Re: [Gnu-arch-users] [BUG] tla dosen't handle '--' as end of option argument

[David Allouche]

On Sun, 2004-09-12 at 08:25 -0500, John Meinel wrote:
> Miles Bader wrote:
> 
> > On Sun, Sep 12, 2004 at 07:50:57AM -0500, John Meinel wrote:
> > 
> [...]
> > 
> > I agree that in cases where it makes sense, the `--' ought to be optional.
> > 
> > However the way tla typically uses `--', it separates file lists (and some
> > other sorts of lists) -- but non-file arguments such as revision names still
> > precede it.  This is non-posixy, but it's in fact damned useful, because it
> > allows one to specify such lists while _still_ omitting preceding non-file
> > arguments (so they get their default value); since these preceding arguments
> > are by virtue of restricted syntax not confusable with options, this is a
> > non-ambiguous, and very handy system.
> > 
> 
> I had forgotten about archive names, etc. Though I thought they all
> followed formats such that you could detect if it was a revision versus
> a file, etc. And that there were some priority rules (at least for some
> commands) such that it checked if it matched a filename first.

Please don't. Non-ambiguity in command-line syntax is essential.

If anything, arguments which look like arch namespace should be
interpreted as such regardless of the existence of a file by the name
name, because file name can be escaped (with "./") while arch namespaces
cannot.

But even that would cause problems when extending the way we can specify
namespaces. For example, it could cause changes in behaviour if the
tree-version is always used as a default (and "patch-42" would become an
acceptable revision specification) or when aliases are introduced.

Say no to ambiguous command lines!

-- 
                                                            -- ddaa