I'm running valgrind with the following switches:

valgrind --tool=memcheck --leak-check=full --leak-resolution=high --show-reachable=no --undef-value-errors=yes --track-origins=yes --num-callers=20 --child-silent-after-fork=yes --trace-children=yes --xml=yes --xml-file=program.%p.valgrind

Because my program forks, I get a new xml file with a different pid number. In the xml file, the Id field will start counting at 1 again, overlapping with other runs.
The valgrind plugin currently combines all the reports into one big report, and can't distinguish between id 1 from pid x and id 1 from pid y anymore, resulting in detail loss.

Grouping by <pid> field from the xml file will fix this problem.

Maybe for another ticket, but caused by the same single huge report in the plugin: having a per program overview first, and then for one program viewing all the errors grouped like they are now would probably be better.