About gmp and mpfr dependencies for gdb building

About gmp and mpfr dependencies for gdb building

[lijiang via Gdb]

Hi,

For the gmp and mpfr dependency libs, is it possible to provide an option
to enable/disable the gmp and mpfr instead of forcely enabling them when
building gdb?

Sometimes, we would like to cross compile the gdb in the crash-utility, but
we did not run into any issues before upgrading to gdb 16.2.

BTW: crash-utility used gdb-10.2 before upgrading to gdb-16.2

Any ideas or suggestions?


Thanks
Lianbo

Re: About gmp and mpfr dependencies for gdb building

[Iru Cai via Gdb]

On 3/13/25 16:14, lijiang via Gdb wrote:
> Hi,
>
> For the gmp and mpfr dependency libs, is it possible to provide an option
> to enable/disable the gmp and mpfr instead of forcely enabling them when
> building gdb?
>
> Sometimes, we would like to cross compile the gdb in the crash-utility, but
> we did not run into any issues before upgrading to gdb 16.2.
>
> BTW: crash-utility used gdb-10.2 before upgrading to gdb-16.2
>
> Any ideas or suggestions?
>
>
> Thanks
> Lianbo

Hi Lianbo,

It seems that gdb cannot be built without gmp and mpfr. You can download 
the gmp and mpfr tarball and unpack them to the binutils-gdb source 
tree, rename gmp-<version>, mpfr-<version> to gmp, mpfr, then building 
gdb will also build the bundled gmp and mpfr, just like the gcc build 
system.

Iru

Re: About gmp and mpfr dependencies for gdb building

[Tom Tromey]

> For the gmp and mpfr dependency libs, is it possible to provide an
> option to enable/disable the gmp and mpfr instead of forcely enabling
> them when building gdb?

Not currently.

You could probably try to implement it.  One way might be to replace the
use of GMP with a custom wide-int implementation; or with (assuming
IIRC) the slower single-header GMP implementation.  Not sure about MPFR.

Probably use of wide-int in gdb is just going to grow due to 128-bit
ints becoming more common.

thanks,
Tom

Re: About gmp and mpfr dependencies for gdb building

[Andrew Pinski via Gdb]

On Thu, Mar 13, 2025 at 1:31 PM Tom Tromey <tom@tromey.com> wrote:
>
> > For the gmp and mpfr dependency libs, is it possible to provide an
> > option to enable/disable the gmp and mpfr instead of forcely enabling
> > them when building gdb?
>
> Not currently.
>
> You could probably try to implement it.  One way might be to replace the
> use of GMP with a custom wide-int implementation; or with (assuming
> IIRC) the slower single-header GMP implementation.  Not sure about MPFR.
>
> Probably use of wide-int in gdb is just going to grow due to 128-bit
> ints becoming more common.

Well and BitInt increases the wide-int even further past the 128bit
wide length too.
GCC has its own wide-int implementation which might be useful to pull
into GDB if needed. It is used for BitInt and wide integer too.

MPFR might be hard to replace for floating point types are hard to
emulate exactly.
GCC does have some code to do some but most of the folding (except for
simple things) are handled via MPFR also.

Thanks,
Andrew Pinski

>
> thanks,
> Tom

Re: About gmp and mpfr dependencies for gdb building

[Tom Tromey]

Andrew> GCC has its own wide-int implementation which might be useful to pull
Andrew> into GDB if needed. It is used for BitInt and wide integer too.

FWIW I tried this once but it was a pain, and also the API seemed pretty
gross.  I don't remember the details though.

Tom

Re: About gmp and mpfr dependencies for gdb building

[lijiang via Gdb]

On Fri, Mar 14, 2025 at 6:09 AM Tom Tromey <tom@tromey.com> wrote:

> Andrew> GCC has its own wide-int implementation which might be useful to
> pull
> Andrew> into GDB if needed. It is used for BitInt and wide integer too.
>
> FWIW I tried this once but it was a pain, and also the API seemed pretty
> gross.  I don't remember the details though.
>

Thank you so much, Tom and Andrew.

I tried to cross compile the gmp and mpfr for the crash-utility, and
succeeded.

https://gcc.gnu.org/pub/gcc/infrastructure/gmp-6.2.1.tar.bz2
https://gcc.gnu.org/pub/gcc/infrastructure/mpfr-4.1.0.tar.bz2

Looks like this is easier than implementing an option to enable/disable gmp
and mpfr. :-)

Thanks
Lianbo


>
> Tom
>
>