Re: [xml] win32 thread safety patch

[Igor Zlatkovic <igor stud fh-frankfurt de>]

Hello everyone.

On Mon, 7 Jan 2002, Daniel Veillard wrote:

>   Well I can't comment on the Windows API specific aspects of your code,
> looking at the thread.c patch this seems clean. It seems the activation
> of the code needs to get a couple of defines set in the xmlwin32version.h
> header, like HAVE_WIN32_THREADS and LIBXML_THREADS_ENABLED, this is minor.

Aye, this is not really a headache :-)

>   I'm tempted to include your code on the basis that it looks clean and that
> adding it is probably the best way to get fixes fast if there is need for
> fixes :-)

Please wait for a day. Serguei, no offense meant, it seems to me that the
patch is wrong. You have put #ifdef _MSC_VER instead of #ifndef on at least
two places. This is a conclusion drawn from the first glance, not from the
actual compile. I don't have Windows machine available at the moment, I'll
check this later today.

>   I'm also wondering if this should be activated by default, we decided not
> to do it on the Unix compilation, but if there is no noticeable overhead
> on Windows this might be a good idea.

Well, this we shall find out quite soon. When I build the next binary, I'll
profile the parser with and without threading. I think about doing the
profile on Solaris on the same machine as well, just out of curiosity about
the results which would be directly comparable given the hardware is
exactly the same.

Now, I would prefer pthread implementation to native one. I have nothing
against Serguei's patch, I just have nighmare visions of the source which
someday contains more preprocessor directives than C code, and that just for
a performance gain which converges against zero on today's processors.
Arguing about this would however cause another flame war, so I'll give in
right here. I'll take the blue pill this time :-) :-)

Ciao
Igor