On Mon, 3 Nov 2003, Brad Cavanagh wrote:
> This leads me to a question about PRIMDAT. Whenever there's
> platform-specific code pulled in I use the ix86_Linux variant. It seems to
> work well. When I run the tests for PRIMDAT it says, after testing the
> divide by zero, "Error: NUM_TRAP error handler not responding." Sure
> enough, the comment at the top of num_han.c_ix86_Linux says that all of
> the error handling functions are supposed to do nothing. Presumably this
> isn't something to worry about, is it? Is this another one of those cases
> where I can ignore the failing test (like for TCLSYS)?
Hi Brad,
yes it's probably easiest to just let this one go.
Unless there is a numeric handler system in BSD/OS X this means we'll see
as many instances of the various NaNs in our data as are seen under Linux
(if you're interested see the sun4_Solaris variants for an OS that does
the right thing, but much of this is rooted back in the days of VAX/VMS).
Peter.
|