[pacman-dev] [PATCH] scripts/library: rewrite parse_options

[Dave Reisner]

On Thu, Aug 18, 2011 at 08:57:55AM +1000, Allan McRae wrote:
> On 18/08/11 08:27, Dan McGee wrote:
> >On Thu, Jul 21, 2011 at 3:40 PM, Dave Reisner<d at falconindy.com>  wrote:
> >>In addition to being what I feel is a cleaner and faster implementation,
> >>we avoid the use of eval by normalizing option arguments into a global
> >>array which is then set after a successful call to parse_options.
> >>
> >>This trims out the idea of having multiple arguments to a single option,
> >>making our parsing algorithm a little more sane. We never took advantage of
> >>this in makepkg (for the one option that feasibly supports it), and I
> >>think we've overlooked a much simpler solution in pacman-key. Since
> >>actions are limited to 1 at a time the leftover positional parameters
> >>become the keys or keyfiles which are acted upon.
> >>
> >>Also added is a new test directory test/scripts with a harness for
> >>parse_options, run as part of make check.
> >>
> >>Signed-off-by: Dave Reisner<dreisner at archlinux.org>
> >>---
> >>Thoughts? I know Allan wasn't quite sold on this, as the downside is that we
> >>sort of pigeonhole ourselves into using the non-optional parameter as arguments
> >>to our action. This has zero effect on makepkg.
> >>
> >>I've also thought to add --option=arg syntax parsing, but I'm not sure we need
> >>this.
> >
> >Allan, was deferring to you on this.
> 
> 
> Well, I was deferring to you as Dave is right that I was never sold
> on this...
> 
> Unless I am missing something, this does have a minor effect on
> makepkg.  In git "makepkg --pkg foo bar" builds only foo and bar
> from a split package.

And, imo, this introduces bizzare unexpected behavior. With the
"standard" getopt{,_long}, passing something such as:

  --pkg foo bar

I'd expect bar to be completely unrelated to the flag. Not the case
here, and this behavior isn't really documented clearly at all. We don't
even properly handle arguments with whitespace anymore. Not such a big
deal for makepkg, but I think it's reasonable that pacman-key might some
day need to import a key from a file with a space in the name.

> I guess with this patch we would have to quote the package names in
> some way (like is needed on the current maint release).  So we would
> need to revert changes made to the makepkg documentation when the
> multiple arguments stuff was added.

We can (should?) follow pacman here and separate multiple arguments by
commas. I'm going to channel Dan and call out "consistency" here.

> But my main issue is that we lose the ability to specify multiple
> arguments with optional parameters.  We do not use this anywhere at
> the moment, so it is not a big loss but that is not to say we will
> never use it in the future.

My problem with this is that you're then forced, in a calling
application, to change the behavior of your parsing of normalized
options. It's a little strange.

> Basically, I am not sure what the actual advantage of rewriting this is...

Faster, cleaner, safer (no eval!!), and more standard. That's what I'm
selling here.

d