[Nptltracetool-cvs] trace TODO,1.7,1.8

[Matthieu C. <ma...@us...>]

Update of /cvsroot/nptltracetool/trace
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv31072

Modified Files:
	TODO 
Log Message:
clean

Index: TODO
===================================================================
RCS file: /cvsroot/nptltracetool/trace/TODO,v
retrieving revision 1.7
retrieving revision 1.8
diff -u -d -r1.7 -r1.8
--- TODO	25 Jul 2005 15:09:59 -0000	1.7
+++ TODO	29 Jul 2005 13:47:44 -0000	1.8
@@ -1,21 +1,21 @@
-buffer indexes
-
-cleanning header
+poll of buffer
 
-extend the trace point coverage
+extend the trace point coverage : rw-lock, ...
 
 add a mecanisme to give a name to trace object (thread, ...)
 
-don't follow pid
+option for don't follow pid (like -f in strace)
 
 when we get the name of an object at the INIT, you should consider not only to register the adress of the object but some of its propriety. For example, a condvar INIT should keep adress of the mutex and the cond-var.
 Idem, manage correctly after a fork.
-Remove duplicate entry of trace point.
+
+Remove duplicate entry of trace point if possible (EVENT A and B use mutex count, then merge it).
 
 
-TRACE review : try to remove duplicate trace point : for example thread_state should be in lowlevellock routines, ...
-prefer C routine : for mutex set trace in __pthread_mutex_lock instead of in assembly...
+TRACE review : 
+  try to remove duplicate trace point : for example thread_state should be in lowlevellock routines, ...
+  /!\ with this we can't merge these events with others -> need to be studied.
 
-Manage a way to include it in libc and select it with configure option
+  prefer C routine : for mutex set trace in __pthread_mutex_lock instead of in assembly...
 
-Les futures versions 2.4 ont une archi pour auditer la biblioteques, LD_AUDIT. Voir si on ne peut pas l'exploiter pour diminuer l'intrusion dans nos patch dans la libc.
+Manage a way to include it in libc and select it with configure option