[Xcb] [SRM] proposed upload of libxcb to lenny

[Ian Osgood]

On May 27, 2009, at 11:32 AM, Julien Cristau wrote:

> Hi,
>
> some bugs in libxcb cause Xlib performance regressions in lenny.   
> Those
> seem to hit ltsp users in particular pretty badly, so I'd like to fix
> this in a stable update.
>
> Proposed debdiff follows, with the following changes:
> - use a 16k buffer instead of 4k; this is maybe not strictly  
> necessary,
>   but would make lenny's version consistent with both traditional Xlib
>   (as used in etch and before) and recent libxcb (>= 1.2)

...

> --- libxcb-1.1.orig/src/xcbint.h
> +++ libxcb-1.1/src/xcbint.h
> @@ -72,7 +72,7 @@
>      pthread_cond_t cond;
>      int writing;
>
> -    char queue[4096];
> +    char queue[16384];
>      int queue_len;
>
>      unsigned int request;

I'm surprised this is a raw literal. Shouldn't this be a symbol,  
possibly configurable?

I think Jamey and Bart originally wanted a small value to make the  
footprint smaller on memory-limited devices. The natural solution in  
my eyes is to make this a constant having a 16K default so as not to  
surprise Xlib users, and have a configure option to build libxcb with  
a different sized buffer if desired.

Ian