Calling bt_split_name with a name that has too many
commas swaps spaces for the extra commas as expected,
but then afterward, bt_postprocess_string is called to
collapse those extra spaces to make it a well-formed name.
The bug is that when bt_postprocess_string is called, the
constant 1 is passed as the flag, and that doesn't do
anything. On just the wrong string, it'll dump core. The
appropriate thing is to send in BTO_COLLAPSE, and then it
works fine on all my test data.
Here's a diff of my names.c:
324,325c324
< bt_postprocess_string(name,BTO_COLLAPSE);
<
---
> bt_postprocess_string (name, 1);
and here's a single entry that should crash the current
version of dumpnames:
@misc{usingPAPILinux,
Author = {Jack Dongarra, Kevin London, Shirley
Moore, Philip Mucci and Dan
iel Terpstra},
Keywords = {PAPI},
Local-Url = {/Users/mike/Documents/hpcl/Papers/
linux-rev2001.pdf},
Month = {July},
Title = {Using PAPI for hardware performance
monitoring on Linux systems},
Url = {http://icl.cs.utk.edu/projects/papi/documents/
pub-papers/2001/linux
-rev2001.pdf},
Year = {2001}}
