Re: [Numpy-discussion] Please test the SVN branch

[Stefan v. d. W. <st...@su...>]

On Thu, Oct 12, 2006 at 12:38:51AM -0600, Travis Oliphant wrote:
>=20
> I made some fixes to the "asbuffer" code which let me feel better about=
=20
> exposing it in NumPy (where it is now named int_asbuffer).
>=20
> This code takes a Python integer and a size and returns a buffer object=
=20
> that points to that memory.  A little test is performed by trying to=20
> read (and possibly write if a writeable buffer is requested) to the=20
> first and last elements of the buffer.  Any segfault is trapped and use=
d=20
> to raise a Python error indicating you can't use that area of memory.=20
>=20
> It doesn't guarantee you won't shoot yourself, but it does make it more=
=20
> difficult to segfault Python.  Previously a simple int_asbuffer(3423423=
,=20
> 5) would have segfaulted (unless by chance you the memory area 3423423=20
> happens to be owned by Python).

Some argument checking and a docstring would be helpful.  I guessed
wrong the first time:

In [1]: x =3D N.array([1,2,3])

In [4]: y =3D N.int_asbuffer(x.size,x.__array_interface__['data'][0])

In [5]: y[0]

Segmentation fault

Out of curiosity, why does the following outputs differ:

In [39]: y =3D N.int_asbuffer(0L,x.size)
-------------------------------------------------------------------------=
--
exceptions.SystemError                               Traceback (most rece=
nt call last)

/home/stefan/<ipython console>=20

SystemError: error return without exception set

In [40]: y =3D N.int_asbuffer(1L,x.size)
-------------------------------------------------------------------------=
--
exceptions.ValueError                                Traceback (most rece=
nt call last)

/home/stefan/<ipython console>=20

ValueError: cannot use memory location as a buffer.

Thanks for your work on this.

Cheers
St=E9fan