Re: FYI - RFC about making GMP a required dep for GDB

Re: FYI - RFC about making GMP a required dep for GDB

[Eli Zaretskii]

> Date: Sun, 11 Mar 2018 11:32:53 -0400
> From: Joel Brobecker <brobecker@adacore.com>
> 
> Hi Eli,
> 
> I just wanted to make sure you saw the following proposal, in case
> you wanted to provide feedback about it. No worries if you don't!

I missed that somehow, or maybe didn't think my input was important.
Thanks for the heads up.

> But if you do, could you send it on the list?  So far, I got one answer,
> which was positive.
> 
> https://www.sourceware.org/ml/gdb-patches/2018-03/msg00148.html

I have no problems with adding this dependency.  A GDB built with MPFR
already indirectly depends on GMP as well, so it sounds like that ship
has sailed already.  MinGW ports of GMP are available, so portability
to native MS-Windows should not be an issue.

(Btw, would it make sense to provide the features via MPFR?  Than we'd
have only one explicit dependency.  Not that it matters too much.)

HTH

Re: FYI - RFC about making GMP a required dep for GDB

[Joel Brobecker]

> I have no problems with adding this dependency.  A GDB built with MPFR
> already indirectly depends on GMP as well, so it sounds like that ship
> has sailed already.  MinGW ports of GMP are available, so portability
> to native MS-Windows should not be an issue.

Excellent news! :)

> (Btw, would it make sense to provide the features via MPFR?  Than we'd
> have only one explicit dependency.  Not that it matters too much.)

Unless I'm mistaken, the dependency on MPFR is currently optional.
I think Ulrich was able to do that by limiting as much as possible
the operations he wanted to perform, and by abstracting those out.
The abstraction layer kind of makes sense for floats, because there
is always going to be at least 2 kinds of floating points (binary
floating points, and decimal floating points, which are handled by
two different libraries). For GMP, it makes less sense, IMO.

That being said - I think the code would be slightly less complex
if we made both GMP and MPFR required, and this would ensure that
users always get the MPFR-capable version, which gives more accurate
results.

-- 
Joel