[pacman-dev] [PATCH 0/3] repo-add desc/depends creation cleanup

[Dave Reisner]

On Wed, Jun 22, 2011 at 06:05:06PM +1000, Allan McRae wrote:
> On 21/06/11 05:46, Dave Reisner wrote:
> >This is basically what I referred to in my last patch about refactoring
> >desc/depends writing. I snuck in an extra winner (the first) to get rid
> >of the insane IFS tweaking and iteration of the .PKGINFO file with a for
> >loop.
> >
> >Dan, I'm not sure if I can break this up any further, but I'm happy to refactor
> >if it would make you more comfortable. I know it's an insanely important code
> >path so I encourage you (and anyone else reading) to put on your nerd glasses
> >and nitpick this.
> >
> >Dave Reisner (3):
> >   repo-add: bashify reading of .PKGINFO file
> >   repo-add: store multi-value fields as arrays
> >   repo-add: use format_entry for all desc/depends fields
> >
> >  scripts/repo-add.sh.in |  105 +++++++++++++++++++++++------------------------
> >  1 files changed, 51 insertions(+), 54 deletions(-)
> >
> 
> No ack.
> 
> Something here is entirely screwed under bash-3.2.  We still have
> people using pacman with that version of bash.  I think it does not
> output any of these fields...
> 
> -%CONFLICTS%
> -%DEPENDS%
> -%GROUPS%
> -%LICENSE%
> -%OPTDEPENDS%
> -%PROVIDES%
> -%REPLACES%
> 
> That looks like anything in an array...

Good catch. I had rebased this at some point and somewhere along the
way, the arrays were no longer declared as arrays. fixed locally.

> To test, I created a repo with all packages in Arch starting with
> "a". Under bash-4.2, there was only a small difference in the
> output:
> 
> e.g. asymptote-2.11-1
> 
>  %OPTDEPENDS%
> -python2: for the xasy GUI
> -python-imaging: for the xasy GUI
> -tix: for the xasy GUI
> +python2:           for the xasy GUI
> +python-imaging:    for the xasy GUI
> +tix:               for the xasy GUI
> 
> This makes the formatting the same as that used in the PKGBUILD.
> Not sure I like that either...
> 
> Allan
> 

I don't like this easier. It looks like I underestimated what the sed
expression did here -- we can normalize whitespace another way though.

Thanks for the test case.

d