Command framework should say which arguments want glob expansion

[Martin Pool]

On Windows, we need to expand globs/wildcards in command line arguments in bzr. This is done ad-hoc for some commands like add, but maybe should be done in the Command framework.

[Alexander Belchenko (bialix)]

glob expansion should be avoided for 'ignore' command.
So if we want to do glob expasion, I'd suggest to
have specal attribute for each command, that control
this behavior on/off.

[µ]

[Jelmer Vernooij (jelmer)]

On Thu, 29 Mar 2007 04:36:11 -0000
Alexander Belchenko <email address hidden> wrote:

> glob expansion should be avoided for 'ignore' command.
> So if we want to do glob expasion, I'd suggest to
> have specal attribute for each command, that control
> this behavior on/off.
There is no way to turn it off on POSIX (as it's done by the shell,
not the application), so I think it makes for it to be on for all
commands on Windows for consistency.

Cheers,

Jelmer

--
Jelmer Vernooij <email address hidden> / Jabber: <email address hidden>
http://samba.org/~jelmer/

[Martin Pool (mbp)]

On 3/29/07, Jelmer Vernooij <email address hidden> wrote:
> > glob expansion should be avoided for 'ignore' command.
> > So if we want to do glob expasion, I'd suggest to
> > have specal attribute for each command, that control
> > this behavior on/off.
> There is no way to turn it off on POSIX (as it's done by the shell,
> not the application), so I think it makes for it to be on for all
> commands on Windows for consistency.

So the glob-expander on Windows would need to be aware of quotes and
not expand quoted globs. That's probably not bad.

I wonder, should it expand globs that don't match anything, or error,
or leave them unexpanded?

  echo unmatched*

will say different things in different unix shells.

--
Martin

[Alexander Belchenko (bialix)]

Jelmer Vernooij пишет:
> On Thu, 29 Mar 2007 04:36:11 -0000
> Alexander Belchenko <email address hidden> wrote:
>
>> glob expansion should be avoided for 'ignore' command.
>> So if we want to do glob expasion, I'd suggest to
>> have specal attribute for each command, that control
>> this behavior on/off.
> There is no way to turn it off on POSIX (as it's done by the shell,
> not the application), so I think it makes for it to be on for all
> commands on Windows for consistency.

I disagree with you. Because as other people often say: "you need
quoting glob to prevent expand by shell". I assume on POSIX
you need to do something like:

$ bzr ignore '$.obj'

and it works.

[µ]

[Jelmer Vernooij (jelmer)]

On Thu, Mar 29, 2007 at 12:42:07PM +0300, Olexandr Byelchenko wrote:
> Jelmer Vernooij ??????????:
> > On Thu, 29 Mar 2007 04:36:11 -0000
> > Alexander Belchenko <email address hidden> wrote:
> >> glob expansion should be avoided for 'ignore' command.
> >> So if we want to do glob expasion, I'd suggest to
> >> have specal attribute for each command, that control
> >> this behavior on/off.
> > There is no way to turn it off on POSIX (as it's done by the shell,
> > not the application), so I think it makes for it to be on for all
> > commands on Windows for consistency.
> I disagree with you. Because as other people often say: "you need
> quoting glob to prevent expand by shell". I assume on POSIX
> you need to do something like:

> $ bzr ignore '$.obj'

> and it works.
Sorry, I should've been more clear. I meant there's no way to turn it
off in POSIX /from the application/.

Cheers,

Jelmer

--
Jelmer Vernooij <email address hidden> - http://jelmer.vernstok.nl/
 12:52:40 up 5 days, 16:22, 1 user, load average: 0.25, 0.32, 0.50

[Alexander Belchenko (bialix)]

Jelmer Vernooij пишет:
> On Thu, Mar 29, 2007 at 12:42:07PM +0300, Olexandr Byelchenko wrote:
>> I assume on POSIX
>> you need to do something like:
>
>> $ bzr ignore '*.obj'
>
>> and it works.
> Sorry, I should've been more clear. I meant there's no way to turn it
> off in POSIX /from the application/.

Because Martin propose to emulate shell expansion behavior on Windows
inside bzrlib, I was trying to say that in this case we need
to turn it off from application, and only on Windows of course.

[µ]

[John A Meinel (jameinel)]

Well, another thing to be aware of is that Windows doesn't respect the single quote as a quoting character. So doing:

bzr commit -m "foo bar baz"

works, but doing

bzr commit -m 'foo bar baz'

tries to commit using the message 'foo, and tries to commit the files bar and baz'.

We have bug 59302 open to remind us of that.

Also, I want to remind everyone that because Windows doesn't handle '*' but *does* handle '"' we have some real problems detecting if the user did or did not quote their request.

Specifically, I wrote a script which was just:
import sys
print sys.argv

And this is what I got:

> python test_args.py
['test_args.py']

> python test_args.py foo bar
['test_args.py', 'foo', 'bar']

> python test_args.py "foo bar"
['test_args.py', 'foo bar'] # Notice that the shell interpreted the "" for us

> python test_args.py 'foo bar'
['test_args.py', "'foo", "bar'"] # Notice that it didn't do anything with ''

> python test_args.py "bar*"
['test_args.py', 'bar*']

> python test_args.py bar*
['test_args.py', 'bar*']

So we have no way of telling the difference between:
  bzr ignore foo*
and
  bzr ignore "foo*"

If we ask users to use single quotes:
  bzr ignore 'foo*'

Then we really should change the command processor to handle single quotes in general
  bzr ignore './foo bar'
  bzr add 'foo bar'
  bzr commit -m 'foo bar'

Or we can say... "Do a reasonably good approximation, but windows cmd.exe is too different to implement true POSIX semantics."
And instead focus our energies on getting bzr good, and Olive/TortoiseBzr working for people on Windows.

[Alexander Belchenko (bialix)]

Kuno Meyer wrote the patch that use heuristic about which types of arguments should be globbed on Windows (i.e. filename* etc.). His patch was marked to resubmit, but apparently he did not sent new version. If heuristic is enough for this purpose, than I think we could close this bug.