several impnotes additions and corrections

[Hoehle, Joerg-C. <Joe...@t-...>]

Hi,

here's what I came across recently:

o impnotes =A75.1.1 "It is an error if a DEFCONSTANT variable is bound =
at the moment the DEFCONSTANT is executed, but DEFCONSTANT does not =
check this."
Misleading, as a warning is issued.

o impnotes =A75.1.6 mentions ext:get-setf-method and
-multiple-value. Perhaps it should generally say that ext: contains
symbols and (working) functionality from CLtL2 removed from ANSI-CL
(and list these, perhaps as chapter 29.3).  E.g. for a few minutes
I was wondering whether get-setf-method was a CLISP-specific addition.
Then I got my CLtLII book.

It could then explain that get-setf-method differs from =
get-setf-expansion in ensuring that a single store form is produces, =
otherwise an error is signaled.

o impnotes =A75.1.8 "Calling SYMBOL-VALUE on a symbol defined as a =
SYMBOL-MACRO
returns the value of the expansion. Calling SET on a symbol defined as
a SYMBOL-MACRO calls SETF on the expansion."
Wrong, there was a change 2005-01-05 to remove that logic.  These
days, SYMBOL-VALUE signals an error, SET accesses the symbol-value cell.
Just remove that paragraph or mention slot access and/or document an
incompatible change? (impnotes doesn't seem to document changes).

Imnotes is wrong again:
"Calling BOUNDP on a symbol defined as a SYMBOL-MACRO returns T."
These days:
CL-USER> (define-symbol-macro weird-2005 *print-case*)
WEIRD-2005
CL-USER> (boundp 'weird-2005)
NIL
CL-USER> weird-2005
:UPCASE
So (boundp foo) presumably still implies (symbol-value foo) succeeds, =
but
impnotes need be changed to reflect past 2005 behaviour.

o impnotes does not mention that and how PUSH, POP and PUSHNEW operate =
on
VALUES places, although the code explicitly deals with multiple stores
(i.e. it's not a side effect of the implementation), while e.g. sbcl and =
cmu
error out.

; push and pop on VALUES places
(multiple-value-list (let (a b) (push (values 1 2) (values a b))))
((1) (2))
(let (a b) (list (multiple-value-list (pushnew (values 1 2) (values a =
b))) a b))
(((1) (2)) (1) (2))

o impnotes =A76.2 "The macros DOLIST and DOTIMES establish a single
  binding for the iteration variable and assign it on each iteration."
However, DOLIST's result form introduces a separate binding:
(dolist (i '(1 2 3) i) (push i a)) ->
(DO* ((#:LIST-3945 '(1 2 3) (CDR #:LIST-3945)) (I NIL))
 ((ENDP #:LIST-3945) (LET ((I NIL)) (DECLARE (IGNORABLE I)) I))
 (DECLARE (LIST #:LIST-3945)) (SETQ I (CAR #:LIST-3945)) (PUSH I A))

I don't remember the discussion on this issue, but I understand that
one doesn't want type conflicts with declarations and the final case
where "At the time result-form is processed, var is bound to nil.".
I.e. the list element type may not encompass NULL, while NULL is the
type needed for the result form.

I think impnotes must document the special treatment in the result form.

o impnotes 30.11.2 LETF
Add(?) something along "A single UNWIND-PROTECT is
used to restore the original places, thus in case of errors therein
[the restore execution], a nested series of (LETF* # (LETF* ...))
behaves differently from a single (LETF* (1+2) ...)."

Regards,
	J=F6rg H=F6hle