I see no reason not to change the order if it improves performance.
Unless I am reading the code incorrectly, the filter is indeed only
applied once. What *is* done for each filter is the comparison with the
data (this may just be semantics). I do not see any functional
difference if you do the change, though.
I just did a quick test and it looks fine. The messages in colors.c
need to be changed to reflect the new processing of the list. I should
probably get rid of the ifdef debuggery in file.c, as well. I can take
care of it Monday, or if you feel inclined, Gilbert, you can do it.
All I did to test it was add a break:
Index: file.c
===================================================================
RCS file: /cvsroot/ethereal/file.c,v
retrieving revision 1.84
diff -C5 -r1.84 file.c
*** file.c 1999/08/28 01:51:57 1.84
--- file.c 1999/08/28 22:29:16
***************
*** 506,515 ****
--- 506,516 ----
cf->pd)){
color = crow;
#ifdef DEBUG_COLOR_FILTERS
fprintf(stderr,"yes\n");
#endif
+ break;
}
#ifdef DEBUG_COLOR_FILTERS
else {
fprintf(stderr,"no\n") ;
fflush(stderr);
--john
Gilbert Ramirez wrote:
>
> On Thu, Aug 26, 1999 at 04:19:33PM -0500, Jeff Jahr wrote:
> >
> > On the subject of performance, I believe [1] that the color filter
> > code is set up to always try and apply each color filter in the list to
> > each packet being decoded.
> >
> > Changing that function to 'the first matching color filter in list
> > is the one applied' from 'all matching color filters are applied in order'
> > would be good optimization with no loss of functionality...
>
> Yes, that's true. John, would we lose any functionality if we stopped
> applying the color display filters when we first find a matching filter?
> It would mean that the user would have to sort his color filters in reverse
> order to how they are now. That is, any filters matching on a high-level
> protocol should come first, and filters matching on the lower layers should
> come last:
>
> FTP = red on white
> TCP = blue on white
> IP = white on read
> TRMAC = white on blue
>
> Would the colorization still provide what it provides now?
>
> --gilbert
--
John McDermott jjm@xxxxxxxxxx
Writer and Computer Consultant
J-K International, Ltd.
+1 505/377-6293 - V
+1 505/377-6313 - F
