[mercury-users] Ever valground a mercury program?

[Nicholas Nethercote]

On Wed, 23 May 2007, Ondrej Bojar wrote:

> My rotd-2007-03-07 generates executables that make valgrind shout and shout 
> and shout. In fact, the main(io,io) does not start in valgrind at all and the 
> detstack overflows instead.
>
> Have any of you ever used valgrind on Mercury programs? Are there any special 
> caveats?

Mercury's garbage collector does lots of low-level nasty stuff that violate 
Valgrind's assumptions about memory behaviour.  For example, if a program 
wants to know if a page is addressable, it could set up a SIGSEGV/SIGBUS 
signal handler and write to it -- it it faults, then the page is not 
addressable.  But Valgrind assumes that programs won't do this kind of 
thing.

Also, because the GC is conservative it may look for pointers in memory that 
is uninitialised, which is reasonable in this case but again violates 
Valgrind's assumptions.  I'm guessing that's what the messages you got 
related to.

Unless there's a way to disable the garbage collector in Mercury, I don't 
think you'll have much luck using Valgrind with it.  Can you test the C code 
in a harness without hooking it up to Mercury?

The stack issue could be because Valgrind limits the stack to 8MB, if I 
recall correctly.

Nick
--------------------------------------------------------------------------
mercury-users mailing list
Post messages to:       mercury-users at csse.unimelb.edu.au
Administrative Queries: owner-mercury-users at csse.unimelb.edu.au
Subscriptions:          mercury-users-request at csse.unimelb.edu.au
--------------------------------------------------------------------------