Re: ‘match’ and “k or more” patterns

[Ludovic Courtès]

Hi,

[Re-adding Cc: address@hidden

Alex Shinn <address@hidden> writes:

> On Mon, Sep 6, 2010 at 12:04 AM, Ludovic Courtès <address@hidden> wrote:
>>
>> GNU Guile 1.9 now uses your implementation of ‘match’ as a nice
>> replacement for Wright’s implementation, so thank you!
>>
>> I stumbled upon this incompatibility: Wright’s ‘match’ supports ‘..1’,
>> ‘..2’, etc., which mean “1 or more”, “2 or more”, etc., and the
>> associated variable (when there’s one) is bound to the list that
>> matches:
>>
>>  (match '(a 1 2) (('a x ..1) x))
>>  => (1 2)
>>
>> AFAICS these patterns aren’t implemented in your ‘match’.
>>
>> Do you have plans to implement them?
>
> Yes, these can't be implemented in syntax-rules.

Well, since there are only 9 of them, they could probably be implemented
as special cases, with an augmented ‘match-gen-ellipses’, which would be
told the minimum number of elements expected?

> It would be straightforward to implement an alternate
> syntax such as
>
>   (match '(a 1 2) (('a x .. 1) x))
>
> or generalize it to
>
>   (match '(a 1 2) (('a x <M> .. <N>) x))
>
> where the <N> could be #f or left out to mean infinity,
> which would be strictly more powerful than Wright's
> syntax.

Yes, this would be nice, too.

> The main reason I haven't bothered adding this is
> I've never needed it, and was waiting to hear reports
> from people who do.
>
> Do you have any code which actually uses the ..k
> patterns? :)

I do!  :-)

  http://git.sv.gnu.org/cgit/guile-rpc.git/tree/modules/rpc/compiler.scm#n312

Well it uses only ‘..1’.  The same code would work with ‘..1’ replaced
by ‘...’, but then errors in the input wouldn’t be detected as nicely.

Thanks,
Ludo’.