Re: [PATCH] Require full GMP for big integer module functions.

[Philipp Stephani]

Am Do., 25. Apr. 2019 um 23:00 Uhr schrieb Paul Eggert <address@hidden>:
>
> On 4/25/19 10:09 AM, Philipp Stephani wrote:
> > This allows mini-gmp to not stay binary-compatible with full GMP.
>
> Isn't this overkill? I can see many practical scenarios where using
> extract_big_integer etc. will work just fine even if Emacs is compiled
> with mini-gmp and the module with full GMP (or vice versa). I can't see
> many practical scenarios where these additional checks will cause more
> good than harm.

The problem with binary compatibility is that if emacs-mini-gmp.h ever
becomes incompatible, the behavior will silently become undefined,
leading to extremely subtle bugs. So even a bit of harm here is too
much.

>
> Would it help to write to the GMP maintainer and ask how compatible
> mini-gmp is intended to be? My impression is that compatiblity concerns
> are pretty high on his priority list.

For full GMP definitely. My understanding is that emacs-mini-gmp.h is
maintained as part of Emacs, so *we* need to make sure to never become
incompatible if we want to continue to support this use case.