On 9/7/21 12:39 AM, Juha Nieminen wrote:
....
> As for what the goal is, I already said it: The library is quite clearly
> intended to be one of the most, if not the most efficient and useful
> bignum libraries for scientific research and other similar purposes
> that require such calculations. An enormous amount of research and
> work has been put into it in order to make it as efficient as possible.
> Who knows how many hours in total has been put into making it the
> greatest and best library at what it's intended to do.
>
> Thus, if they were to disregard one of the biggest, if not the biggest
> desktop operating systems out there (which is the platform most likely
> used by those in need of such a library), they are being rather stupid.

No, they're just disagreeing with you about that judgement call. They
know, in far more detail than you do, who their users are, and in their
judgement the number of prospective users who need a Windows version and
can't take advantage of the options to use it through Cygwin, MSYS2, or
WSL is too small to justify the effort required to provide a native
Windows version. How can you be so certain that they're wrong about
that? Do you know of any significant number of people complaining about
the absences of a native Windows version of GMP? Keep in mind that it's
their own effort that would be required, so it's entirely their own
business how they expend that effort.

On Tuesday, 7 September 2021 at 16:32:20 UTC+1, james...@alumni.caltech.edu wrote:
> On 9/7/21 12:39 AM, Juha Nieminen wrote:
> ...
> > As for what the goal is, I already said it: The library is quite clearly
> > intended to be one of the most, if not the most efficient and useful
> > bignum libraries for scientific research and other similar purposes
> > that require such calculations. An enormous amount of research and
> > work has been put into it in order to make it as efficient as possible.
> > Who knows how many hours in total has been put into making it the
> > greatest and best library at what it's intended to do.
> >
> > Thus, if they were to disregard one of the biggest, if not the biggest
> > desktop operating systems out there (which is the platform most likely
> > used by those in need of such a library), they are being rather stupid.
>
> No, they're just disagreeing with you about that judgement call. They
> know, in far more detail than you do, who their users are, and in their
> judgement the number of prospective users who need a Windows version and
> can't take advantage of the options to use it through Cygwin, MSYS2, or
> WSL is too small to justify the effort required to provide a native
> Windows version. How can you be so certain that they're wrong about
> that? Do you know of any significant number of people complaining about
> the absences of a native Windows version of GMP? Keep in mind that it's
> their own effort that would be required, so it's entirely their own
> business how they expend that effort.
>
Windows is mainly a platform for business software. A high precision
numerical library is of little interest to most developers of business
software. The other main use for Windows is probably games, where
again you are unlikely to need mathematical operations that aren't supported
by C's basic types.
Geeks doing hobby programming and some scientists (though not many)
will need bignums, but they largely use Linux for the desktop. If you need
bignums for academic purposes, you'll likely need a lot of them, so you'll
want a mainframe.

On 07/09/2021 16:32, James Kuyper wrote:
> On 9/7/21 12:39 AM, Juha Nieminen wrote:
> ...
>> As for what the goal is, I already said it: The library is quite clearly
>> intended to be one of the most, if not the most efficient and useful
>> bignum libraries for scientific research and other similar purposes
>> that require such calculations. An enormous amount of research and
>> work has been put into it in order to make it as efficient as possible.
>> Who knows how many hours in total has been put into making it the
>> greatest and best library at what it's intended to do.
>>
>> Thus, if they were to disregard one of the biggest, if not the biggest
>> desktop operating systems out there (which is the platform most likely
>> used by those in need of such a library), they are being rather stupid.
>
> No, they're just disagreeing with you about that judgement call. They
> know, in far more detail than you do, who their users are, and in their
> judgement the number of prospective users who need a Windows version and
> can't take advantage of the options to use it through Cygwin, MSYS2, or
> WSL is too small to justify the effort required to provide a native
> Windows version. How can you be so certain that they're wrong about
> that? Do you know of any significant number of people complaining about
> the absences of a native Windows version of GMP? Keep in mind that it's
> their own effort that would be required, so it's entirely their own
> business how they expend that effort.

How much effort /would/ be required, by the people who are experts in
the product, in using Linux, and who would probably know how to knock
off a cross-compiled DLL?

Either it's trivial, and it would be churlish to refuse to support that
platform.

Or it is decidely non-trivial, and it would be unfair to dump it all on
users who have little knowledge of its organisation, of its build
system, or products such as Cygwin and MSY2 which would need to be
specially installated by every user.

And who would have little interest in doing so. All they should need is
a .dll file and, if using it from C, a .h file. Using GMP itself is
challenging enough, without having to go hustling for the actual library.

On 07/09/2021 16:54, Malcolm McLean wrote:
> On Tuesday, 7 September 2021 at 16:32:20 UTC+1, james...@alumni.caltech.edu wrote:
>> On 9/7/21 12:39 AM, Juha Nieminen wrote:
>> ...
>>> As for what the goal is, I already said it: The library is quite clearly
>>> intended to be one of the most, if not the most efficient and useful
>>> bignum libraries for scientific research and other similar purposes
>>> that require such calculations. An enormous amount of research and
>>> work has been put into it in order to make it as efficient as possible.
>>> Who knows how many hours in total has been put into making it the
>>> greatest and best library at what it's intended to do.
>>>
>>> Thus, if they were to disregard one of the biggest, if not the biggest
>>> desktop operating systems out there (which is the platform most likely
>>> used by those in need of such a library), they are being rather stupid.
>>
>> No, they're just disagreeing with you about that judgement call. They
>> know, in far more detail than you do, who their users are, and in their
>> judgement the number of prospective users who need a Windows version and
>> can't take advantage of the options to use it through Cygwin, MSYS2, or
>> WSL is too small to justify the effort required to provide a native
>> Windows version. How can you be so certain that they're wrong about
>> that? Do you know of any significant number of people complaining about
>> the absences of a native Windows version of GMP? Keep in mind that it's
>> their own effort that would be required, so it's entirely their own
>> business how they expend that effort.
>>
> Windows is mainly a platform for business software. A high precision
> numerical library is of little interest to most developers of business
> software. The other main use for Windows is probably games, where
> again you are unlikely to need mathematical operations that aren't supported
> by C's basic types.
> Geeks doing hobby programming and some scientists (though not many)
> will need bignums, but they largely use Linux for the desktop. If you need
> bignums for academic purposes, you'll likely need a lot of them, so you'll
> want a mainframe.
>

I needed such a library to support bignums in my interpreted language.

If GMP was an easy dependency that could be reliably downloaded by
anyone, then I might have considered it. But I couldn't even create the
DLL for my own use.

Anyone using such a language (eg. Python) could be using a bignum
library without be aware of it.

You can't really set the 1 billion users of Windows aside and say, that
as a class, none of them will need arbitrary precision either directly
or indirectly.

And anyway this was just an example of one library. Look at the vast
numbers of libraries that /are/ available for Windows.

On Tuesday, September 7, 2021 at 12:39:25 PM UTC-4, Bart wrote:
....
> If GMP was an easy dependency that could be reliably downloaded by
> anyone, then I might have considered it. But I couldn't even create the
> DLL for my own use.

The website contains instructions for creating a dll. Doing so probably requires
skills that you lack.

> Anyone using such a language (eg. Python) could be using a bignum
> library without be aware of it.
>
> You can't really set the 1 billion users of Windows aside and say, that
> as a class, none of them will need arbitrary precision either directly
> or indirectly.

If GMP developers are ignoring the needs of people interested in using
GMP, but unable to do so because they're using Windows and can't use
Cygwin, MSYS2, or WSL, it's probably because those people haven't bothered
expressing their interest to the GMP developers, at least not in large numbers.

On 08/09/2021 03:17, james...@alumni.caltech.edu wrote:
> On Tuesday, September 7, 2021 at 12:39:25 PM UTC-4, Bart wrote:
> ...
>> If GMP was an easy dependency that could be reliably downloaded by
>> anyone, then I might have considered it. But I couldn't even create the
>> DLL for my own use.
>
> The website contains instructions for creating a dll. Doing so probably requires
> skills that you lack.

What skills should be necessary? I can create a DLL for my own library
using:

mm -dll bignum

mm.exe is one of my compilers.

What appear to be lacking in such projects (of which GMP is only one
example I've encountered) are the skills required to make it equally
simple, without adding more layers of software and more third party
dependencies. And more points of failure.

>> Anyone using such a language (eg. Python) could be using a bignum
>> library without be aware of it.
>>
>> You can't really set the 1 billion users of Windows aside and say, that
>> as a class, none of them will need arbitrary precision either directly
>> or indirectly.
>
> If GMP developers are ignoring the needs of people interested in using
> GMP, but unable to do so because they're using Windows and can't use
> Cygwin, MSYS2, or WSL, it's probably because those people haven't bothered
> expressing their interest to the GMP developers, at least not in large numbers.

It's just the wrong mindset. The developers need to jump through N hoops
while working on the product, because it is after all quite a
complicated one with T diverse targets.

So they seem to think that every end user needs to jump through the same
N hoops, even if most of those don't exist on the user's machine, and
even if the user is interested on just 1 target.

I accept that GMP can't be in 100% C, and needs some target-specific ASM
code. That might mean needing to use a gcc installation (to either deal
with gcc-style inline assembly, or just to process .s files).

But I have gcc; I still can't build the project. Why not? What other
magic is needed, and why?

Why does no one else ask these questions?

Bart <bc@freeuk.com> wrote:
> On 07/09/2021 16:32, James Kuyper wrote:
> > On 9/7/21 12:39 AM, Juha Nieminen wrote:
> > ...
> >> As for what the goal is, I already said it: The library is quite clearly
> >> intended to be one of the most, if not the most efficient and useful
> >> bignum libraries for scientific research and other similar purposes
> >> that require such calculations. An enormous amount of research and
> >> work has been put into it in order to make it as efficient as possible.
> >> Who knows how many hours in total has been put into making it the
> >> greatest and best library at what it's intended to do.
> >>
> >> Thus, if they were to disregard one of the biggest, if not the biggest
> >> desktop operating systems out there (which is the platform most likely
> >> used by those in need of such a library), they are being rather stupid.
> >
> > No, they're just disagreeing with you about that judgement call. They
> > know, in far more detail than you do, who their users are, and in their
> > judgement the number of prospective users who need a Windows version and
> > can't take advantage of the options to use it through Cygwin, MSYS2, or
> > WSL is too small to justify the effort required to provide a native
> > Windows version. How can you be so certain that they're wrong about
> > that? Do you know of any significant number of people complaining about
> > the absences of a native Windows version of GMP? Keep in mind that it's
> > their own effort that would be required, so it's entirely their own
> > business how they expend that effort.
>
>
> How much effort /would/ be required, by the people who are experts in
> the product, in using Linux, and who would probably know how to knock
> off a cross-compiled DLL?
>
> Either it's trivial, and it would be churlish to refuse to support that
> platform.
>
> Or it is decidely non-trivial, and it would be unfair to dump it all on
> users who have little knowledge of its organisation, of its build
> system, or products such as Cygwin and MSY2 which would need to be
> specially installated by every user.
>
> And who would have little interest in doing so. All they should need is
> a .dll file and, if using it from C, a .h file. Using GMP itself is
> challenging enough, without having to go hustling for the actual library.

Well, you got half of dynamic here right, namely why no Windows users
provide easily buildable kits. But you got wrong the other part.

First of all, to have viable port to environment X you need access
to that enviroment. Which means limited but not entirely trivial effort
to set up such environment and keep it running. Do not forget
about version differences: to support several versions you need
several versions of environment. Users of environment X presumably
have this set up so for them this part comes "for free". For
others it is extra burden. Some time ago I decided that keeping
copy of Windows only for puprose of porting my programs to
Windows makes no sense. I am trying to write portable code
and I am even willing to write Windows specific code based on
available documentation, but it is for Windows users to tell me
if there are any problems and to test possible fixes. If no
Windows users is willing to do that, then from my point of
view it is as if Windows had no users at all.

Coming back to GMP, it is widely used on Windows and nobody
told me that there are any troubles. Users who are willing
to follows instructions can easily get it. Some get it
via Cygwin, some get it with Mingw/Msys (first Google
hit for "Msys GMP" gives you link to corresponding package).
AFAIK people using MSVC know what to do to use GMP.
In the past I have cross-compiled libraries intended to
run on Windows using Mingw gcc on Linux as cross-compiler.
I think that I would be able to cross-compile GMP as
well. But since I do not have Windows I would be unable
to test it. And you already have several places where
you can get binary GMP, so this would really change nothing
for you.

Let me add a little story. Several years ago I was asked
to help with installing a program. The program was add-on
on top of Oracle database. At first person needing the
program asked two bright students to install it. The
students told me that that they followed supplied
installation instruction, but there were errors. They
tried to work around but got stuck. They said that
program must be completely broken. I looked at error
messages and messages were cryptic confirming what
students said. So I decided to start fresh, deleting
everthing that students managed to install. In the
instructions there were one ugly looking point. It
was natural to do it differently and I first tried
such modified instructions. I got similar problems
as reported by students. So I started fresh again,
this time literally following instructions. The
instalation process went fine. So, clearly students
were unable to follow simple instructions (probably
they either did not notice that they deviated or
thought that deviation can not make difference).

My conclusion from the story: significant fraction
of general population can not follow simple
instructions. One can try to make things
"foolproof", but in case of software fools are
creative enough to find unexpected ways of
breaking things.

--
Waldek Hebisch

Bart <bc@freeuk.com> wrote:
> On 07/09/2021 16:32, James Kuyper wrote:
....
> > No, they're just disagreeing with you about that judgement call. They
> > know, in far more detail than you do, who their users are, and in their
> > judgement the number of prospective users who need a Windows version and
> > can't take advantage of the options to use it through Cygwin, MSYS2, or
> > WSL is too small to justify the effort required to provide a native
> > Windows version. How can you be so certain that they're wrong about
> > that? Do you know of any significant number of people complaining about
> > the absences of a native Windows version of GMP? Keep in mind that it's
> > their own effort that would be required, so it's entirely their own
> > business how they expend that effort.
>
>
> How much effort /would/ be required, by the people who are experts in
> the product, in using Linux, and who would probably know how to knock
> off a cross-compiled DLL?

I don't know, and the GMP developers do - and they're the ones who made
the decision about how to spend their own time. Even those developers
don't necessarily all know - if they only have Linux systems, not only
would getting a Windows system to test on be an unnecessary extra
expense, but they might also lack the Windows experience needed to
accurately judge how difficult porting it to Windows might be.
If they decided that the number of potential users who needed a native
Windows build is too small to justify making that effort, on what
grounds do you challenge that judgement? Do you know how many requests
they've received for such a port?

> Or it is decidely non-trivial, and it would be unfair to dump it all on
> users who have little knowledge of its organisation, of its build
> system, or products such as Cygwin and MSY2 which would need to be
> specially installated by every user.

If it's sufficiently non-trivial, and if the number of potential users
who want it done is sufficiently small (as appears to be the case), then
surely the right response is to spare everyone the necessity of doing
it, by not doing it at all?

On Thursday, September 9, 2021 at 6:41:13 PM UTC+3, james...@alumni.caltech.edu wrote:
> Bart <b...@freeuk.com> wrote:
> > On 07/09/2021 16:32, James Kuyper wrote:
> ...
> > > No, they're just disagreeing with you about that judgement call. They
> > > know, in far more detail than you do, who their users are, and in their
> > > judgement the number of prospective users who need a Windows version and
> > > can't take advantage of the options to use it through Cygwin, MSYS2, or
> > > WSL is too small to justify the effort required to provide a native
> > > Windows version. How can you be so certain that they're wrong about
> > > that? Do you know of any significant number of people complaining about
> > > the absences of a native Windows version of GMP? Keep in mind that it's
> > > their own effort that would be required, so it's entirely their own
> > > business how they expend that effort.
> >
> >
> > How much effort /would/ be required, by the people who are experts in
> > the product, in using Linux, and who would probably know how to knock
> > off a cross-compiled DLL?
> I don't know, and the GMP developers do - and they're the ones who made
> the decision about how to spend their own time. Even those developers
> don't necessarily all know - if they only have Linux systems, not only
> would getting a Windows system to test on be an unnecessary extra
> expense, but they might also lack the Windows experience needed to
> accurately judge how difficult porting it to Windows might be.
> If they decided that the number of potential users who needed a native
> Windows build is too small to justify making that effort, on what
> grounds do you challenge that judgement? Do you know how many requests
> they've received for such a port?

I would imagine that lots of potential GMP users that develop with Microsoft
Visual Studio tools end up using boost::multiprecision instead of GMP exactly
because of "friction" described by Bart.
Now, boost::multiprecision is not crap, but every module that uses it compiles MUCH
slower that it would with GMP. Also, boost::multiprecision is not tested as well as GMP
so it's quite possible that it still has bugs in rarely used corners.
On the positive side, it's probably a little faster at run time, esp. for relatively small numbers,
but I would guess that majority of users care about slowness of compilation more than
about small speed up at run time.

> > Or it is decidely non-trivial, and it would be unfair to dump it all on
> > users who have little knowledge of its organisation, of its build
> > system, or products such as Cygwin and MSY2 which would need to be
> > specially installated by every user.
> If it's sufficiently non-trivial, and if the number of potential users
> who want it done is sufficiently small (as appears to be the case), then
> surely the right response is to spare everyone the necessity of doing
> it, by not doing it at all?

On Tuesday, September 7, 2021 at 6:54:45 PM UTC+3, Malcolm McLean wrote:
> On Tuesday, 7 September 2021 at 16:32:20 UTC+1, james...@alumni.caltech.edu wrote:
> > On 9/7/21 12:39 AM, Juha Nieminen wrote:
> > ...
> > > As for what the goal is, I already said it: The library is quite clearly
> > > intended to be one of the most, if not the most efficient and useful
> > > bignum libraries for scientific research and other similar purposes
> > > that require such calculations. An enormous amount of research and
> > > work has been put into it in order to make it as efficient as possible.
> > > Who knows how many hours in total has been put into making it the
> > > greatest and best library at what it's intended to do.
> > >
> > > Thus, if they were to disregard one of the biggest, if not the biggest
> > > desktop operating systems out there (which is the platform most likely
> > > used by those in need of such a library), they are being rather stupid.
> >
> > No, they're just disagreeing with you about that judgement call. They
> > know, in far more detail than you do, who their users are, and in their
> > judgement the number of prospective users who need a Windows version and
> > can't take advantage of the options to use it through Cygwin, MSYS2, or
> > WSL is too small to justify the effort required to provide a native
> > Windows version. How can you be so certain that they're wrong about
> > that? Do you know of any significant number of people complaining about
> > the absences of a native Windows version of GMP? Keep in mind that it's
> > their own effort that would be required, so it's entirely their own
> > business how they expend that effort.
> >
> Windows is mainly a platform for business software. A high precision
> numerical library is of little interest to most developers of business
> software. The other main use for Windows is probably games, where
> again you are unlikely to need mathematical operations that aren't supported
> by C's basic types.
> Geeks doing hobby programming and some scientists (though not many)
> will need bignums, but they largely use Linux for the desktop. If you need
> bignums for academic purposes, you'll likely need a lot of them, so you'll
> want a mainframe.

Your last suggestion is funny.

Bart <bc@freeuk.com> wrote:
> On 08/09/2021 03:17, james...@alumni.caltech.edu wrote:
> > On Tuesday, September 7, 2021 at 12:39:25 PM UTC-4, Bart wrote:
> > ...
> >> If GMP was an easy dependency that could be reliably downloaded by
> >> anyone, then I might have considered it. But I couldn't even create the
> >> DLL for my own use.
> >
> > The website contains instructions for creating a dll. Doing so probably requires
> > skills that you lack.
>
> What skills should be necessary? I can create a DLL for my own library
> using:
>
> mm -dll bignum
>
> mm.exe is one of my compilers.
>
> What appear to be lacking in such projects (of which GMP is only one
> example I've encountered) are the skills required to make it equally
> simple, without adding more layers of software and more third party
> dependencies. And more points of failure.
>
> >> Anyone using such a language (eg. Python) could be using a bignum
> >> library without be aware of it.
> >>
> >> You can't really set the 1 billion users of Windows aside and say, that
> >> as a class, none of them will need arbitrary precision either directly
> >> or indirectly.
> >
> > If GMP developers are ignoring the needs of people interested in using
> > GMP, but unable to do so because they're using Windows and can't use
> > Cygwin, MSYS2, or WSL, it's probably because those people haven't bothered
> > expressing their interest to the GMP developers, at least not in large numbers.
>
> It's just the wrong mindset. The developers need to jump through N hoops
> while working on the product, because it is after all quite a
> complicated one with T diverse targets.
>
> So they seem to think that every end user needs to jump through the same
> N hoops, even if most of those don't exist on the user's machine, and
> even if the user is interested on just 1 target.
>
> I accept that GMP can't be in 100% C, and needs some target-specific ASM
> code. That might mean needing to use a gcc installation (to either deal
> with gcc-style inline assembly, or just to process .s files).
>
> But I have gcc; I still can't build the project. Why not? What other
> magic is needed, and why?
>
> Why does no one else ask these questions?

Apparently they do not have problems that you have. Or if they have,
they ask for hlpe in right place. I know folks using GMP on Windows,
and if there are problems they can resolve them without making great
fuss: I know they you GMP and I heard nothing about troubles...

--
Waldek Hebisch

On 10/09/2021 12:28, antispam@math.uni.wroc.pl wrote:
> Bart <bc@freeuk.com> wrote:

>> Why does no one else ask these questions?
>
> Apparently they do not have problems that you have. Or if they have,
> they ask for hlpe in right place. I know folks using GMP on Windows,
> and if there are problems they can resolve them without making great
> fuss: I know they you GMP and I heard nothing about troubles...

How you tried it yourself? If not then you can't really cast doubt on my
claims.

Before posting, I thought I'd download MSYS2 and try it once more.

I did that (all 378MB of it across 16000 files and 1000 directories -
but no C compiler!).

If I start msys2.exe, then I get a prompt with $. I can then navigate to
a place where the GMP sources are.

If I type ./configure, after 20 seconds it says it can't find a working
compiler. I need to install one. Apparently the first step is to type:

$ pacman -Syu

Which I try, and I get:

-bash: pacman: command not found

Nothing else works either. That's Problem One.

So, now what? I won't give up immediately; if I start bash.exe instead
of msys2.exe, it stays within my Windows file system, and it can see my
tdm/gcc installation, but it can't see any of the commands used by
configure.

If I add C:\msys64\usr\bin to the search paths, then it keeps configure
happy for 10-15 minutes, doing lots of pointless tests to do with the
minutiae of C functionality, until it says this:

configure: error: no version of invert_limv_table found in path
x64_64/k10 x64_64/k8 x86_64 generic

That's Problem Two.

So NOW what? No doubt if I somehow fix that, there will be more. But
would I even know if it was successful - will it tell me what the output
file is called (hopefully a DLL, but who knows), and where it will be
located?

This is of course all my fault!

However, how many of those issues do you think would come up if my
starting point had been a precompiled binary DLL?