Ulf Lamping wrote:
Where I'm now lacking knowledge of is the best way to implement the pipe
related stuff for the unix side - I need some help here. The current
implementation uses a dummy function which runs into a g_assert(),
therefore the buildbot test step and thark capturing fails now :-(
The problem here is how to read from the pipe that transfers messages
from dumpcap to tshark (messages like: "10 new packets arrived"). The
actual message handling is no big deal, the problem is to read mechansim
(block/poll) on the pipes file descriptor.
The unix specific implementation in Wireshark (gtk/gui_utils.c) uses
gtk_input_add_full(), but as we don't want to include GTK/GDK with
tshark, we need a different way to handle this.
The Win32 implementation that I've done in tshark.c uses 200ms polling
and Win32 specific PeekNamedPipe/GetExitCodeProcess and GStaticMutex to
deal with this - this is basically the same implementation as in
Wiresharks gui_utils.c (except for the mutex, which is new to prevent
reentrance problems due to changed timer mechanism).
So what we need now is something like this on the unix side. tshark.c
line 1682 contains already a function pipe_timer_cb() that will be
called every 200ms and currently only contains a g_assert() - but this
is only a thought. Would a select() call on the pipe_input.source file
descriptor do the trick here?
Could someone with more knowledge about this stuff could have a look at
this?
First, thanks for doing it! :-)
Second: I don't know much about it but I just committed something that
works for me.
It seems (based on my somewhat limited testing) that we don't need to
worry about blocking in tshark so the *NIX side is now just blocking
trying to read from the child. Could the Windows side not do the same?
Could other *NIX users test tshark to see if it works (I'll see in a
while if the buildbots are happy about it)?
