cl-loop extensibility is impossible with current documented interfaces

cl-loop extensibility is impossible with current documented interfaces

[Daniel Colascione]

[-- Attachment #1: Type: text/plain, Size: 670 bytes --]

cl-loop defines cl-loop-for-handler and cl-loop-handler as public
interfaces for extending the set of keywords that loop supports.
There is no public interface, however, for doing anything useful in
the functions these handlers describe.  We should document
cl--loop-args and, for cl-loop-for-handler, loop-for-bindings,
loop-for-sets, and loop-for-steps so that loop extenders don't have
to understand the cl-macs source to understand what's going on.
Additionally, we should *dynamically* bind loop-for-bindings,
loop-for-sets, and loop-for-steps so that handler functions can
actually modify these variables as they used to be able to do before
lexbind.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: cl-loop extensibility is impossible with current documented interfaces

[Stefan Monnier]

> cl-loop defines cl-loop-for-handler and cl-loop-handler as public
> interfaces for extending the set of keywords that loop supports.
> There is no public interface, however, for doing anything useful in
> the functions these handlers describe.  We should document
> cl--loop-args and, for cl-loop-for-handler, loop-for-bindings,
> loop-for-sets, and loop-for-steps so that loop extenders don't have
> to understand the cl-macs source to understand what's going on.
> Additionally, we should *dynamically* bind loop-for-bindings,
> loop-for-sets, and loop-for-steps so that handler functions can
> actually modify these variables as they used to be able to do before
> lexbind.

Hmm... never realized there was such an extension mechanism.  No wonder
I broke it when I moved the code to lexical-binding.

If we want to document it, I think it would be good to change some of the
existing built-in functionality to make use of it (i.e. move it out of
the built-in code).  Maybe that could be done for the code that handle
looping over key bindings and over hash tables?

Would you like to take a crack at it?  The changes you suggest above
sound OK.  Except all the dynamically-scoped vars we might document
should use the "cl-loop-" prefix.

Do you know of existing code that used this functionality?


        Stefan