|
From: Solomon, B. <be...@ug...> - 2008-03-24 23:42:21
|
I am not quite sure where I am supposed to post patches. This is the one removing the excessive checks though of course other options are possible (check every so often, have a command line switch etc.) I was using 3.3 but SVN seems the same. The 64 bit change I'll send separately if this is the right place to send it. Bernie -----Original Message----- From: Julian Seward [mailto:js...@ac...] Sent: Monday, March 17, 2008 6:25 PM To: val...@li... Cc: Solomon, Bernard Subject: Re: [Valgrind-users] Couple of questions > We have custom allocators so I used the mempool > interface as that matches our internals nicely - but I found the > check_mempool_sane > calls made things excessively slow so commented them out of alloc and > free which works > fine for me. These checks seemed overly paranoid for a well tested > allocator > to me and was wondering why they are there all the time (at least after > my > integration of the custom allocator is tested and working for us). I don't know. > The second issue is the user requests like VALGRIND_CHECK_MEM_IS_DEFINED > are documented to return the address of the first bad byte - yet > internally > they use a unsigned int type is which does not seem correct on 64 bit > machines > I would have expected that to be unsigned long or perhaps void *. That sounds like a straightforward 32-vs-64-bit bug. Can you fix both of these in a way that seems right to you, and send along a patch? Also, what version of V was this? J |