[21.4] Patch for consistent crash with Linux native sound on AMD64

[Stephen J. Turnbull]

Aidan Kehoe writes:

 >> The following patch fixes things for me and makes sense since prtn and rrtn
 >> are later recast as a size_t.

 > Yeah, there’s an unfortunate history with XEmacs and size_t, 

That has nothing to do with this, though.  The problem with size_t is
that it is a C library standard type that is diastrously badly
designed for use as a dimension type in user programs.

 > so a question: does changing the declaration to long work as well?

As long as the size_t data is solely used to communicate with external
code such as device drivers, it's best to use size_t explicitly (and
probably document it with a comment).  If it is going to get mixed
with XEmacs data, then it should be cast to EMACS_INT or perhaps
EMACS_UINT (the latter should be avoided, though).  Typically this
should be done in a wrapper function (inline if there's need for speed).