[F.J.Wright@qmul.ac.uk: rcs2log]

[F.J.Wright@qmul.ac.uk: rcs2log]

[Richard Stallman]

Anyone interested in working on this?

------- Start of forwarded message -------
Envelope-to: emacs-pretest-bug@gnu.org
Delivery-date: Fri, 25 Oct 2002 07:13:15 -0400
From: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
To: <emacs-pretest-bug@gnu.org>
Subject: rcs2log
Date: Fri, 25 Oct 2002 12:12:07 +0100
Organization: Mathematical Sciences, Queen Mary, University of London
X-Priority: 3
X-MSMail-Priority: Normal
X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2800.1106
X-Spam-Status: No, hits=-1.3 required=5.0
	tests=INVALID_MSGID,NOSPAM_INC,SPAM_PHRASE_00_01,USER_AGENT_OE
	version=2.41
X-Spam-Level: 

This is a multi-part message in MIME format.

- ------=_NextPart_000_002D_01C27C1F.BCCCBD40
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

GNU Emacs 21.2.91.1 (i386-mingw-nt4.0.1381) of 2002-10-13 on VALETTA

Everything seems to work fine on both Windows NT and XP, except that the
command `vc-update-change-log' does not work.  It tries to execute a UNIX
shell script called rcs2log in the bin directory, but the script is not
there in NTEmacs, and even if it were there NTEmacs could not execute it
directly

I suggest the following solution.  Modify the NTEmacs installation procedure
to include the file rcs2log in the bin directory, and also include in the
bin directory the attached file rcs2log.bat (with any modifications deemed
appropriate), which NTEmacs can execute directly and which acts as an
interface to rcs2log.  I have written rcs2log.bat to be as portable across
different versions of Windows as possible, without losing the advantages of
NT.  However, I have tested it only on Windows XP, where it seems to work
well.

Of course, this all assumes a suitable UNIX-like environment on the Windows
platform, and I use Cygwin, which works well.  I think it is reasonably
likely that anyone trying to execute the Emacs command
`vc-update-change-log' in NTEmacs will have a suitable environment for
executing UNIX shell scripts.

An alternative solution to using rcs2log.bat would be to make a trivial
modification to vc.el so that it explicitly calls sh to interpret rcs2log
instead of relying on the kernel to interpret the #! and call sh indirectly,
which is the part that does not work on NTEmacs.  I tested this solution,
which also works well.  The advantage of using rcs2log.bat is that it has no
impact on any other platform.  However, if you would prefer to modify vc.el
I would be happy to provide diffs and a changelog entry if you want.  Since
rcs2log must be interpreted by sh anyway, I don't think that making this
explicit in the Emacs source code would lose anything, so this is probably
the most elegant solution.

Francis

- ---

Dr Francis J. Wright
School of Mathematical Sciences, Queen Mary
University of London, Mile End Road, London E1 4NS, UK
Tel: 020 7882 5453 (direct);  Fax: 020 8981 9587 (dept.)
F.J.Wright@qmul.ac.uk;  http://centaur.maths.qmul.ac.uk/

- ------=_NextPart_000_002D_01C27C1F.BCCCBD40
Content-Type: application/x-gzip;
	name="rcs2log.bat.gz"
Content-Transfer-Encoding: base64
Content-Disposition: attachment;
	filename="rcs2log.bat.gz"

H4sICFRluD0AA3JjczJsb2cuYmF0AI2RX0/bMBTF3yPlO5wHrEjVEgT7w6gGoqvoNjQ6BK26t+E4
N7FFYhfbaZdvP7dJgce9XV3f3znnXl+RkAamLOPIUgMr3GltqiznHmmK+eK64cJBaU+25ILgzWEG
TlJdwwmr1r6nJ62Xxo4xs1wL5XCTYWVVJT2+zLKbrK+vnpu2zrjI2qfLHls6KpB38JLQ+wnTNFwX
eNyItF0X3FMqJNcVpcE4yXrsq/HyJcxu2svgWaqa0LTOI6eQexDMQ1UoS8Ib2w38PT23oeWwnP/4
DdIbZY1uSPt3e7V9d0PWKaPDgeCMHdZ8UXXErZBY8xBklFNpLI0C20Ebnb7lB8eVsU8O293Ztiow
067aBrF1UN45fJsv38YYoDtrcp7XHewh7/T+5wy10hSmC6Urh5BwpXRhtg7nfwMXR6pEwn49sOTi
Ihme/swXCSoTfvC1EUdOIhVIjq5vJ9OHu8ni+/HhpOwE7BTsPdgHsI9gn8DOwD6DnSdxtNcJ9nE0
/k+1UaDGe+IfeRjJ8HQCAAA=

- ------=_NextPart_000_002D_01C27C1F.BCCCBD40--
------- End of forwarded message -------

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Sat, 26 Oct 2002 16:16:08 -0400, Richard Stallman <rms@gnu.org> wrote:

> Anyone interested in working on this?

I'll talk with Francis.

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

> I suggest the following solution.  Modify the NTEmacs installation procedure
> to include the file rcs2log in the bin directory, and also include in the
> bin directory the attached file rcs2log.bat (with any modifications deemed
> appropriate), which NTEmacs can execute directly and which acts as an
> interface to rcs2log.  I have written rcs2log.bat to be as portable across
> different versions of Windows as possible, without losing the advantages of
> NT.  However, I have tested it only on Windows XP, where it seems to work
> well.

> Of course, this all assumes a suitable UNIX-like environment on the Windows
> platform, and I use Cygwin, which works well.  I think it is reasonably
> likely that anyone trying to execute the Emacs command
> `vc-update-change-log' in NTEmacs will have a suitable environment for
> executing UNIX shell scripts.

I also think it's reasonable, provided we document it somewhere... CVS
is an Unixish tool that few (if any) Windows users will have in their
environment unless they've installed it. But it'd be good to document it,
or at least talk with Steve Kemp and get it added to the Emacs Windows
FAQ.

I don't entirely like, though, the requeriment that the unix-compatible
sort be found on the PATH before any system one... I like to put
c:\bin\cygwin\bin, c:\bin\mingw\bin and the like *after*
c:\winnt\system32, c:\winnt, etc. and wouldn't like to be forced to
change that just to use rcs2log...

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

On Mon, 28 Oct 2002, Juanma Barranquero wrote:

> I don't entirely like, though, the requeriment that the unix-compatible
> sort be found on the PATH before any system one... I like to put
> c:\bin\cygwin\bin, c:\bin\mingw\bin and the like *after*
> c:\winnt\system32, c:\winnt, etc. and wouldn't like to be forced to
> change that just to use rcs2log...

You could write a small script fragment that would walk PATH, look for a 
version of `sort' that understood some specific option (e.g., --help), 
and invoke that one explicitly, or barf if not found.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Tue, 29 Oct 2002 08:03:17 +0200 (IST), Eli Zaretskii <eliz@is.elta.co.il> wrote:

> You could write a small script fragment that would walk PATH, look for a 
> version of `sort' that understood some specific option (e.g., --help), 
> and invoke that one explicitly, or barf if not found.

That kind of thing strikes me as being fragile.

Wouldn't it be easier to have a configuration variable
(`cvs-sort-program' or something) and have the user configure it if
needed?

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Juanma Barranquero" <lektu@terra.es>
To: "Eli Zaretskii" <eliz@is.elta.co.il>
Cc: <emacs-devel@gnu.org>; "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
Sent: Tuesday, October 29, 2002 7:30 AM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> On Tue, 29 Oct 2002 08:03:17 +0200 (IST), Eli Zaretskii
<eliz@is.elta.co.il> wrote:
>
> > You could write a small script fragment that would walk PATH, look for a
> > version of `sort' that understood some specific option (e.g., --help),
> > and invoke that one explicitly, or barf if not found.
>
> That kind of thing strikes me as being fragile.
>
> Wouldn't it be easier to have a configuration variable
> (`cvs-sort-program' or something) and have the user configure it if
> needed?


I agree that this would be a more elegant solution.  However, it's not
specific to CVS and in fact I use RCS, although it is specific to vc, so I
suggest the variable name should be vc-sort-program.  If such a new variable
is going to be introduced to support non-UNIX users then I suggest that the
vc code should also be edited to call sh explicitly to interpret the rcs2log
script (as I suggested before) and my .bat file ignored.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Tue, 29 Oct 2002 10:21:57 -0000, "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk> wrote:

> I agree that this would be a more elegant solution.  However, it's not
> specific to CVS and in fact I use RCS, although it is specific to vc, so I
> suggest the variable name should be vc-sort-program.

Yes, you're right. I was writing from the top of my head.

> If such a new variable
> is going to be introduced to support non-UNIX users then I suggest that the
> vc code should also be edited to call sh explicitly to interpret the rcs2log
> script (as I suggested before) and my .bat file ignored.

Could you send a patch so we can test it?

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Stefan Monnier]

> On Tue, 29 Oct 2002 10:21:57 -0000, "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk> wrote:
> 
> > I agree that this would be a more elegant solution.  However, it's not
> > specific to CVS and in fact I use RCS, although it is specific to vc, so I
> > suggest the variable name should be vc-sort-program.
> 
> Yes, you're right. I was writing from the top of my head.
> 
> > If such a new variable
> > is going to be introduced to support non-UNIX users then I suggest that the
> > vc code should also be edited to call sh explicitly to interpret the rcs2log
> > script (as I suggested before) and my .bat file ignored.
> 
> Could you send a patch so we can test it?

Can't we change the installation script instead ?
Such that on W32 systems, rcs2log is installed as rcs2log.sh
and then we install a rcs2log.bat wrapper ?
This way the VC code does not need to know whether rcs2log is implemented
as an sh script or anything else (it really doesn't need to know: I played
with the code and didn't know until now that it was an sh script rather
than a binary or a perl script).


	Stefan

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Tue, 29 Oct 2002 09:47:36 -0500, "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu> wrote:

> Can't we change the installation script instead ?
> Such that on W32 systems, rcs2log is installed as rcs2log.sh
> and then we install a rcs2log.bat wrapper ?
> This way the VC code does not need to know whether rcs2log is implemented
> as an sh script or anything else (it really doesn't need to know: I played
> with the code and didn't know until now that it was an sh script rather
> than a binary or a perl script).

I don't know what's the right way, but as I've said, detecting at
installation time whether there is a GNU-compatible sort in the path and
where is it located seems very fragile to me. Not to mention that it
doesn't adapt very well to situations where you modify the PATH or
add/remove items to it (something I do quite frequently).

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Stefan Monnier]

> > Can't we change the installation script instead ?
> > Such that on W32 systems, rcs2log is installed as rcs2log.sh
> > and then we install a rcs2log.bat wrapper ?
> > This way the VC code does not need to know whether rcs2log is implemented
> > as an sh script or anything else (it really doesn't need to know: I played
> > with the code and didn't know until now that it was an sh script rather
> > than a binary or a perl script).
> 
> I don't know what's the right way, but as I've said, detecting at
> installation time whether there is a GNU-compatible sort in the path and
> where is it located seems very fragile to me. Not to mention that it
> doesn't adapt very well to situations where you modify the PATH or
> add/remove items to it (something I do quite frequently).

Sorry, I replied to the wrong part of the thread.  I was only
replying to the original part of the thread that's trying to solve the
problem that (call-process "rcs2log" ...) doesn't work under
W32 because rcs2log happens to be an sh script.

As for your sorting problem, I have no idea what a vc-sort-program
could do since VC doesn't use `sort' (try `grep sort lisp/vc*.el').
The problem you have should obviously be fixed somewhere else.


	Stefan

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Tue, 29 Oct 2002 10:19:20 -0500, "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu> wrote:

> As for your sorting problem, I have no idea what a vc-sort-program
> could do since VC doesn't use `sort' (try `grep sort lisp/vc*.el').

No, but vc uses rcs2log which in fact seems to need sort.

> The problem you have should obviously be fixed somewhere else.

Yeah, but I can't think of a clear way.


                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Stefan Monnier]

> > As for your sorting problem, I have no idea what a vc-sort-program
> > could do since VC doesn't use `sort' (try `grep sort lisp/vc*.el').
> No, but vc uses rcs2log which in fact seems to need sort.

No, VC does not *use* rcs2log AFAIK.  It does provide some way to run it
for you, tho.  But it can also be run separately from Emacs, so the
problem with `sort' should not be fixed in any elisp file, but in rcs2log
(or maybe in rcs2log.bat instead.  That would be even better so that
all this ugly w32 business doesn't stain the rest of the code).


	Stefan

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
To: "Juanma Barranquero" <lektu@terra.es>
Cc: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>; "Dr Francis J.
Wright" <F.J.Wright@qmul.ac.uk>; "Eli Zaretskii" <eliz@is.elta.co.il>;
<emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 3:43 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> > > As for your sorting problem, I have no idea what a vc-sort-program
> > > could do since VC doesn't use `sort' (try `grep sort lisp/vc*.el').
> > No, but vc uses rcs2log which in fact seems to need sort.
>
> No, VC does not *use* rcs2log AFAIK.  It does provide some way to run it
> for you, tho.  But it can also be run separately from Emacs, so the
> problem with `sort' should not be fixed in any elisp file, but in rcs2log
> (or maybe in rcs2log.bat instead.  That would be even better so that
> all this ugly w32 business doesn't stain the rest of the code).
>
>
> Stefan


I see the way that vc uses rcs2log as being the same as the way that hexl
mode uses an external helper program.  Here are two possible ways forward.

One is Eli's suggestion, namely a more sophisticated replacement for my
rcs2log.bat file that deduces the correct search path to use for sort.
Clearly this should not use any non-standard Windows technology other than
what is necessary anyway to run rcs2log.  (So, for example, a perl script
would not be not a good solution, although using Windows Script Host might
be.)  A good solution would be to pull the appropriate Cygwin path out of
the Registry, but I'm not sure how feasible that is.

The other is to modify vc.el to allow the user to specify a directory
containing the correct sort program and to prepend that to exec-path before
trying to execute rcs2log.  This could be made to happen only on Windows
platforms.

I'll think about these two options and see whether I can produce a solution
that people find more acceptable than my original rcs2log.bat solution.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Stefan Monnier]

> I see the way that vc uses rcs2log as being the same as the way that hexl
> mode uses an external helper program.

There's a subtle difference: hexl-mode is completely unusable without the
`hexl' program, whereas VC is fully functional even without rcs2log.
I.e. support for rcs2log is an added feature to VC, but it is not the
main purpose of VC.

> One is Eli's suggestion, namely a more sophisticated replacement for my
> rcs2log.bat file that deduces the correct search path to use for sort.
> Clearly this should not use any non-standard Windows technology other than
> what is necessary anyway to run rcs2log.  (So, for example, a perl script
> would not be not a good solution, although using Windows Script Host might
> be.)  A good solution would be to pull the appropriate Cygwin path out of
> the Registry, but I'm not sure how feasible that is.
> 
> The other is to modify vc.el to allow the user to specify a directory
> containing the correct sort program and to prepend that to exec-path before
> trying to execute rcs2log.  This could be made to happen only on Windows
> platforms.

I think modifying vc.el is completely wrong.  Right now vc.el doesn't know
anything about rcs2log apart from some arguments it can take and what
things it might output.  It doesn't know it's written in `sh', it doesn't
know it uses `sort'.  And indeed, rcs2log might be rewritten in Perl
some day and stop using `sort' and `sh'.

Also rcs2log is actually not particularly linked to Emacs (which
is why it's also distributed along with CVS), so any fix for w32
should allow it to work correctly even when it's not run from Emacs.

> I'll think about these two options and see whether I can produce a solution
> that people find more acceptable than my original rcs2log.bat solution.

I think an rcs2log.bat wrapper is the right approach, I believe.
I actually haven't seen any disagreement with it.  The only problem
mentioned so far has to do with `sort' which is basically unrelated
to the problem at hand (except for the fact that it is related to the
same script and to the same platform).


	Stefan

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

> From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
> Date: Tue, 29 Oct 2002 14:22:13 -0500
> 
> I think modifying vc.el is completely wrong.

I think there's a misunderstanding: I didn't suggest to modify vc.el,
I suggested to modify the rcs2log script itself to make it find the
right `sort' executable.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
To: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
Cc: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>; "Juanma
Barranquero" <lektu@terra.es>; "Eli Zaretskii" <eliz@is.elta.co.il>;
<emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 7:22 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> There's a subtle difference: hexl-mode is completely unusable without the
> `hexl' program, whereas VC is fully functional even without rcs2log.
> I.e. support for rcs2log is an added feature to VC, but it is not the
> main purpose of VC.

OK.  Point taken.

> I think modifying vc.el is completely wrong.  Right now vc.el doesn't know
> anything about rcs2log apart from some arguments it can take and what
> things it might output.  It doesn't know it's written in `sh', it doesn't
> know it uses `sort'.  And indeed, rcs2log might be rewritten in Perl
> some day and stop using `sort' and `sh'.

If anyone were going to rewrite rcs2log then I would suggest it should be
rewritten in Emacs Lisp.  Bit it's probably not worth the effort.

> Also rcs2log is actually not particularly linked to Emacs (which
> is why it's also distributed along with CVS), so any fix for w32
> should allow it to work correctly even when it's not run from Emacs.
>
> > I'll think about these two options and see whether I can produce a
solution
> > that people find more acceptable than my original rcs2log.bat solution.
>
> I think an rcs2log.bat wrapper is the right approach, I believe.
> I actually haven't seen any disagreement with it.  The only problem
> mentioned so far has to do with `sort' which is basically unrelated
> to the problem at hand (except for the fact that it is related to the
> same script and to the same platform).

I agree, which is why I proposed the wrapper approach in the first place.
Any customization can be done with environment rather than Lisp variables.
I will send a revised version that does that shortly.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Richard Stallman]

    There's a subtle difference: hexl-mode is completely unusable without the
    `hexl' program, whereas VC is fully functional even without rcs2log.
    I.e. support for rcs2log is an added feature to VC, but it is not the
    main purpose of VC.

The converse is true also.  rcs2log is useful on its own, but hexl is not.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

> From: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
> Date: Tue, 29 Oct 2002 17:06:27 -0000
> 
> One is Eli's suggestion, namely a more sophisticated replacement for my
> rcs2log.bat file that deduces the correct search path to use for sort.
> Clearly this should not use any non-standard Windows technology other than
> what is necessary anyway to run rcs2log.

None is needed.  You can iterate thru PATH with a FOR builtin, for
example.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

[-- Attachment #1: Type: text/plain, Size: 1716 bytes --]

From: "Eli Zaretskii" <eliz@is.elta.co.il>
To: <F.J.Wright@qmul.ac.uk>
Cc: <emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 7:50 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> > From: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
> > Date: Tue, 29 Oct 2002 17:06:27 -0000
> >
> > One is Eli's suggestion, namely a more sophisticated replacement for my
> > rcs2log.bat file that deduces the correct search path to use for sort.
> > Clearly this should not use any non-standard Windows technology other
than
> > what is necessary anyway to run rcs2log.
>
> None is needed.  You can iterate thru PATH with a FOR builtin, for
> example.


I think that's probably overkill.  Here is a simpler approach that seems to
work well for me.  It requires only that a suitable sh program is found in
the search path used within Emacs.  If the environment variable VC_SORT_PATH
is set to a search path appropriate to the sh program then it is prepended
to PATH; otherwise /bin is prepended to PATH.  In a reasonably recent
default Cygwin installation, sort is in /bin (and Cygwin knows where its own
root directory is within the Windows file system).  Hence I think this
should work correctly by default for most users of Cygwin and probably any
other sensible UNIX-like environment.  If it doesn't then the user can
specify an appropriate search path by setting VC_SORT_PATH.  This could in
fact be used to specify the location of any of the programs run by rcs2log,
although I think only sort clashes with the name of a standard Windows
program and on my system they are all in /bin.  (In fact, they appear to be
in /usr/bin, but /bin is mounted on /usr/bin by default and the files are
physically in /bin.)

Francis

[-- Attachment #2: rcs2log.bat.gz --]
[-- Type: application/x-gzip, Size: 506 bytes --]

Re: rcs2log.bat

[Dr Francis J. Wright]

From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
To: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
Sent: Wednesday, October 30, 2002 4:00 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> Looks good to me (except that it would have been easier to read if
> it weren't gzipped).

I didn't want it to get mangled in transit.

> Oh, and don't use `VC_SORT_PATH': rcs2log is not vc-specific, so using
> the `VC' prefix there is just confusing.
>
> As for re-implementing rcs2log in elisp, I agree that it wouldn't
> make much sense.  I'd much rather have support for cvs2cl in vc-cvs.el.

Whoever installs rcs2log.bat into the CVS repository is welcome to change
that variable name to something more appropriate.  Perhaps `UNIX_SORT_PATH'
would be better.

If there is a similar problem with cvs2cl under Windows (I don't know since
I don't use CVS) then presumably it could be solved by a similar wrapper.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

> From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
> Date: Tue, 29 Oct 2002 10:19:20 -0500
> 
> the problem that (call-process "rcs2log" ...) doesn't work
> under W32 because rcs2log happens to be an sh script.

Perhaps we should fix the Windows version of call-process so that it
could run shell scripts automatically.  The DJGPP (a.k.a. DOS) port,
for example, already does that.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Eli Zaretskii" <eliz@is.elta.co.il>
To: <monnier+gnu/emacs@rum.cs.yale.edu>
Cc: <F.J.Wright@qmul.ac.uk>; <emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 7:47 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> > From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
> > Date: Tue, 29 Oct 2002 10:19:20 -0500
> >
> > the problem that (call-process "rcs2log" ...) doesn't work
> > under W32 because rcs2log happens to be an sh script.
>
> Perhaps we should fix the Windows version of call-process so that it
> could run shell scripts automatically.  The DJGPP (a.k.a. DOS) port,
> for example, already does that.

That might be a good idea, especially if the necessary code is essentially
already there in the DJGPP port.  But it wouldn't solve the path problem
unless it also tweaked the path.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
To: "Juanma Barranquero" <lektu@terra.es>
Cc: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>; "Eli Zaretskii"
<eliz@is.elta.co.il>; <emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 2:47 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> > On Tue, 29 Oct 2002 10:21:57 -0000, "Dr Francis J. Wright"
<F.J.Wright@qmul.ac.uk> wrote:
> >
> > > I agree that this would be a more elegant solution.  However, it's not
> > > specific to CVS and in fact I use RCS, although it is specific to vc,
so I
> > > suggest the variable name should be vc-sort-program.
> >
> > Yes, you're right. I was writing from the top of my head.
> >
> > > If such a new variable
> > > is going to be introduced to support non-UNIX users then I suggest
that the
> > > vc code should also be edited to call sh explicitly to interpret the
rcs2log
> > > script (as I suggested before) and my .bat file ignored.
> >
> > Could you send a patch so we can test it?
>
> Can't we change the installation script instead ?
> Such that on W32 systems, rcs2log is installed as rcs2log.sh
> and then we install a rcs2log.bat wrapper ?
> This way the VC code does not need to know whether rcs2log is implemented
> as an sh script or anything else (it really doesn't need to know: I played
> with the code and didn't know until now that it was an sh script rather
> than a binary or a perl script).
>
>
> Stefan

I'll be happy to suggest a patch once we have agreed what we want to
achieve.

Why do you want to rename rcs2log to rcs2log.sh?  That seems to me to make
the Windows installation more non-standard than necessary.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Stefan Monnier]

> Why do you want to rename rcs2log to rcs2log.sh?  That seems to me to make
> the Windows installation more non-standard than necessary.

I just assumed that if rcs2log stays as it is, it might be run
instead of rcs2log.bat.  Whereas adding a .sh makes sure it won't
be considered as an executable, so rcs2log.bat will be run instead.
I agree that if it's not necessary, there's no point in making that
change.


	Stefan

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Dr Francis J. Wright]

From: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>
To: "Dr Francis J. Wright" <F.J.Wright@qmul.ac.uk>
Cc: "Stefan Monnier" <monnier+gnu/emacs@rum.cs.yale.edu>; "Juanma
Barranquero" <lektu@terra.es>; "Eli Zaretskii" <eliz@is.elta.co.il>;
<emacs-devel@gnu.org>
Sent: Tuesday, October 29, 2002 4:40 PM
Subject: Re: [F.J.Wright@qmul.ac.uk: rcs2log]


> > Why do you want to rename rcs2log to rcs2log.sh?  That seems to me to
make
> > the Windows installation more non-standard than necessary.
>
> I just assumed that if rcs2log stays as it is, it might be run
> instead of rcs2log.bat.  Whereas adding a .sh makes sure it won't
> be considered as an executable, so rcs2log.bat will be run instead.
> I agree that if it's not necessary, there's no point in making that
> change.


At first, I had exactly the same thought.  But a file with no extension is
never executable under Windows (as far as I am aware) and experiment
confirmed that it was not necessary to rename rcs2log.

Francis

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

> Date: Tue, 29 Oct 2002 08:30:00 +0100
> From: Juanma Barranquero <lektu@terra.es>
> 
> > You could write a small script fragment that would walk PATH, look for a 
> > version of `sort' that understood some specific option (e.g., --help), 
> > and invoke that one explicitly, or barf if not found.
> 
> That kind of thing strikes me as being fragile.

Why is it fragile?

> Wouldn't it be easier to have a configuration variable
> (`cvs-sort-program' or something) and have the user configure it if
> needed?

This would require users to customize the variable.  Without a proper
customization, the relevant feature will be broken.  So it's IMHO
better to find the right `sort' automatically.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Tue, 29 Oct 2002 22:39:53 +0300, "Eli Zaretskii" <eliz@is.elta.co.il> wrote:

> Why is it fragile?

Because you are executing programs that you don't control (what if
trying a sort.exe causes a crash or any other kind of problem), and
because probably it'd fail if I happen to modify the PATH afterwards.
If I have to manually configure a variable and after a while I change
the path I'm more prone to remember it that if Emacs somehow configured
it behind my back.

> This would require users to customize the variable.  Without a proper
> customization, the relevant feature will be broken.

Sure. But we already have cases like these now, and they're resolved by
having -program variables. The first "find.exe" in my PATH is
C:\WINNT\system32\find.exe. Same for ftp and telnet. Ange-FTP didn't try
to find what was the "right" ftp, I manually configured
`ange-ftp-ftp-program'.

Having good defaults helps here.

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Eli Zaretskii]

On Wed, 30 Oct 2002, Juanma Barranquero wrote:

> On Tue, 29 Oct 2002 22:39:53 +0300, "Eli Zaretskii" <eliz@is.elta.co.il> wrote:
> 
> > Why is it fragile?
> 
> Because you are executing programs that you don't control (what if
> trying a sort.exe causes a crash or any other kind of problem)

We already run programs from various Emacs features, without fear of a 
crash.  Also, running programs to probe their functionality is a standard 
way of testing things in configure scripts.  So I don't see anything 
especially fragile here.

> and
> because probably it'd fail if I happen to modify the PATH afterwards.

??? I suggested to do this as part of the rcs2log script itself.  So PATH 
changes will be noted the next time the script runs, and acted upon at 
that very moment.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Juanma Barranquero]

On Wed, 30 Oct 2002 16:10:44 +0200 (IST), Eli Zaretskii <eliz@is.elta.co.il> wrote:

> So I don't see anything  especially fragile here.

Well, I do. Anyway, when I say "fragile" I'm thinking more in the fact
that the detection method works for relatively stable environments. I
often modify my PATH and also quite often I have several CMD sessions
open with different PATHs. I'd much prefer to configure a variable (be
that an Emacs variable or an environment variable) and have it set up
forever.

Still, you say:

> ??? I suggested to do this as part of the rcs2log script itself.  So PATH 
> changes will be noted the next time the script runs, and acted upon at 
> that very moment.

Detecting which sort is the correct one *every* time rcs2log runs seems
weird to me.

And still holds true what I've said: we have quite a few -program
variables for very similar reasons. sql-mode doesn't do any detecting
'round to find if I have sqlplus.exe or plus80.exe or whatever.

                                                           /L/e/k/t/u

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Francesco Potorti`]

Juanma Barranquero <lektu@terra.es> writes:

>And still holds true what I've said: we have quite a few -program
>variables for very similar reasons. sql-mode doesn't do any detecting
>'round to find if I have sqlplus.exe or plus80.exe or whatever.

I do not know if that helps, but for example man.el does autodetection
to choose from different syntax sed scripts.  It has a variable
Man-sed-script that is initialised the first time man is called in an
Emacs sssion.  

You could set up a -program variable as a defvar inited to nil.  The
first time your function is called in an Emacs session it checks whether
the -program variable is initialized and, if not, it does autodetection
and initializes it.  This way users can force the value of the variable
in their .emacs file.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Richard Stallman]

    We already run programs from various Emacs features, without fear of a 
    crash.  Also, running programs to probe their functionality is a standard 
    way of testing things in configure scripts.

It is undesirable to do this in configure scripts because that makes
cross-compilation impossible.  Doing it when running Emacs is better
because it does not cause such a problem.

Re: [F.J.Wright@qmul.ac.uk: rcs2log]

[Richard Stallman]

    Wouldn't it be easier to have a configuration variable
    (`cvs-sort-program' or something) and have the user configure it if
    needed?

That way, each user would have to set the variable to get proper
results.  That is very very undersirable.  To make the system really
"work" means making it work without the user's having to know what
variable to set.