moar examples needed

[nicferrier]

I had fun using this for a matcher for the stuff produced by org-elements... it's pretty tricky because there aren't enough examples.

One particular problem I had was how I could match against something deep in a list, org-elements has stuff like this:

(headline (:raw-value "Jacintha Love" :begin 1 :end 198 :pre-blank 0 :hiddenp nil :contents-begin 17) ...)

How to match just :contents-begin for example?

[VincentToups]

This is similar to the plist discussion going on over on the dash.el repository.

I would probably do something like

(defun extract/c (key) (lexical-let ((key key)) (lambda (plist) (plist-get key plist))))

(match '(headline (:raw-value "Jacintha Love" :begin 1 :end 198 :pre-blank 0 :hiddenp nil :contents-begin 17) ...) 
    ((_ (call (extract/c :contents-begin) value)) value))

You can generalize this into a custom plist matcher using defpattern pretty easily:

 (defpattern plist (&rest kp-pairs) 
     `(and ,@(loop for (k v . rest) on kv-pairs collect `(call (extract/c ,k) ,p))))

Of course this is not so performant for large plists. You could coerce the plist into a hash table and do the same thing using hash lookup if you were so concerned.

[nicferrier]

So pretty much do the matching elsewhere?

I really like dash's end of list mechanism by the way:

(a b . c) as opposed to shadchen's (a b (tail c))

So now I'm wondering whether it might be good to take both approaches and combine them.

[VincentToups]

The more idiomatic match for the tail of a list is list* in accord with the function of the same name. You cannot, unfortunately, use . unless all you want to do is match against a symbol, because syntactically (list a b . (and stringp s)) is read by the reader as(list a b and stringp s). There is no way, post-read-time, to know whether the denotation was intending the last form to be a sub-pattern or a simple list of symbols.

One could forbid the use of symbols denoting patterns in match expressions and correctly interpret (list a b and stringp s) but this doesn't seem like a good solution.

With the plist defpattern defined above, you can match against a plist pretty idiomatically:

(match '(:a 'a :b 'b)
((plist :a (and symbolp x)) (message "matched a symbol at key :a" x)))

The point of extensibility in Shadchen is that any match semantics you might want for custom objects can be given a concise form which hides the details of the pattern match.

[nicferrier]

Ah, right. Gotcha.

But it would be nice if those things were all there. plist and alist matching out of the box would be fantastic.

Ah well, perhaps I'll have a play and send you some PRs.

[VincentToups]

By the way, I am an enormous fan of you emacs work. I definitely recognize a kindred spirit re making emacs do insane things. You are inspiring me to brush off some of my old projects!

[nicferrier]

Awww. Shucks. I just wish I understood your code. I can't possibly hope to keep up with you.

[VincentToups]

If you can't understand my code, it is definitely my problem, not yours. I would welcome pull requests and I am in particular interested in modernizing shadchen el a bit. I sort of delayed moving to Emacs 24 for quite a long time, for various reasons, and as such it is kind of ugly or at least old looking, I would guess.

Plus, I am a bit embarrassed, as I've said before, about the quality of the code here, so I kind of stopped promoting Shadchen when I realized there was a lot for me to learn about software engineering. I would be nice to go back and clean it up. Thanks for giving me the reminder to do so.

[nicferrier]

I don't think you need to clear it up. If I could give you any advice it would be "if it works, ship it".

I am torn between this and -let. -let is really nice but this is more powerful in the end. I'll get this added to marmalade and then I'll try and send you PRs.

[nicferrier]

So:

(defun extract/c (key)
  (lambda (plist)
    (plist-get key plist)))

(defpattern plist (&rest kv-pairs) 
  `(and ,@(loop for (k v . rest) on kv-pairs
             collect
               `(funcall (extract/c ,k) ,v))))

(match
 '(:one 1 :two 2 :three 3)
 ((plist :two a) a)) => nil

what am I doing wrong?

[nicferrier]

Aha. The definition of extract/c was wrong:

(defun extract/c (key)
  (lambda (plist)
    (plist-get plist key)))

(defpattern plist (&rest kv-pairs) 
  `(and ,@(loop for (k v . rest) on kv-pairs
             collect
               `(funcall (extract/c ,k) ,v))))

(match
 '(:one 1 :two 2 :three 3)
 ((plist :two a) a))

[nicferrier]

... and alists too:

(defun extract/a (key)
  (lambda (alist) (cdr-safe (assoc key alist))))

(defpattern alist (&rest kv-pairs)
  (cl-labels ((alist-get (key)
                         (lambda (alist)
                           (cdr (assoc key alist)))))
    `(and ,@(loop for (k v . rest) on kv-pairs
               collect
                 `(funcall (extract/a ,k) ,v)))))

(match
 '((a . 1)(b . 2)(c . 3))
 ((alist 'c a) a))

I wonder how many other types could be provided for. Windows? Frames?

One other thing about this lib is that it seems like it's a straight port from your CL version. I found a few doc nits but the doc looks the same between the two. Can I submit PRs to this? or do you generate one from the other?

[VincentToups]

Nice work! But I think we also need a by #'cddr on the loop in the pattern
expansion of the plist macro (it was my mistake).

I wish I generated one from the other, but they are divergent enough that
it's not so easy. Plus, there are many great matches in Common Lisp and
fewer in Emacs, so the el version got more of my attention.

What advice can you give me about how people are doing testing in Emacs
Lisp these days? I'd like to build a test suite.
On Oct 8, 2014 1:16 AM, "Nic Ferrier" notifications@github.com wrote:

... and alists too:

(defun extract/a (key)
(lambda (alist) (cdr-safe (assoc key alist))))

(defpattern alist (&rest kv-pairs)
(cl-labels ((alist-get (key)
(lambda (alist)
(cdr (assoc key alist)))))
(and ,@(loop for (k v . rest) on kv-pairs collect (funcall (extract/a ,k) ,v)))))

(match
'((a . 1)(b . 2)(c . 3))
((alist 'c a) a))

I wonder how many other types could be provided for. Windows? Frames?

One other thing about this lib is that it seems like it's a straight port
from your CL version. I found a few doc nits but the doc looks the same
between the two. Can I submit PRs to this? or do you generate one from the
other?

—
Reply to this email directly or view it on GitHub
#4 (comment)
.

[VincentToups]

Now that I am at a real computer, I can test these things out. We want something that looks like this:

(defun extract/c (key)
  (lexical-let ((key key))
      (lambda (plist)
        (plist-get plist key))))

(defpattern plist (&rest kv-pairs) 
  `(and ,@(loop for (k v . rest) on kv-pairs by #'cddr
                collect
                `(funcall (extract/c ,k) ,v))))


(match (list :x 10 :y 11 :z 12)
       ((plist :x x :y y) (list x y)))

[nicferrier]

Great! I'll send a pr on that.

Personally for testing I use ert which is built in, it works pretty well, I see no reason to use anything else.

I have built my own continuous integration stuff for that using a tool I call elpakit. Other people use cask for the same task.

[nicferrier]

I'll take a look at this myself I guess... but to what extent could match be a macro? then one could debug the evaluated forms.

[nicferrier]

You were asking about tests. A complex set of tests is here:

https://github.com/joddie/macrostep/blob/master/macrostep-test.el

[nicferrier]

Is the inspiration for shadchen the matcher in racket? It seems so. If so I'll try to use that as inspiration for any extra patterns I define?

I'd like to add the ___ to the list matcher for example.

[VincentToups]

It is definitely based on Racket's pattern matcher. I will have to refresh my memory on the ___ pattern.

[nicferrier]

It's something that let's you match many things:

(match (list 1 2 3 4 5 6)
  ((list 1 ___ (tail a)) a))) => 6

[VincentToups]

That should probably be (6) rather than 6. Tail performs a pattern match
on the remainder of the list not matched by the previous patterns.

The append pattern does something like this already.

(append ... ) produces the first match
satisfying all the list patterns (there may, in general, be many). Append
has a crazy performance profile because it has to back track and sometimes
blows the stack, though, so it might be better to have a separate pattern.
You would write something like this:

(append (list 1)
(___)
(list last))

And I would expect this pattern to match last to 6 if used against the
value in your example.

If you want to take a crack at making append better behaved, that might be
my ideal way forward.

On Mon, Oct 13, 2014 at 11:06 PM, Nic Ferrier notifications@github.com
wrote:

It's something that let's you match many things:

(match (list 1 2 3 4 5 6)
((list 1 ___ (tail a)) a))) => 6

—
Reply to this email directly or view it on GitHub
#4 (comment)
.

[nicferrier]

Hum. Ok. I'll look.

Slightly related - how about declaring a specific index for the macros?

(match  value
  (matches...))

eg:

(defmacro ...
   (declare (indent 1) ...)  ...)

[nicferrier]

Yeah, so further to your comment, this works:

(match
 (number-sequence 1 10)
 ((append (list 1) _ (list y)) y))  => 10

[VincentToups]

I think its worth looking into the implementation, because it might be sort
of wacky.

I don't object to putting in a (___) pattern which is active just inside of
list matches for that specific case, if it is pretty common. Adding indent
declarations is a great idea, by the way. Thanks for taking so much time
to improve shadchen!

[nicferrier]

I definitely will look into the implementation. Particularly if I can
find a problem that causes it to be a pain.

notifications@github.com writes:

I think its worth looking into the implementation, because it might be sort
of wacky.

I don't object to putting in a (___) pattern which is active just inside of
list matches for that specific case, if it is pretty common. Adding indent
declarations is a great idea, by the way. Thanks for taking so much time
to improve shadchen!

---

Reply to this email directly or view it on GitHub:
#4 (comment)

[nicferrier]

Bizarrely when I come to use it again it doesn't work! weird!

[VincentToups]

What doesn't?

On Wed, Oct 29, 2014 at 8:56 PM, Nic Ferrier notifications@github.com
wrote:

Bizarrely when I come to use it again it doesn't work! weird!

—
Reply to this email directly or view it on GitHub
#4 (comment)
.

[nicferrier]

The append. I need to add a test I guess.

[VincentToups]

It is a kind of tricky thing, concat and append. I have often thought
about removing them. Super handy, but also kind of "stupid programming
tricks". Using Shadchen and a few macros one can easily write a really
nice monadic parsing framework that handles this sort of thing much more
efficiently. So I don't know.

I think making concat and append efficient (or just one, as concat can be
implemented in terms of append) would be neat little project in writing a
performant back tracker in Emacs Lisp.

-V

On Wed, Oct 29, 2014 at 10:40 PM, Nic Ferrier notifications@github.com
wrote:

The append. I need to add a test I guess.

—
Reply to this email directly or view it on GitHub
#4 (comment)
.