bug#11649: 24.1.50; Some compilation woes

[Eli Zaretskii <eliz@gnu.org>]

> From: Juanma Barranquero <lekktu@gmail.com>
> Date: Fri, 8 Jun 2012 15:19:27 +0200
> 
> 1) Bootstrapping now generates a few "lisp nesting exceeds
> max-lisp-eval-depth" errors:
> 
> In toplevel form:
> progmodes/make-mode.el:324:1:Error: Lisp nesting exceeds `max-lisp-eval-depth'
> 
> In toplevel form:
> textmodes/rst.el:271:1:Error: Lisp nesting exceeds `max-lisp-eval-depth'
> 
> In toplevel form:
> cedet/semantic/bovine/c-by.el:183:1:Error: Lisp nesting exceeds
> `max-lisp-eval-depth'

I fixed this in revision 108536, by enlarging the stack depth of the
byte compiler when it is run interpreted.  The new size is 2200; the
largest value I needed to use to compile the above files was 2000.

AFAICS, this problem doesn't show on Posix hosts because there the
byte compiler is run interpreted only for a small set of Lisp files
that are preloaded into Emacs.  After these are compiled, the
bootstrap process produces a new Emacs executable with those files
preloaded, then compiles the rest of Lisp files with this new
executable, where the byte compiler no longer runs interpreted.

By contrast, on MS-Windows, all of the Lisp files are compiled with
the bootstrapped Emacs executable, so they all are compiled with the
interpreted byte compiler.

Volunteers are welcome to change the Windows bootstrap procedure to
include this optimization.  In the meantime, a larger stack will have
to be the solution.

> 2) Byte-compilation generates lots of new "function X might not be
> defined at runtime". AFAICS, they are generated for functions used
> before they are defined (with defun).

I don't see this with the latest trunk.  If you do, can you show a
complete recipe with the messages it produces?

> 3) Byte-compilation seems quite fragile. After a successful bootstrap,
> touch'ing every *.el in lisp/** and doing
> 
>   cd lisp
>   make recompile
> 
> produces quite a few errors:
> 
>   Generating autoloads for ibuf-macs.el...
>   Autoload cookie error in ibuf-macs.el:129 (wrong-type-argument integerp nil)
>   Autoload cookie error in ibuf-macs.el:160 (wrong-type-argument integerp nil)
>   Autoload cookie error in ibuf-macs.el:264 (wrong-type-argument integerp nil)
>   Autoload cookie error in ibuf-macs.el:299 (wrong-type-argument integerp nil)
>   Generating autoloads for ibuf-macs.el...done

Do you still see this one?

>   In toplevel form:
>   textmodes/rst.el:120:1:Error: Invalid byte code in
> c:/emacs/trunk/lisp/emacs-lisp/cl-macs.elc
>   Checking c:/emacs/trunk/lisp/url...
>   Checking c:/emacs/trunk/lisp/vc...
>   Compiling c:emacs/repo/lisp/vc/pcvs.el...
>   Compiler-macro error: (error "Invalid byte code in
> c:/emacs/trunk/lisp/emacs-lisp/cl-macs.elc")
>   Compiler-macro error: (error "Invalid byte code in
> c:/emacs/trunk/lisp/emacs-lisp/cl-macs.elc")
>   etc

I confirm these.  They also happen on GNU/Linux.  My interpretation of
this is that some of the files compiled by bootstrap-emacs produce
invalid .elc files (e.g., cl-macs.elc, cl-lib.elc, and pvcs-util.elc),
because once they are compiled by the non-bootstrap Emacs executable,
the errors go away.  Stefan, can you please look into this?