Re: [Sbcl-devel] Macroexpansion strips pretty printing info

[Douglas K. <do...@go...>]

> But what if quasiquote expanded to a quasiquote-identity form?  I'm not
> sure if it's being
> proposed for real, but it's interesting nonetheless.
>

This was definitely a real suggestion. I think it's not terribly deceiving
since you can always write a form as non-pretty.

But it sounds like opinions lean toward macroexpand-all just doing what I
had proposed as macroexpand-all-except-quasiquote, is that right?  While
we're saying this is legitimate in account of possible treatment of
quasiquote as a special operator, there is actually no special handler in
either the compiler or the interpreter. We're justifying why it would be ok
for macroexpand-all to decide that any old random macro could be
transformed into a different use of that very same macro as long as
expansion terminates somewhere. That's fine with me.

Also, it seems that formerly macroexpand-all's operation on comma-fied
forms prior to the backq change was doing what we've said users shouldn't
rely on.
``(,,(pop (car x)) ,(incf result) ,a) is
(SB-IMPL::BACKQ-LIST* (QUOTE SB-IMPL::BACKQ-LIST) (POP (CAR X)) (QUOTE
((INCF RESULT) A)))
But macroexpand-all works because stuffing the expansion of (pop (car x))
into the second arg of backq-list* does something sane.  It's hard to
imagine why that would be untrue, but I think it is nifty to provide
macroexpand-all hooks that allow transforming (some-macro (pop x)) ->
(some-macro (let* ((#:x ....))).
You'd have to write the partial expander for each macro that you wanted to
have this capability for.
quasiquote is but one example where it becomes extremely desirable.
-all-except-quasiquote was too ad-hoc for my liking.

One last thing: I'm pretty sure that quasiquote's expansion needs to use
only the standard list constructors when being called for COMPILE and not
COMPILE-FILE.  (Or rather, any functions would be ok - ANSI or non-ANSI -
that aren't foldable)
COMPILE is not allowed to collapse constants, so we should skip the IR1
transforms for foldable list constructors.
[Unless someone has a different interpretation of the issue about exactly
how constant backquoted lists are, I think it's safer to remove the
interesting behavior from COMPILE. And Paul's original change should have
done that but it would have been impossible]
As such, I've changed the code so that DEFINE-COMPILER-MACRO QUASIQUOTE
expands using the SB-IMPL functions but DEFMACRO QUASIQUOTE expands using
ordinary LIST,APPEND,etc.  This change is orthogonal.

I have no new patch yes. Just waiting for more clarity on which direction
to go.