Profiling trouble

Simon Marlow
Thu, 30 Jan 2003 18:51:36 -0000

> something interesting turned up again.  Setting cut=3D8 in
> Show2.hs and running it with ./show +RTS -h -i0.01 gives
> show: fatal error: main thread has been GC'd
> It may be silly, since 0.01 < 1/50, but this error message
> isn't too helpful.  Btw, +RTS -? says:
>   -i<msec>       Time between heap samples (msec, default: 100)
> which is wrong (time is given in seconds really).

Both of these bugs were fixed in 5.04.2.

> One more question: is there a way not to truncate the call
> stacks?  Ie in the hp file I see lines like
> (144)showData2/showData/ma...   12
> and I'd like to see showData2/showData/main or so.

I'm afraid there's no way at the moment, but I'll look into adding a
flag.  The problem is that the longer the stacks get, the less room
there is for the graph - hp2ps tends to squeeze the graph up to make
room for the labels, which is why we truncate them.

Your other problem (misattribution of costs in the profile) seems to be
related to the use of overloaded functions, and the compiler isn't
translating the definition of getStates properly.  I haven't found a fix