bf is small and easy to build; couple of sources and a header, which I had
no difficulty integrating. It also has a cleanly-designed api. The
optimization is not quite as good as gmp & co, but it is still
competitive, and it gets the asymptotics right.
It also has arbitrary-precision floating point support, which I would like
to integrate as well. (The other 'heavy-hitting' fp libraries--arb,
mpfr--depend on gmp.) Main concern there is that it is missing complex
numbers. But those can be faked, at the cost of a few extra roundings.
On Wed, 9 Mar 2022, Eric Iverson wrote:
I don't know about libbf, but I do have reservations about gmp. The first
step is to make sure the choice is right.
On Wed, Mar 9, 2022 at 7:15 PM Eric Iverson <[email protected]>
wrote:
There is no need to static link. A dynamic link is easier. The 'standard'
build of the dll would be distributed in the J bin folder and would work
just as would the static link. But if you wanted a different build, it is
easy to do.
On Wed, Mar 9, 2022 at 4:09 PM Elijah Stone <[email protected]> wrote:
jconsole/etc. already dynamically link to libj, so I think dynamic
linking
is fine. Not hard to bundle the requisite libs. That said, my
preference
is for libbf rather than gmp (and I started on something oriented
thattaways), and it happens to be permissively licensed.
On Wed, 9 Mar 2022, Julian Fondren wrote:
> I got very little through February for various reasons, but I'm back on
> it. I think static linking is the best way as well, including to avoid
> newer builds of J having new dependencies that make J not work on
> machines where it previously did, but if so it probably can't be GMP
for
> licensing reasons. My plan's just to get something working, with the
> expectation that various issues like this will be easy to decide later.
>
> ------- Original Message -------
>
> On Wednesday, March 9th, 2022 at 1:21 PM, Raul Miller <
[email protected]> wrote:
>
>> On Tue, Jan 25, 2022 at 8:38 PM Julian Fondren
[email protected] wrote:
>>
>> > I'll get it done.
>>
>> How is this proceeding?
>>
>> (I have been tempted to tackle this integration myself, but between
>>
>> laptop failures and some other issues, I have not gotten around to it,
>>
>> yet.)
>>
>> I will say this:
>>
>> Because of how J is typically installed, I think that the right
>>
>> approach here would be to statically link against libgmp. Dynamic
>>
>> links are appropriate in a variety of cases, but not this one, not for
>>
>> the initial port. (Once libgmp is supported, it would be possible to
>>
>> build J against a dynamically linked libgmp, but while that might be
>>
>> right for some people, most people are not going to want to deal with
>>
>> the consequences of that approach.)
>>
>> Anyways, ... I might yet get around to tackling this by myself, but if
>>
>> you're making good progress, I should probably either just wait or try
>>
>> to pitch in on issues which are eating too much of your time.
>>
>> Thanks,
>>
>> --
>>
>> Raul
>>
>> ----------------------------------------------------------------------
>>
>> For information about J forums see http://www.jsoftware.com/forums.htm
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
