Compiled function annotations.

[Ben Wing]

On Mon, Feb 8, 2010 at 8:05 AM, Aidan Kehoe <kehoea at parhasard.net> wrote:
>
>  Ar an triú lá de mí Feabhra, scríobh Ben Wing:
>
>  > add more byte-code assertions and better failure output
>  >
>  > -------------------- ChangeLog entries follow: --------------------
>  >
>  > src/ChangeLog addition:
>  >
>  > 2010-02-03  Ben Wing  <ben at xemacs.org>
>
>  >      [...]
>  >      Turn on the "compiled-function annotation hack".  Implement it
>  >      properly by hooking into Fdefalias().  Note in the docstring to
>  >      `defalias' that we do this.
>
> [...]
>
>  > [...]
>  > @@ -716,12 +718,19 @@
>  >  DEFUN ("define-function", Fdefine_function, 2, 2, 0, /*
>  >  Set SYMBOL's function definition to NEWDEF, and return NEWDEF.
>  >  Associates the function with the current load file, if any.
>  > +If NEWDEF is a compiled-function object, stores the function name in
>  > +the `annotated' slot of the compiled-function (retrievable using
>  > +`compiled-function-annotation').
>  >  */
>
> This is noticeably different from what the compiled function annotations
> used to do, and breaks help.el, which says, e.g. that #'eql was loaded from
> `eql'. What’s the upside here? Just that the function name is available in
> the function object? If you have a compiled function object, and an idea of
> what directory it was loaded from, you can grep for the constants vector or
> some sufficiently distinct subset of it in the .elc files, and that’ll tell
> you where it was defined.

I wrote the annotation hack originally and it was always designed to
put the function name in the compiled function object.  This way it
makes backtraces more readable -- the print method for compiled
functions automatically prints the annotation.  I don't know where
that code in help.el came from.  It might have been this: If the
function name couldn't be identified by the very hacky method formerly
used in lread.c, the filename was substituted.  Then compiled code was
changed to use `defalias' instead of `fset', and the filename would
always be substituted -- then someone saw this and wrote the code in
help.el.

I just deleted the code -- load-history does the trick fine.