Re: Rethinking octave_idx_type

[John W. Eaton]

Handling of --enable-64 in the configure script will change slightly:

  Native build:

    * with --enable-64: test that installed BLAS library we are using 
is built to use 64-bit integers.  Fail if it is not.  Check to see what 
options (if any) are needed so that the default Fortran INTEGER type is 
8 bytes.

    * without --enable-64: test to determine the size of integers used 
by the BLAS library that we are using and use that for our F77_INT type.

  Cross build:

    * with --enable-64: trust that using 64-bit integers is the right 
thing to use for our F77_INT type.  Choose default values for the 
Fortran compiler options to set the default INTEGER size to 8 bytes.

    * without --enable-64: trust that 32-bit integers is the right 
thing to use for our F77_INT type.

Existing user code that interfaces to Fortran code may need to be 
changed to work properly.  But as of the current release, we recommend 
using F77_INT instead of octave_idx_type in declarations of Fortran 
interfaces and people passing integer arrays would already need to use 
some care to ensure that their libraries are compiled with the correct 
options to use either 32- or 64-bit INTEGER types.

Other than fixing our own interfaces to Fortran code and checking sizes 
where it matters, is there anything else we need to do?

jwe