Only ~100 valgrind errors left?

[Dan Kegel]

I'm getting back to valgrinding again.
In preparation for adding valgrind runs to patchwatcher,
I'm running the test suite through valgrind zillions
of times, looking at how repeatable the results are.
At the moment it looks like there is an annoyingly
long tail of less likely failures, which is going to
be a pain.

On the bright side, it is looking like there aren't
that many frequent valgrind errors left.  See
http://kegel.com/wine/valgrind/2008-11-13.txt
for a manually deduplicated list.  It only shows
about 100 problems.  Here's the breakdown by
error type:
      1 InvalidJump
      3 InvalidWrite
      7 Leak_DefinitelyLost
      7 UninitValue
     11 SyscallParam
     25 InvalidRead
     52 UninitCondition

A word of warning: don't go fixing the problems behind
these warnings unless you really know what you're
doing.  Superficial fixes are worse than no fix.
That said, if anyone has a really good grasp of
any of these problems, please go ahead and fix,
or let me know which of these should be suppressed.

For no particularly good reason, I've omiitted filenames,
the list only includes function names and line numbers.
That should have the unintended benefit of
discouraging people unfamiliar with the code from
jumping in and fixing things...

Thanks,
Dan