From: David M. C. <co...@ph...> - 2006-06-15 05:44:58
|
On Thu, 15 Jun 2006 15:22:57 +1000 (EST) Damien Miller <dj...@mi...> wrote: > Hi, > > I'm trying to make an OpenBSD package on numpy-0.9.5, but it receives a > malloc fault in the check_types() self-test as it tries to free() a junk > pointer. In case you are not aware, OpenBSD's malloc() implementation > does a fair bit of randomisation that makes it (deliberately) sensitive > to memory management errors. > > Instumenting the check_types test and scalartypes.inc.src's > gen_dealloc() and gen_alloc() functions I noticed that the error occurs > up after calling gen_dealloc() on a complex128scalar that was created as > check_types's "valb" variable as it is GC'd. > > The check_types tests work fine on the complex64scalar type and all > the other preceeding types. I'm not familiar with the guts of numpy > at all (and I can't even find the declaration of the complex128scalar > type in the source). What difference between complex64scalar and > complex128scalar should I look for to debug this further? Can you update to the latest svn? We may have fixed it already: valgrind is showing up nothing for me. A complex128scalar is a complex number made up of doubles (float64); a complex64 is one of floats (float32). -- |>|\/|< /--------------------------------------------------------------------------\ |David M. Cooke http://arbutus.physics.mcmaster.ca/dmc/ |co...@ph... |