[Libreoffice] Helgrind results for LO

Julian Seward jseward at acm.org
Wed Apr 6 03:53:19 PDT 2011

> gtk a11y stack, while the initial helgrind ones end on in gdbusprivate.c
> etc.

The reports of races are not going to be accurate unless Helgrind sees
all the sync events done by the lower level libraries.  If they just use
the pthread_ primitives that's fine, but if they roll their own locks
etc then there will be noise until such time as the libraries are annotated
to tell Helgrind what's going on.  It's a not-insubstantial project.

I see also you're not using the Helgrind dev branch.  If you chase this
more, it's worth using, since:

* it shows the locks held by each thread in a race

* it shows all 4 stacks involved in a lock order violation, not 3

* output format is easier to read

* somewhat faster(*) and less memory-eating

Branch = svn://svn.valgrind.org/valgrind/branches/HGDEV2


(*) as in "now = as fast as treacle flowing uphill";
"before = as fast as treacle flowing uphill in a Canadian midwinter"

More information about the LibreOffice mailing list