🚀 go-pugleaf

>>>>> On 7 Sep 90 03:02:27 GMT, jbw@bucsf.bu.edu (Joseph Wells) said:

> In article <9008310740.AA04666@csvax.cs.caltech.edu> daveg@CSVAX.CS.CALTECH.EDU (David Gillespie) writes:

>                                                       ... there is a
>    window between the call to make-local-variable and the first setq in
>    which a setq-default's effect is incorrectly visible.

> 	   (setq foo 'one)
> 	   (make-variable-buffer-local 'foo)
> 	   (make-local-variable 'foo)
> 	   foo  => 'one
> 	   (setq-default foo 'two)
> 	   foo  => 'two    [this is inconsistent]
> 	   (setq foo 'three)
> 	   foo  => 'three
> 	   (setq-default foo 'four)
> 	   foo  => 'three  [correct behavior]

> This is "correct" behavior.  make-variable-buffer-local and
> make-local-variable leave the buffer-local value of the symbol unbound.
> When the buffer-local value of a symbol is unbound, it is shadowed by the
> default value of the symbol.

Actually, make-local-variable *does* copy the default value of the
variable to the buffer-local binding.  The above experiment was
inconsistent with the following, which I just tried with Emacs 18.55:

	(setq bar 'one)
	(make-local-variable 'bar)
	bar  => 'one
	(setq-default bar 'two)
	bar  => 'one

That is, once you have made a variable local to a buffer, it no
longer sees the default...  *unless* the variable was flagged to
become buffer-local automagically but had not yet actually done so.

In node "Buffer Local Variables" my copy of the Lisp manual says,
with no if's, and's, or but's:

    The buffer-local value of SYMBOL starts out as the same value
    SYMBOL previously had.


> Strange fact: if a symbol has any buffer-local values, even unbound
> values, then that symbol's default value can not be unbound.

> Fact 2: you can undo the effects of any make-variable-buffer-local and
> make-local-variable calls with makunbound.

Hey, you're right!

There are explicit tests in both Fmake_variable_buffer_local and
Fmake_local_variable to change an unbound variable to nil.  I
wonder why; it looks like it would work fine leaving the variable
unbound.

The second fact could, as far as I can see, be fixed by this
patch to Fmakunbound in data.c:

----
  CHECK_SYMBOL (sym, 0);
! XSYMBOL (sym)->value = Qunbound;
  return sym;
----
  CHECK_SYMBOL (sym, 0);    /* Actually redundant now. */
! Fset(sym, Qunbound);      /* Looks like a kludge, but unbind_to() */
  return sym;               /*    does this, too! */
----

Gee, I wonder if there should be a "makunbound-default" function?  :-)

The Fboundp and specbind() functions would have to be enhanced to be able
to understand unbound buffer-local variables.  One creative solution would
be to define a "safe-symbol-value" function that doesn't mind returning
Qunbound as the symbol value; where Qunbound would be interned under
some suitably obscure name.  Then Fboundp and specbind() could use
this function, and there have even been a couple of times I could have
used it myself...

However, all of this is beginning to bring to mind the saying,
"if it works [well enough], don't fix it."  :-/

								-- Dave
--
Dave Gillespie
  256-80 Caltech Pasadena CA USA 91125
  daveg@csvax.cs.caltech.edu, ...!cit-vax!daveg