[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [MIT-Scheme-devel] Default floating-point traps

From: Chris Hanson
Subject: Re: [MIT-Scheme-devel] Default floating-point traps
Date: Sun, 21 Oct 2018 01:19:18 -0700

My argument for having them on by default is that, if your computation doesn't trap, that means it didn't overflow or underflow or something like that. For a naive user, who expects FP numbers to "just work" (that IS naive!) it's a great big flag that what they are doing is wrong.

But I don't feel very strongly about this.

On Sun, Oct 21, 2018 at 12:26 AM Taylor R Campbell <address@hidden> wrote:
I spend way too much time fighting with floating-point traps when I
practically never want them on by default, particularly when trying to
print (e.g.) subnormals at the REPL.

Does anyone care that they're on by default?  Can we disable them?
One can always detect them, or explicitly trap, in particular
computations anyway.

It would also be nice if, e.g., (exp -1000.) gave 0. and set the
inexact-result exception bit, rather than barfing -- that's something
the ucode does explicitly which the floating-point trap control has no
power over.

MIT-Scheme-devel mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]