vc-cvs-global-switches and -f

vc-cvs-global-switches and -f

[Juanma Barranquero]

I was running "emacs.exe -Q" today and was puzzled to see that the
output of `vc-diff' talked about "conflicting specifications of output
style", until I remembered that I have an entry on ~/.cvsrc which
says: "diff -u2", and the default style for VC is "-c".

So the question is: why does Emacs by default assume that it is safe
to use .cvsrc? It seems  to me that having "-f" as default for
`vc-cvs-global-switches' would be more logical, specially because in a
situation like mine above, having it set on my .emacs wouldn't have
helped.

-- 
                    /L/e/k/t/u

Re: vc-cvs-global-switches and -f

[Stefan Monnier]

> I was running "emacs.exe -Q" today and was puzzled to see that the
> output of `vc-diff' talked about "conflicting specifications of output
> style", until I remembered that I have an entry on ~/.cvsrc which
> says: "diff -u2", and the default style for VC is "-c".

> So the question is: why does Emacs by default assume that it is safe
> to use .cvsrc? It seems  to me that having "-f" as default for
> `vc-cvs-global-switches' would be more logical, specially because in a
> situation like mine above, having it set on my .emacs wouldn't have
> helped.

I for one would be annoyed to have to re-specify in Emacs the options I've
already set globally in my ~/.cvsrc.

In PCL-CVS I slove this dilemma by parsing the ~/.cvsrc file in order to set
the default value of config vars.


        Stefan

Re: vc-cvs-global-switches and -f

[Juanma Barranquero]

On 6/21/05, Stefan Monnier <monnier@iro.umontreal.ca> wrote:

> I for one would be annoyed to have to re-specify in Emacs the options I've
> already set globally in my ~/.cvsrc.

You can always set `vc-cvs-global-switches' to nil in your .emacs. The
situation I'm trying to fix is precisely the one where the user is not
using local configs (i.e., emacs -Q) and there are .cvsrc entries
which conflict with Emacs defaults (as it happened with my "diff -u2"
and the "-c" default of VC).

So my proposed fix does not incommode you, other than forcing you to
add an entry to your .emacs file *once*, while the default requires
the user to remember doing "M-x set-variable vc-cvs-global-switches
RET nil RET" *every time* he starts Emacs without loading .emacs.

-- 
                    /L/e/k/t/u

Re: vc-cvs-global-switches and -f

[Stefan Monnier]

> RET nil RET" *every time* he starts Emacs without loading .emacs.

Are you speaking seriously here?
The "emacs -Q" scenario is not a normal use scenario, AFAIK, so its
convenience doesn't really matter that much.  At least not nearly as much as
the convenience of the normal situation.


        Stefan

Re: vc-cvs-global-switches and -f

[Juanma Barranquero]

> Are you speaking seriously here?
> The "emacs -Q" scenario is not a normal use scenario, AFAIK, so its
> convenience doesn't really matter that much.  At least not nearly as much as
> the convenience of the normal situation.

Let's put it other way: even on normal situation, it seems safer to
make VC *not* use customized defaults it knows nothing about (and
which can cause trouble), and let the user who uses a .cvsrc and knows
what he's doing to put a simple

  (setq vc-cvs-global-switches nil)

on their .emacs.

For my use case to happen, it is not necessary to use -Q, it's just
enough to have a .cvsrc with "diff -u" and not having configured any
vc-cvs-* entry on .emacs.

-- 
                    /L/e/k/t/u

Re: vc-cvs-global-switches and -f

[Andre Spiegel]

On Wed, 2005-06-22 at 05:51 +0200, Juanma Barranquero wrote:

> Let's put it other way: even on normal situation, it seems safer to
> make VC *not* use customized defaults it knows nothing about (and
> which can cause trouble), and let the user who uses a .cvsrc and knows
> what he's doing to put a simple
> 
>   (setq vc-cvs-global-switches nil)
> 
> on their .emacs.

If we put implementation issues aside for the moment, I think the
following behaviour would be desirable for VC:

(1) Settings from .cvsrc should be effective during VC commands, unless
    a particular VC command would not work with that setting, in 
    which case VC must override it

(2) Customizations that the user makes inside Emacs should override
    settings in .cvsrc.  In some cases, it may be good or tolerable to 
    see a warning message if the settings are inconsistent, e.g. in 
    your case where Emacs and .cvsrc don't agree on the diff format.

(3) Generally, users who customize VC settings or who create .cvsrc 
    files can be considered advanced users, and they can be bothered 
    with some effort to make sure that both kinds of settings are
    consistent.

Given this, I don't like the idea of making "-f" the default in VC,
because it violates (1).

A real solution would require something along the lines of Stefan's
parsing of .cvsrc from within Emacs.  Stefan, do you see a possibility
that VC could use your code from pcvs.el to do this?

However, given that we are in a feature freeze, and given (3), and given
the fact that this is a rather exotic situation, I don't think we should
change anything right now.

Re: vc-cvs-global-switches and -f

[Juanma Barranquero]

> (1) Settings from .cvsrc should be effective during VC commands, unless
>     a particular VC command would not work with that setting, in
>     which case VC must override it

I'm not going to fight over vc-cvs settings (no, really :-) but if we
were creating VC from scratch, I would push for a
`vc-cvs-use-cvsrc-settings', defaulting to nil. The equivalent
functionality as of today is vc-cvs-global-settings '("-f").

> (2) Customizations that the user makes inside Emacs should override
>     settings in .cvsrc.  In some cases, it may be good or tolerable to
>     see a warning message if the settings are inconsistent, e.g. in
>     your case where Emacs and .cvsrc don't agree on the diff format.

Just a question: how do you do Emacs override a setting? Because if
you've got "update -dP", how you would make Emacs do just "update -d"
without passing -f?

> (3) Generally, users who customize VC settings or who create .cvsrc
>     files can be considered advanced users, and they can be bothered
>     with some effort to make sure that both kinds of settings are
>     consistent.

I'm not entirely sure why being knowledgeable on CVS should make you
knowledgeable on Emacs.

> However, given that we are in a feature freeze, and given (3), and given
> the fact that this is a rather exotic situation, I don't think we should
> change anything right now.

Other than the (3) part, yeah, I agree with this.

-- 
                    /L/e/k/t/u

Re: vc-cvs-global-switches and -f

[Andre Spiegel]

> > (2) Customizations that the user makes inside Emacs should override
> >     settings in .cvsrc.  In some cases, it may be good or tolerable to
> >     see a warning message if the settings are inconsistent, e.g. in
> >     your case where Emacs and .cvsrc don't agree on the diff format.
> 
> Just a question: how do you do Emacs override a setting? Because if
> you've got "update -dP", how you would make Emacs do just "update -d"
> without passing -f?

As I said, I was not referring to the current implementation.
Currently, Emacs cannot do this, and it likely requires "-f", along with
a way to parse .cvsrc from within Emacs as pcvs.el does it.

I was just saying that this seems the most reasonable behaviour,
regardless of whether it can easily be implemented (especially during a
feature freeze).

Re: vc-cvs-global-switches and -f

[Stefan Monnier]

> I'm not going to fight over vc-cvs settings (no, really :-) but if we were
> creating VC from scratch, I would push for a `vc-cvs-use-cvsrc-settings',
> defaulting to nil. The equivalent functionality as of today is
> vc-cvs-global-settings '("-f").

I'm curious why you think so.  What kind of settings do you put in ~/.cvsrc?
The nature of the ~/.cvsrc file is to contain settings which should
basically always apply.

I can't believe you'd spend so much time on this thread only for the -c/-u
thingy with `cvs diff'.  BTW, I consider it a bug in CVS that it complains
when ~/.cvsrc says -u but the command line says -c (or vice-versa): the
command line arg should take precedence.


        Stefan

Re: vc-cvs-global-switches and -f

[Juanma Barranquero]

> I'm curious why you think so.  What kind of settings do you put in ~/.cvsrc?
> The nature of the ~/.cvsrc file is to contain settings which should
> basically always apply.

Nothing special:

cvs -z9
diff -u2
tag -c
edit -c
update -dP
release -d
status -v

"release -d", for example, is something I don't want a program to do
for me, only when I explicitly type "cvs remove xxxx" on the command
line.

As I said earlier in the thread, this does not cause a problem for me;
I have my customizations on .emacs and .cvsrc, and I knew enough to do
M-x set-variable vc-cvs-global-options '("-f"). It just strikes me as
a bad default.

> BTW, I consider it a bug in CVS that it complains
> when ~/.cvsrc says -u but the command line says -c (or vice-versa): the
> command line arg should take precedence.

Yes, I agree with that. But even if it didn't complain, I would still
think the same (but I wouldn't have realized VC was using .cvsrc :-)

-- 
                    /L/e/k/t/u