[WIP] Coprocessor with evaluated arguments #428
Conversation
Now, instead of coprocessors taking the arguments from the s-expression, it will retrieve the arguments from the (dynamic) environment, meaning they will take evaluated arguments. We have to be careful to wrap coprocessors around lambdas so that it behaves as expected.
| return IO { | ||
| expr: args, | ||
| // TODO: instead of `env` put coprocessor symbol | ||
| expr: env, |
| @@ -10,34 +10,29 @@ fn trie_lang() { | |||
|
|
|||
| install(s, &mut lang); | |||
|
|
|||
| let expr = "(let ((trie (.lurk.trie.new))) | |||
| // TODO wrap .lurk.trie.new, .lurk.trie.lookup and .lurk.trie.insert in lambdas | |||
There was a problem hiding this comment.
I think this should be handled at the coprocessor level. Having this test case is good because it's an example of what the ux should be like for a coprocessor user (here) and implementer (elsewhere in the Trie implementation, which should not require significant — or ideally any — changes to absorb the corrected/improved coprocessor behavior.
|
|
||
| let expr3 = "(.lurk.trie.insert 0x1cc5b90039db85fd519af975afa1de9d2b92960a585a546637b653b115bc3b53 123 456)"; | ||
| let expr3 = "((lambda (_x _y _z) .lurk.trie.insert) .lurk.trie.new 123 456)"; |
There was a problem hiding this comment.
This is fine as an intermediate example of what the coprocessor 'macroexpansion' should potentially look like. I think when we are done, this should revert back to the code I would like to have written in the first place for the trie test:
(let ((trie (.lurk.trie.new))
(.lurk.trie.insert trie 123 456))
| _ => return None, | ||
| } | ||
| } | ||
| _ => return None, |
There was a problem hiding this comment.
This should probably distinguish between the malformed list case and the case in which the list is shorter than anticipated. Certainly the name 'truncate' suggests that a shorter-than-expected list should be returned whole, rather than as None.
If you look at the implementation of fetch_list, which this morally replaces, you'll see an example of the expected behavior.
In other words, I think this should only return None if the input is malformed up to the first len elements.
|
Achieved via LEM |
I've modified coprocessors so that it extracts arguments from the environment instead of the argument list (which are unevaluted). So instead of
(foo-coproc x y z)you do((lambda (_x _y _z) foo-coproc) x y z). I've also updated the Trie coprocessor and showed that this strategy works. All that is needed is that we automatically wrap the coprocessors in these lambdas. I'm thinking maybe add a new method to coprocessors to install these expressions correctly