[Emacs-diffs] master 1d2df2f: Improve bignum comparison

[Emacs-diffs] master 1d2df2f: Improve bignum comparison

[Eli Zaretskii]

This commit does this:

    * src/data.c (isnan): Remove, as we can assume C99.

We indeed assume a C99 compliant compiler, but not necessarily a C99
compliant library.  Does the above assumption still valid if the
library is not C99-compliant?

Re: [Emacs-diffs] master 1d2df2f: Improve bignum comparison

[Paul Eggert]

Eli Zaretskii wrote:
> This commit does this:
> 
>      * src/data.c (isnan): Remove, as we can assume C99.
> 
> We indeed assume a C99 compliant compiler, but not necessarily a C99
> compliant library.  Does the above assumption still valid if the
> library is not C99-compliant?

Other parts of Emacs have been assuming isnan since July, so this should be safe 
here.

In practice isnan support is universal nowadays in the non-Microsoft world, and 
even MSVC has it, as I understand it. If there are problems (when compiling for 
MS-Windows 95, say?) we can fix them by doing something like adding "#define 
isnan _isnan" in src/w32.h. It's better to isolate portability hacks to 
platform-specific code, when the problem is platform-specific.

Re: [Emacs-diffs] master 1d2df2f: Improve bignum comparison

[Eli Zaretskii]

> Cc: emacs-devel@gnu.org
> From: Paul Eggert <eggert@cs.ucla.edu>
> Date: Wed, 22 Aug 2018 08:06:34 -0700
> 
> In practice isnan support is universal nowadays in the non-Microsoft world, and 
> even MSVC has it, as I understand it. If there are problems (when compiling for 
> MS-Windows 95, say?) we can fix them by doing something like adding "#define 
> isnan _isnan" in src/w32.h. It's better to isolate portability hacks to 
> platform-specific code, when the problem is platform-specific.

I wasn't worried about MSVC or Windows (there are no problems there),
I was worried about a more general assumption of a C99 library.  I
didn't think we assume that yet.

Re: [Emacs-diffs] master 1d2df2f: Improve bignum comparison

[Paul Eggert]

Eli Zaretskii wrote:
> I wasn't worried about MSVC or Windows (there are no problems there),

Good to know.

> I was worried about a more general assumption of a C99 library.  I
> didn't think we assume that yet.

Although we don't assume full C99 support in library functions, we do assume 
some level of C99 compatibility and isnan should be in the safe camp.

The same is true for the language proper. We don't assume full support for C99 
varying-length arrays, for example, and this is independent of the library.