prof core-dumping

[Fred Fish]

In article <252 at metaware.metaware.com> ken at metaware.UUCP (ken) writes:
>In article <1050 at tuewsd.lso.win.tue.nl> wsinpdb at lso.win.tue.nl (Paul de Bra) writes:
>>the application runs fine, but when I try to look at the profiling data
>>with prof, prof dumps core.
>
>I too have seen this. I (of course) use a MetaWare compiler on our very large
>and convoluted internal test suites and my prof will core dump. I'm now less
>likely to assume that I'm doing something wrong in the compiler and that prof
>is broken.

This may not help you unless you have access to source, and even then it
may not be the same problem.

First, look through the prof source until you find a function that examines
symbol table entries and keeps a count of the auxiliary entries for a 
particular symbol.  It keeps this count in a static variable and decrements
the variable on each count, doing an early return whenever the count is still
greater than zero (thus skipping aux entries when called with sequential
addresses).  Now, find where this function is called in the code.  You should
find one location where it is called with indices in REVERSE order, I.E.
something like "n=numsyms; while(--n) {func(n)}".  This causes is almost
guaranteed to cause some aux entries to be interpreted as non-aux entries.
Depending upon the actual values in the aux entries, bad things can happen...

Sorry I can't be more specific.

-Fred
-- 
# Fred Fish, 1835 E. Belmont Drive, Tempe, AZ 85284,  USA
# 1-602-491-0048           asuvax!{nud,mcdphx}!estinc!fnf