc's malloc such a fine piece of art, that you don't see
> negative effects from these large number of mallocs?

  Gcc does not provide malloc, this is a libc and hence system dependant
operation. If you know how to make the same result without going through
the same number of allocations, please feel free to share your knowledge,
I will definitely take patches !

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/