Re: [Quilt-dev] quilt new overrides SUBDIR

[Joe Green]

Andreas Gruenbacher wrote:
> On Monday 12 July 2004 18:13, Joe Green wrote:
> > How about something like the attached patch?  It assumes that if both
> > the QUILT_PATCHES directory and the SERIES file have been found, the
> > SUBDIR value can be trusted.  Otherwise it generates an error, and the
> > user must specify "-f" to create a new root in the current directory.
> > This seems like it should be reliable, unless maybe QUILT_SERIES is set
> > to an absolute path.
>
> The series file may also be in the QUILT_PATCHES directory, and this is often 
> useful, too.

I agree, that's where I usually put it, but the patch doesn't assume or 
exclude any particular location for the series file.  It uses the value 
of the SERIES variable.

It simply requires that the SERIES file already exist, so it wouldn't 
work until the first patch is created.  I think it's reasonable that the 
 first patch be created in the actual root directory, especially if new 
warns you it's going to do that and makes you specify "-f".

> So checking for both doesn't give a lot more confidence about 
> the directory to choose.

Using the default names, it means you've located an existing directory 
with the name "patches", and an existing file named "series" either in 
that directory, or in the parent directory of "patches", or in a ".pc" 
directory in the parent directory of "patches".  That seems reasonably 
unlikely to occur outside of a quilt tree.

Unless you want to support creating a new root directory when you're 
already in a quilt tree?

It could require that the QUILT_PC directory be present as well.  That 
seems like overkill to me, and it wouldn't work until the first patch is 
applied (as opposed to created), but that would still cover the great 
majority of uses.

> I really wanted to avoid creating patches with `quilt new' above the current 
> directory in the default case.

When you're down deep in a directory tree, it's nice to be able to do 
"new", "edit" and "refresh" on a file right there without having to go 
back to the top of the tree.

Since "new" is one of the only commands that doesn't honor SUBDIR, I 
think users will be surprised when it creates a new root in this 
situation; I was.

> How does an option sound that turns off this
> check, and accepts the next higher-level patches directory it finds? (I'm 
> thinking about a long option, e.g., --check-parent-dirs that people could add 
> to quiltrc when needed.)

If you think it's safer to enable this only with an option, that's OK 
with me, since your recent changes have made it easy to turn this on in 
the quiltrc.  I think it's a good idea to check for the existence of the 
SERIES file even if that option is given, though, and that seems to be 
safe enough for the option to be superfluous.

> > What do you think about making the users always specify "-f" for the
> > first patch, to make sure they really want to create a new root in the
> > current directory?  That could be implemented with small modifications
> > to this patch.
>
> Unfortunately this doesn't solve the problem.

No, that wasn't my intention.  This was a separate suggestion that if 
the other checks fail  (e.g. either QUILT_PATCHES or SERIES doesn't 
exist), you'd have to specify "-f", which would mean it's OK to create a 
new root in the current directory (overriding SUBDIR as it does now, if 
necessary).  Basically, you'd have to specify "-f" for the first patch 
in a tree, in order to confirm that you want to start a new tree.

> I am quite happy that quilt does not require and special setup steps, and I 
> would prefer to keep it that way.

I didn't intend this as a setup step, more as a simple protection from 
having a new "patches", ".pc" and "series" created when that wasn't the 
intention.

The suggested the behavior would be:

  - You'd have to specify "new -f" or "import -f" to create the first
    patch, because no existing root can be located.

  - After that, "new" would be able to reliably locate the root and
    would work safely in subdirectories.

  - If "new" (or "import") ever got into a situation where it couldn't
    locate an existing root (with various checks to be sure it's the
    root), it would exit with an error saying what amounts to "If you
    wish to proceed, I'm going to create a new root in the current
    directory.  Specify -f if that's what you want."

Sorry if this is a bit long; I just wanted to explain my thinking 
clearly.  I'm not trying to force this on you.

--
Joe Green <address@hidden>
MontaVista Software, Inc.