scheme directory

scheme directory

[Satoru Takabayashi]

Hi, I found the description of INSTALL file in the cvs
repository was inappropriate.

Original:

  # mkdir /usr/local/share/emacs/scheme
  # cp -r emacs /usr/local/share/emacs/scheme

My suggestion:

  # mkdir /usr/local/share/emacs/20.6/scheme
  # cp -r emacs /usr/local/share/emacs/20.6/scheme

  or 

  # mkdir /usr/local/share/emacs/site-scheme
  # cp -r emacs /usr/local/share/emacs/site-scheme


Anyway, why should we install *.scm files into
"/usr/local/share/emacs/20.6/scheme/emacs" instead of
"/usr/local/share/emacs/20.6/scheme"?  I think `emacs' at
the end of the path is redundant.

Moreover, it might be fine to install *.scm files
automatically by `make install' of Guile Emacs.

p.s.
I am new to Scheme. I have been studying Scheme with Guile
Emacs.  scheme-interaction-mode is great. I hope
interrupting the Scheme process with C-g will be possible.

Thanks,
-- Satoru Takabayashi

Re: scheme directory

[Kalle Olavi Niemitalo]

Satoru Takabayashi <satoru-t@is.aist-nara.ac.jp> writes:

>   # mkdir /usr/local/share/emacs/20.6/scheme
>   # cp -r emacs /usr/local/share/emacs/20.6/scheme

This would be best, I think.

>   # mkdir /usr/local/share/emacs/site-scheme
>   # cp -r emacs /usr/local/share/emacs/site-scheme

Actually, this is how it was before I changed "site-scheme" to
"scheme".  ;-)

> Anyway, why should we install *.scm files into
> "/usr/local/share/emacs/20.6/scheme/emacs" instead of
> "/usr/local/share/emacs/20.6/scheme"?  I think `emacs' at
> the end of the path is redundant.

The files contain Guile modules.  For example, emacs/guile.scm is
loaded with (use-module (emacs guile)).  In the module name,
`emacs' is not redundant.  If we installed the *.scm files
directly in .../scheme/, we would have to either change Guile's
module-loading procedures or install a symlink somewhere.

> Moreover, it might be fine to install *.scm files
> automatically by `make install' of Guile Emacs.

True.

Re: scheme directory

[Ken Raeburn]

Eventually, it'd be nice to have Scheme code installed parallel to
Lisp code -- i.e., the 20.6/scheme directory and a site-scheme
directory, both....

interrupting the Scheme process

[Keisuke Nishida]

Satoru Takabayashi <satoru-t@is.aist-nara.ac.jp> writes:

> I hope interrupting the Scheme process with C-g will be possible.

I have no idea of how this can be done right now.  Emacs checks for
quit in some loops such as eval.  When the process is in the Scheme
interpreter, we can't use the same way; instead, we have to generate
a user interrupt.  I guess it is possible for Guile Emacs to bind C-g
to generate an interrupt before calling the Scheme evaluator, but I'm
not sure about this part.  Ken, do you have any idea?

Also, once we come to support multi-thread programming with Emacs,
things become more complicated.  Probably we have to maintain threads
in the same way as shells; that is, Emacs will have a "current thread"
and "background threads", and the user may quit only the current thread
by typing C-g, whereas background threads must be killed by a special
command like M-x kill-thread.

-- Kei

Re: interrupting the Scheme process

[Kalle Olavi Niemitalo]

We might be able to use something like the "emergency escape"
(double `C-g') feature.  Add a C flag `scheme_evaluator_active'
and make `scheme-eval' and `lisp-eval' set and reset it.  Then in
the `C-g' handler which usually tests `quit-flag', interrupt
Guile instead if `scheme_evaluator_active' is set.

Emacs would also have to convert Lisp quits to Scheme interrupts
and vice versa when returning from the *-eval functions.

Could this work?

Re: interrupting the Scheme process

[Ken Raeburn]

Keisuke Nishida <kxn30@po.cwru.edu> writes:
> I have no idea of how this can be done right now.  Emacs checks for
> quit in some loops such as eval.  When the process is in the Scheme
> interpreter, we can't use the same way; instead, we have to generate
> a user interrupt.  I guess it is possible for Guile Emacs to bind C-g
> to generate an interrupt before calling the Scheme evaluator, but I'm
> not sure about this part.  Ken, do you have any idea?

Guile has some hooks for recording that an interrupt has occurred and,
at certain times, throwing an exception of some sort.  If we could set
that flag at C-g time, if Scheme code is currently running, might that
do the trick?  Then Scheme can trap the exception if it wants, but
otherwise we throw back to the containing Lisp call.  Propagating the
unwind through possibly multiple lisp<->scheme interfaces could be
hairy though; I haven't looked very closely at that stuff.

> Also, once we come to support multi-thread programming with Emacs,
> things become more complicated.  Probably we have to maintain threads
> in the same way as shells; that is, Emacs will have a "current thread"
> and "background threads", and the user may quit only the current thread
> by typing C-g, whereas background threads must be killed by a special
> command like M-x kill-thread.

Hm...that might be a way to do it.  I also wonder if we might have
cases where we want to send the interrupt to multiple threads -- i.e.,
if the "foreground" thread is sitting around waiting for N tasks to
finish (e.g., get new news from news.mycompany.com, get email from
pop.mycompany.com, get new news from news.redhat.com, etc), might we
want to interrupt all of them and unwind the main thread only after
they've died off?

I suppose we could have a "spawn and wait for multiple threads"
function which implements all of this on top of the scheme^Wplan you
describe, by trapping the interrupt...

Ken

Re: interrupting the Scheme process

[Keisuke Nishida]

Ken Raeburn <raeburn@raeburn.org> writes:

> Guile has some hooks for recording that an interrupt has occurred and,
> at certain times, throwing an exception of some sort.  If we could set
> that flag at C-g time, if Scheme code is currently running, might that
> do the trick?  Then Scheme can trap the exception if it wants, but
> otherwise we throw back to the containing Lisp call.  Propagating the
> unwind through possibly multiple lisp<->scheme interfaces could be
> hairy though; I haven't looked very closely at that stuff.

It seems Emacs catches the interrupt by interrupt_signal in keyboard.c,
so we could modify this function so that it also sets the Guile's flag.
If this works, we can convert the signal from Guile into a Emacs's quit
signal.  In case this doesn't work, I don't know what to do.

/* This routine is called at interrupt level in response to C-G.
 If interrupt_input, this is the handler for SIGINT.
 Otherwise, it is called from kbd_buffer_store_event,
 in handling SIGIO or SIGTINT.

 If `waiting_for_input' is non zero, then unless `echoing' is nonzero,
 immediately throw back to read_char.

 Otherwise it sets the Lisp variable  quit-flag  not-nil.
 This causes  eval  to throw, when it gets a chance.
 If  quit-flag  is already non-nil, it stops the job right away.  */

SIGTYPE
interrupt_signal (signalnum)	/* If we don't have an argument, */
     int signalnum;		/* some compilers complain in signal calls. */

> Hm...that might be a way to do it.  I also wonder if we might have
> cases where we want to send the interrupt to multiple threads -- i.e.,
> if the "foreground" thread is sitting around waiting for N tasks to
> finish (e.g., get new news from news.mycompany.com, get email from
> pop.mycompany.com, get new news from news.redhat.com, etc), might we
> want to interrupt all of them and unwind the main thread only after
> they've died off?

Probably.  We should kill the parent thread and its children at the same
time, though I guess we would still need kill-thread to kill detached
threads, which are "background" threads.  We don't want to close a GTK
window, which may run in a separate thread, just by typing C-g, for example.

> I suppose we could have a "spawn and wait for multiple threads"
> function which implements all of this on top of the scheme^Wplan you
> describe, by trapping the interrupt...

Anyway, we can't work on this until your Guile-based Emacs appears :)

-- Kei

Re: interrupting the Scheme process

[Ken Raeburn]

Keisuke Nishida <kxn30@po.cwru.edu> writes:
> It seems Emacs catches the interrupt by interrupt_signal in keyboard.c,
> so we could modify this function so that it also sets the Guile's flag.
> If this works, we can convert the signal from Guile into a Emacs's quit
> signal.  In case this doesn't work, I don't know what to do.

Looks like the right place to start.  Like I said, figuring out
which language should get the interrupt and how to propagate the
interrupt across language boundaries will probably be trickier than
figuring out where to set the flags.

> > I suppose we could have a "spawn and wait for multiple threads"
> > function which implements all of this on top of the scheme^Wplan you
> > describe, by trapping the interrupt...
> 
> Anyway, we can't work on this until your Guile-based Emacs appears :)

Yeah....  Emacs 21.1 is going into feature freeze soon, so it's a good
time for me to stop putting changes into it and focus on merging it
back into my guile-emacs source base.  Then when the final release
comes, I should be able to make diffs available rather quickly.

Ken