ronnie sahlberg wrote:
I have checked in a fix. it should work now.
Thanks for the bugreport.
The problem is that proto_item_append_string() does not play well with
the TRY_TO_FAKE_THIS_ITEM() optimization.
Fortunately proto_item_append_string() is very rare, essentially
only called from WSP.
I have added a comment to proto_item_append_string() and how it
interacts (or fails to interact) with TRY_TO_FAKE_THIS_ITEM() and what
one needs to do if one uses this function.
I have added a change to WSP so that it will unconditionally disable
the optimization for WSP and all subdissectors of WSP.
Hmm no comment :) Well yes, it won't work for my code, I'm creating the
tree structure only once per run...
It won't solve the pb with xxx_set_len, xxx_get_len where ethereal could
set/get the size of a wrong item,
From a quick browsing, It seems that :
packet-quake.c is broken, packet-m2tp.c is ok because it's an _add_text
item.
I have no idea what ptvcursors (use by packet-ncp2222.inc) do.
What about
1) A new parameter to proto_register_protocol
PROTO_NO_OPTIMIZE /* default proto can */
2) use a new flag in struct field_info.flags
FI_something set each time TRY_TO_FAKE_THIS_ITEM() return
If it's set in get_len then THROW an error, saying the dissector need
PROTO_NO_OPTIMIZE
if it"s set in set_len THROW the same error. Currently most (all?)
set_len are on always allocated add_text item.
At least it won't fail silently.
Didier
