5:13 p.m.
Jerry -- how come you have to specify use-number-lib to get bignums? Why
aren't they detected by default by configure?
5:34 p.m.
"Ben Wing" <ben(a)666.com>, on Sat, 12 Feb 2005 11:13:28 -0600 you wrote:
Jerry -- how come you have to specify use-number-lib to get bignums?
Why
aren't they detected by default by configure?
It's still considered an experimental feature. If the Board considers
it to be sufficiently stable, we could vote to make it autodetected.
In my opinion, the GNU MP stuff is quite stable. I've used it a fair
amount over the last 9 months or so with no problems. The BSD MP stuff
is terrible, though. BSD MP is just too limited. We should probably
remove support for it.
We should also consider Hrvoje's suggestion that we make it possible to
unpack a GNU MP tarball into the XEmacs build tree, then detect and use
it if there is no installed GNU MP library. This is somewhat complex
because you really want to build GNU MP with optimizations for the CPU
you actually have. An i386-optimized GNU MP performs terribly on a
Pentium 4, for example. If we can solve that problem, we can eventually
make the move to requiring GNU MP, which would make all kinds of
simplifications possible.
--
Jerry James
http://www.ittc.ku.edu/~james/
3:24 a.m.
"Ben Wing" <ben(a)666.com>, on Sat, 12 Feb 2005
11:13:28 -0600
you wrote:
> Jerry -- how come you have to specify use-number-lib to get
bignums?
> Why aren't they detected by default by configure?
It's still considered an experimental feature. If the Board
considers it to be sufficiently stable, we could vote to make
it autodetected.
I would support this.
In my opinion, the GNU MP stuff is quite stable. I've used
it a fair amount over the last 9 months or so with no
problems. The BSD MP stuff is terrible, though. BSD MP is
just too limited. We should probably remove support for it.
Since it's your code, you make the call.
We should also consider Hrvoje's suggestion that we make it
possible to unpack a GNU MP tarball into the XEmacs build
tree, then detect and use it if there is no installed GNU MP
library. This is somewhat complex because you really want to
build GNU MP with optimizations for the CPU you actually
have. An i386-optimized GNU MP performs terribly on a
Pentium 4, for example. If we can solve that problem, we can
eventually make the move to requiring GNU MP, which would
make all kinds of simplifications possible.
I'm a bit leery of this. In the past I've suggested at various times that
certain libraries should be incorporated into XEmacs, mostly because of
their importance and/or lack of current support (e.g. Xpm). However, others
have convinced me that this is a major maintenance hassle. The same
arguments would apply here. For GNU MP, under MS Windows there isn't even a
native Makefile -- the "prescribed way to build" is to use MinGW and then
manually run some magic commands to generate a Windows import library.
Hmmmmph. And no way to static build.
2:24 p.m.
>>>> "Jerry" == Jerry James
<james(a)eecs.ku.edu> writes:
Jerry> "Ben Wing" <ben(a)666.com>, on Sat, 12 Feb 2005 11:13:28
Jerry> -0600 you wrote:
> Jerry -- how come you have to specify use-number-lib to get
> bignums? Why aren't they detected by default by configure?
Jerry> It's still considered an experimental feature. If the
Jerry> Board considers it to be sufficiently stable, we could vote
Jerry> to make it autodetected.
Why do this? Bignums are a limited audience feature; that audience is
probably more limited than the one for the Canna Japanese input
method. As far as I know they have no utility in editing, and they do
slow everything else down (although probably immeasurably) by adding
checks for bignums in numerical contexts. If we had large file
support, yes, that would be a killer app, but we don't.
--
Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp
University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Ask not how you can "do" free software business;
ask what your business can "do for" free software.
1:07 a.m.
>>>> "Jerry" == Jerry James
<james(a)eecs.ku.edu> writes:
Jerry> We should also consider Hrvoje's suggestion that we make it possible to
Jerry> unpack a GNU MP tarball into the XEmacs build tree, then detect and use
Jerry> it if there is no installed GNU MP library.
I'm not sure about including it directly in the tarball but there are other
situations too where it would be good to be able to use a library that can't
be assumed to be part of the user's default configuration. The recent Carbon
port comes to mind, but Windows might be another one.
Instead of including the source, why not designate a directory in the source
tree (say third_party) in which any source trees found therein are build and
installed with the same paths as XEmacs. That would make a stand-alone
install tree that would also allow the user to mix and match XEmacs and
library versions (which is what pretty much happens already with the installed
libraries).
Jerry> This is somewhat complex because you really want to build GNU MP with
Jerry> optimizations for the CPU you actually have. An i386-optimized GNU MP
Jerry> performs terribly on a Pentium 4, for example.
It's not autodetected by GNU MP's configure? If not then our configure knows
the CPU type pretty early in the piece so there is no reason why it can't pass
it along when doing the GNU MP configure and build.
Malcolm
--
Malcolm Purvis <malcolmp(a)xemacs.org>
4:12 a.m.
>>>> "Malcolm" == Malcolm Purvis
<malcolmp(a)xemacs.org> writes:
>>>> "Jerry" == Jerry James
<james(a)eecs.ku.edu> writes:
Jerry> We should also consider Hrvoje's suggestion that we make it
Jerry> possible to unpack a GNU MP tarball into the XEmacs build
Jerry> tree, then detect and use it if there is no installed GNU
Jerry> MP library.
Malcolm> I'm not sure about including it directly in the tarball
Malcolm> but there are other situations too where it would be good
Malcolm> to be able to use a library that can't be assumed to be
Malcolm> part of the user's default configuration.
I really don't think this is something we want to start on now _if_
we're going to be releasing XEmacs 22.0 in 2005.
--
Institute of Policy and Planning Sciences http://turnbull.sk.tsukuba.ac.jp
University of Tsukuba Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Ask not how you can "do" free software business;
ask what your business can "do for" free software.
4:36 p.m.
Malcolm Purvis <malcolmp(a)xemacs.org> writes:
>>>>> "Jerry" == Jerry James
<james(a)eecs.ku.edu> writes:
Jerry> We should also consider Hrvoje's suggestion that we make it possible to
Jerry> unpack a GNU MP tarball into the XEmacs build tree, then detect and use
Jerry> it if there is no installed GNU MP library.
I'm not sure about including it directly in the tarball but there
are other situations too where it would be good to be able to use a
library that can't be assumed to be part of the user's default
configuration.
The reason I suggested that was to enable us to make bignums an
integral part of XEmacs, like they are in CLISP or Python. If bignum
support remains an "option", its usefulness is IMO questionable.
7001
days inactive
7007
days old
6 comments
5 participants
participants (5)
-
Ben Wing -
Hrvoje Niksic -
Jerry James -
Malcolm Purvis -
Stephen J. Turnbull