|
From: Julian S. <js...@ac...> - 2008-01-05 16:50:53
|
On Saturday 05 January 2008 14:09, Nicholas Nethercote wrote: > On Sat, 5 Jan 2008, Bart Van Assche wrote: > > I'm not entirely happy with hardcoding the stack depth either. Please > > note that the problem of hardcoded stack depths does not only exist in > > drd but also in other Valgrind tools (helgrind/hg_main.c, > > msm__show_state_change()). Other tools (exp-omega) use > > VG_(clo_backtrace_size) as stack depth by including coregrind header > > files, which should be avoided. > > Do you think it would be a good idea to use the stack depth specified > > via the --num-callers=... command line option when printing a call > > stack ? Unfortunately this value (VG_(clo_backtrace_size)) is not > > available to tools. One way to overcome this is to define a special > > value for n_ips that selects the stack depth specified via > > --num-callers=..., e.g. as follows: > > There are a couple of options that are tool-visible, eg. -v and --xml. It > would be reasonable to add --num-callers to that list if multiple tools > need it, and assuming using the --num-callers numbers is reasonable -- eg. > Massif has a --depth parameter which is similar, but different because > reusing --num-callers didn't feel appropriate. It's not a clear-cut > decision either way... using --num-callers is probably ok for DRD and > Helgrind. So it seems to me the simple thing to do is move VG_(clo_backtrace_size) from the core to the tool header file, yes? I'll do that shortly. J |