[Guy Harris <guy@xxxxxxxxxxxx>]

On Nov 21, 2003, at 6:02 AM, didier wrote:

> Ok, I was thinking about all calls to proto_tree_add_text in 
> dissectors.
> cf attached patch for ip.

Those eliminate calls to "check_col()" (by caching the result of the 
first call), and eliminate calls to "proto_item_append_text()", but 
they don't eliminate any "sprintf()"/"snprintf()" calls - the 
"check_col()" calls would have prevented "col_add_fstr()" from being 
called, and "proto_item_append_text()" doesn't do any formatting if the 
"visible" flag isn't set on the tree ("proto_tree_add_text()" calls 
"proto_tree_set_representation()" even if "visible" isn't set, but 
"proto_tree_set_representation()" doesn't do any work if it's not set).

So it'd probably increase performance some, but not a lot - I'll try it 
at some point to see what the performance change is, but as it's only 
patching one dissector, it might not make a big difference, so I'll 
look at the other dissectors that would be used heavily in my big test 
capture.