Erratic behavior of TclBC in parallel

From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Thu Jan 21 2010 - 08:38:35 CST

Hi everyone,

I am seeing random errors from the Tcl interpreter(s) when running
TclBC in parallel.

Using the current CVS version of NAMD, on Linux-x86_64-g++ (gcc
4.4.1), charm++ 6.1.2 (arch: multicore-linux64). The Tcl8.3 static lib
is the x86_64 build provided on the NAMD website.

When running on one node, things work as expected. In parallel though
(on any number of cores from a dual 4-core box), I get erratic errors
such as:

FATAL ERROR: unknown math function "/"
    while compiling
"expr {$F * $x / $r}"

or

Tcl_Release couldn't find reference for 0x2a66310
Aborted

or

FATAL ERROR: can't use empty string as operand of "*"
    while executing
"expr {$x*$x + $y*$y}"

where x and y are local variables in proc calcforces.

This seems to point to a seriously messed up Tcl interpreter. Did
anyone notice anything of the sort? I'll gladly provide anyone willing
to try with a small test case.

Thanks,
Jerome

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:55:21 CST