Tcl Source Code

View Ticket
Login
Ticket UUID: 1160057
Title: Seg fault from usleep
Type: Bug Created on: 2005-03-09 17:49:25
Submitter: s_a_white Assigned to: andreas_kupries
Subsystem: 49. Threading Severity:
Priority: 4 Last modified: 2026-10-12 13:47:05
Status: Pending Closed by: hobbs
Resolution: Works For Me Closed on: 2005-10-05 05:31:25
Version: obsolete: 8.4.9
Description:
Am trying to determine what may be going on in tcl that
causes this code to crash.  Using a main instead of tcl
to run the code it appears to work fine.

The computer is running MandrakeLinux 10.1:

kernel: 2.6.8.1-12mdksmp
gcc: 3.4.3-1mdk
glibc: 2.3.3-21mdk
tcl: 8.4.5-6mdk

Code:

#include <stdint.h>
#include <stdio.h>
#include <pthread.h>
#include <unistd.h>
#include <tcl.h>

class Message
{
public:
   Message();
   ~Message ();
};

Message::Message() {}
Message::~Message () {}

static void *RecallTaskEntry(void *_this)
{
   Message m;
   usleep(2000000);
   return 0;
}

extern "C" int createThread() {
  pthread_t Recallthread;
  pthread_create(&Recallthread,NULL,RecallTaskEntry, 0);
  usleep (50000);
  pthread_cancel (Recallthread);
  pthread_join   (Recallthread, NULL);
  return 0;
}

int TclCreateThread (ClientData clientData, Tcl_Interp
*interp,
                     int argc,CONST84 char *argv[])
{
  createThread ();
  return TCL_OK;
}

extern "C" int Thread_Init(Tcl_Interp *interp) {
   
Tcl_CreateCommand(interp,"createThread",TclCreateThread,
0, 0);
    return TCL_OK;
}

extern "C" int Thread_SafeInit(Tcl_Interp *interp) {
   return (Thread_Init(interp));
}

Code is C++ compiled as a shared library and run in tcl
via:

load libThread.so
createThread

The observed beviour is a segfault which results from
doing a pthread_cancel when the thread is in the
usleep.  The debugger output looks a little confused
like possible stack corruption.  However as explained
it works ok standalone outside of tcl.

Things that seem to make it work are:

- Removing the Message object from the thread.
- Removing Messages destructor.
- As a test using sleep and pause instead of usleep
appear to work.

Any ideas on what is causing this bug?
User Comments:
nobody added on 2005-03-11 05:42:19:
Logged In: NO 

Have updated tcl to 8.4.9 and rebuilt the code.  Ran it up
using tclsh (didn't update tk) and still the seg fault
appears.  Interestingly I tried valgrind on tclsh and get
various complaints about mallocs/free's, but the seg fault
disappeared, hmmm.

nobody added on 2005-03-13 17:20:44:
Logged In: NO 

Additional things I've tried that had no effect, and still
resulted in a segfault.

-Updating glibc to offical release 2.3.4
-Updating kernel to to official release 2.6.11.2

Haven't yet updated the compiler as no newer official releases.

s_a_white added on 2005-09-21 16:59:53:
Logged In: YES 
user_id=59929

This appears to be somehow exception related.  Compiling
with -fno-exceptions prevents the segfault.

hobbs added on 2005-10-05 12:31:25:
Logged In: YES 
user_id=72656

Is this a possible compiler bug?  Are you making sure to
compile Tcl with --enable-threads?  I'm not sure how this
really relates to Tcl, vs. being a C++ issue.

s_a_white added on 2005-10-05 14:48:18:
Logged In: YES 
user_id=59929

Am not sure...  The problem only started when moving to a
newer Linux distribution and as yet have not been able to
try a new one.  The compiler has gone from gcc 3.3 to 3.4
(last distribution used the old compiler).

I believe 3.4 is still consider "stable" and gcc 4 what to
start aiming for but may contain bugs,  The major thing I
note about this Linux distribution was the change to ntpl
which was causing various other problems for us (mainly how
thread cancelation was now being handled during cout.... now
exception rather than cancelation point, sem_wait can now
also return -1... not according to the docs).

To be honest I wasn't sure which list to ask TCL, glibc, C++
or gcc (note I've tried contacting glibc as they do the
nptl, but no response).  As noted the problem only appears
when run in TCL, rather than a standalone main.  As stated
running tcl under valgrind with the same test made the
problem disappear.

Am I running with threads enabled, probably not.  This is a
prebuilt version by the distribution.  All TCL does is make
calls to external functions that start/stop threads and
occasionally grab mutexs, but those threads do not access
any tcl functions or keep track of any tcl memory (objects).

hobbs added on 2005-10-05 15:05:06:
Logged In: YES 
user_id=72656

Reducing prio as I'm still suspicious that Tcl is at fault here.

serhiy.storchaka added on 2026-10-12 13:47:05:

I cannot reproduce this with 8.6.18 and 9.1 (glibc 2.43, g++ 15): the extension from the report, built against them, survives 20 cancellations of the thread during usleep() with the C++ destructor on its stack. The crash was most likely in the reported environment: a Tcl built without threads and a glibc whose pthread_cancel unwinds the stack and needs libpthread to be loaded at startup; a Tcl built without threads works with the current glibc. Tcl is threaded by default since 8.6, and glibc includes libpthread in libc since 2.34.