Re: [Quilt-dev] Quilt 0.60 is coming

[Jean Delvare]

Hi Raphael,

On Saturday 25 February 2012 12:16:04 am Raphael Hertzog wrote:
> On Fri, 24 Feb 2012, Jean Delvare wrote:
> > Then passing QUILT_PC=.fake_pc is kind of a hack, it works in
> > practice [1] but doesn't look too good to my eyes at least. I am
> > wondering if it would make sense to add a --dumb mode to quilt
> > where it wouldn't look for .pc or series up to the filesystem root,
> > but would just assume we are at the root of the project and trust
> > the QUILT_* environment variables. This would be useful for
> > generating the documentation, and there may be other use cases as
> > well, for example nesting quilt trees. Opinions?
> 
> Looks reasonable but its usage would be very limited IMO.

At least the test suite could benefit from it. Currently all our test 
cases start with "mkdir patches" exactly to inhibit the path traversing.

> I think I saw once a bug report where a user was annoyed by this path
> traversing because it found out a quilt tree somewhere else while he
>  was trying to create one in the current directory.

This rings a bell to me too. Obviously one can work around the problem 
with "mkdir patches" (or "mkdir $QUILT_PATCHES") but a dedicated way 
would look better. This could take the form of a --dumb option as I said 
above, or alternatively the form of a "quilt init" command that would 
merely create empty $QUILT_PC and $QUILT_PATCHES directories.

> Note that in the case of the doc generation, we don't have to
>  traverse anything. The Makefile is already in the directory where
>  the .pc is. So it's important to not read .pc/.quilt_patches in that
>  mode if you want to use it it to fix the problem affecting us during
>  a Debian package build.

Yes, this was my idea of the dumb mode: no path traversal and no reading 
of hints in .pc/.quilt_*. But one could argue that this is too many 
things for one option, and also a good way for the users to mess up 
their tree. I'll have to think about it some more, it's too late for 
0.61 anyway, for which I think we should just go with your original 
proposal (QUILT_PC=.fake_pc).

-- 
Jean Delvare
Suse L3