[bug 0.11] bzr ci -m allow no message to be specified

[Jari Aalto]

Martin Pool <mbp at canonical.com> writes:

> On 23/11/2006, at 11:55pm, Matthew Hannigan wrote:
> 
> > On Fri, Nov 24, 2006 at 12:13:51AM +0200, Jari Aalto wrote:
> >> James Westby <jw+debian at jameswestby.net> writes:
> >>
> >>> On (23/11/06 13:53), Jari Aalto wrote:
> >>>> Aaron Bentley <aaron.bentley at utoronto.ca> writes:
> >>>>>
> >>>>> It can happen under any release.  Your command there is "commit
> >>>>> all
> >>>>> files, with the message 'file.txt'", and I doubt that there's an
> >>>>> automatic way to detect that this is not what you meant.
> >>>>
> >>>> Oh, then make this alwasy require argument (like "."), so that
> >>>> this mistake
> >>>> does not happen by accident:
> >>>>
> >>>>     $ bzr ci -m file.txt
> >>>>     Syntax error, needs file or directory "." to commit all
> >>>>
> >>>
> >>> I don't like this idea. I think that the other proposal of asking
> >>> for
> >>> confirmation if the message is the same as a file is a better one.
> >>
> >> If this means interacitve questions, please don't. It will make bzr
> >> inpossible to be used in scripts and backends.
> >
> > That's an important point, but think it's still worth doing.
> > You could error out if not a terminal.
> >
> > The case where the intended commit message is the name of a single
> > file and you want to commit every change must be _vanishingly_
> > rare.
> >
> > The case where you intend to commit just one file but in
> > fact accidentally commit everything is not particularly rare,
> > and the consequences might be severe.  Especially in a script,
> > where the error might not be realised till quite a few commits
> > later.
> 
> I filed this as https://launchpad.net/products/bzr/+bug/73073
> 
> On the other hand, unlike svn, we have an easy and effective uncommit
> command, so perhaps this should just give a warning and proceed.

Not a good idea. In scripts this will go unnoticed. An error is
better.

Jari