Generating Backtrace

Backtrace

When reporting problems with the model it is often useful to obtain a backtrace identifying the source code location of an error. If the model terminated via an internal abort command (an FLAbort) then a backtrace is output to standard error:

 *** FLUIDITY ERROR ***
 Source location: (State.F90, 1528)
 Error message: Density is not a field name in this state
 Backtrace will follow if it is available:
 ../../bin/fluidity(fprint_backtrace_+0x1f) [0x81a7bb5]
 ../../bin/fluidity(__fldebug_MOD_flabort_pinpoint+0x2e0) [0x816d575]
 ../../bin/fluidity(__state_module_MOD_extract_from_one_scalar_field+0xb23) [0x82244
 ../../bin/fluidity(__momentum_equation_MOD_solve_momentum+0x3983) [0x8779c1c]
 ../../bin/fluidity(__momentum_equation_MOD_momentum_loop+0x970) [0x8785bdc]
 ../../bin/fluidity(__fluids_module_MOD_fluids+0xf1f4) [0x819b7fc]
 ../../bin/fluidity(mainfl_+0x155) [0x816cf7d]
 ../../bin/fluidity(main+0x301) [0x8161931]
 /lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xe5) [0xb5c57685]
 ../../bin/fluidity [0x8161541]
 Use addr2line -e <binary> <address> to decipher.
 Error is terminal.

Line numbers for each call in the backtrace can be obtained via:

 addr2line -e fluidity address

where address is one of the addresses in square brackets at the end of a line of the backtrace. If the model aborts via some other error, such as a Fortran run-time error or a segmentation fault, then a backtrace can be obtained with gdb. First make sure you have compiled with debugging, enabled at configure time with the ''--enable-debugging'' argument. Then run Fluidity in the debugger:

 gdb fluidity

Run the debugger command “break exit”:

 (gdb) break exit
 Function "exit" not defined.
 Make breakpoint pending on future shared library load? (y or [n]) y
 Breakpoint 1 (exit) pending.

to abort on any exit calls, and run the model with the required options:

 (gdb) r arguments

where arguments are the arguments passed to Fluidity. If the debugger breaks due to an exit command or segmentation fault you can retrieve a backtrace with:

 (gdb) bt

To obtain a backtrace for a parallel simulation you must launch gdb in parallel, e.g.:

 mpiexec -n 4 xterm -e gdb fluidity

This will launch multiple instances of the debugger. Run the commands above in each instance to retrieve a backtrace for each process.