Re: [PATCH 9/5] Migrate git-checkout.sh to use git-rev-parse --parseopt --keep-dashdash
- From
Pierre Habouzit <madcoder@debian.org>
- Date
- Nov 4, 2007, 14:18 UTC
- Message-ID
- <20071104141809.GG26269@artemis.corp>
- In-Reply-To
- <Pine.LNX.4.64.0711041343470.4362@racer.site>
On Sun, Nov 04, 2007 at 01:48:36PM +0000, Johannes Schindelin wrote:
Show 10 quoted lines
> > Pierre Habouzit <madcoder@debian.org> writes: > > It takes on the standard input the specification of the options to parse > > and understand, and echoes on the standard ouput a line suitable for > > `sh(1)` `eval` to replace the arguments with normalized ones. > > Why not go the full nine yards and output something which when eval'ed > sets the variables correctly (taking the variable names from the option > names; long name if available, otherwise short one)? It can also set the > command line arguments to what's left after option parsing, with a "set" > call.
We could do that, though it's not as great as it looks like at the first glance. If you want -vvv to work like an accumulator, then you need a really more complex approach in the C code. To enter the gory details, git-rev-parse --parseopt uses a callback that deals with options and their arguments one by one, then appends a delimiter to tell the shell script that only arguments follow, and then appends the arguments the option parser left alone.
It does not deal with the semantics that the C has available at all, it's up to the shell script to decide which is better.
My goal with this is not really to do all the work for the shell script author, but rather to the user: it's not because a porcelain is new (or not a builtin yet) that it should have a creepy interface. If it helps the programmer as a side effect, then it's great, but this series really is about usability to me.
Of course we can do what you propose, but it will probably be quite sophisticated and looks to me like an overkill to what shell builtins really are used for: prototyping a new porcelain until it becomes a new full blown C-builtin. I do believe in simplicity after all :)
> And to prevent funny games with "PARSEOPT_OPTS=blabla git xyz", why not > provide a function in git-sh-setup which takes the string as argument, and > is called _after_ sourcing git-sh-setup?
This one is quite a non issue, it's only used for keep-dashdash, and I see no other need in the near future. And if the need arise, it'll still be doable any time. So for now I've taken a non-generic way to do that, and I believe it's fine as it is.
-- ·O· Pierre Habouzit ··O madcoder@debian.org OOO http://www.madism.org