|Anonymous | Login||2021-03-07 17:47 UTC|
|My View | View Issues | Change Log | Roadmap|
|View Issue Details|
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0000363||OMNeT++||simulation kernel||public||2011-06-24 13:33||2011-06-24 13:33|
|Target Version||Fixed in Version|
|Summary||0000363: EventlogFileManager incorrectly assumes that sputn always takes a whole line terminated with '\n' as argument|
|Description||From: Jan Vogelgesang|
i discovered a possible bug in the way the omnet++ event log is written.
I did some investigation and came up with a working patch. But i'am
afraid i didn't solve the root cause, only hide some of the bad effects.
For this reason i'have included my 'findings' in an extra section of
this mail, if anyone wants to continue investigation.
Caution: in-depth iostream-knowledge required ;-)
Version: OMNeT++ 4.2b2, compiled using gcc 4.4.3 on Ubuntu 10.04
Libraries used: Boost 1.45.0
+++ Description: +++
I'am using boost::iostreams::tee_device together with cEnvir's ostream
(ev.getOStream()). I include the actual code snippet this as , but
don't want to go in detail here, because i think the real problem is
unrelated to the boost classes, but related to the way the
The event log (.elog) written from the simulation contains extraneous
'-' characters at each line following an output send to ev.getOStream().
See following .elog snippet:
E # 8 t 1.1 m 6 ce 1 msg 10
- debug output from my application
- BS id 24 tid 23 eid 24 etid 23 c someMessage n someMessage
SH sm 6 sg 1572865 pd 0.1
This .elog file is clearly defective.
+++ Findings +++
I did some investigation, but i'am not an expert in iostream-stuff, so
please read the following with care:
The std::ostream and the boost::iostreams::stream seems to differs in
the way they communicate with the underlaying streambuf when
The boost::iostreams::stream class is issuing the whole string
(including the trailing '\n') at once through
std::basic_streambuf<>::sputn(), which is calling evbuf::xsputn() (see
cenvir.cc:43). This function calls sync(), which eventually results in a
call to EventLogWriter::recordLogLine(...) (see eventlogwriter.cc:25).
Then boost::iostream is calling std::basic_streambuf<>::pubsync() - to
me this seems reasonable beheavior when handling a endl. pubsync() calls
evbuf::sync() and finally EventLogWriter::recordLogLine(...) is called
again, with parameter n==0.
EventLogWriter::recordLogLine(...) then outputs a leading '-' to the
next line, even if it is called with an empty buffer (n==0).
The std::ostream seems to use sputc(...) to output the trailing '\n' -
thus evbuf::xsputn() isn't called and don't issue an extra sync().
Suggestion: place a breakpoint at eventlogwriter.cc:27 when you start to
I assume a sync() on an empty buffer is supposed to do nothing. And
recordLogLine should not be called on each sync().
+++ Solution +++
For me the following simple patch works:
--- omnetpp-4.1-org/src/envir/eventlogwriter.pl 2010-06-11
+++ omnetpp-4.1/src/envir/eventlogwriter.pl 2011-06-23
@@ -188,6 +188,9 @@
void EventLogWriter::recordLogLine(FILE *f, const char *s, int n)
- CHECK(fprintf(f, \"- \"));
- CHECK(fwrite(s, 1, n, f));
+ if(n > 0)
+ CHECK(fprintf(f, \"- \"));
+ CHECK(fwrite(s, 1, n, f));
However, i'am afraid this patch doesn't solve the root cause.
In my opinion one should rethink the way recordLogLine is called - at
the moment it is called on each sync(). Thus additional calls to sync()
will do harm. It seems better to call it only on a complete line.
Hope this helps !
the following are just a condensed code snippet, part
of a complete omnet++ simulation. I think it should work in almost any
oment++ simulation application.
typedef boost::iostreams::stream<TeeDevice> TeeStream;
//instantiate global output stream to cout and omnet ev
//somewhere in a function
debs << "debug output from my application" << endl;
|Tags||No tags attached.|
|2011-06-24 13:33||levy||New Issue|
|2011-06-24 13:33||levy||Assigned To||=> levy|
|Copyright © 2000 - 2021 MantisBT Team|