bootstrap error in c-mode with CANNOT_DUMP

bootstrap error in c-mode with CANNOT_DUMP

[Ken Raeburn]

I'm doing some experimenting with some changes that require linking in  
a new library and calling some code early in the Emacs process.  Since  
the library isn't currently unexec-friendly, I'm trying some builds  
without using unexec.  I find that I can't do a bootstrap.  If I do  
"configure ... CPPFLAGS=-DCANNOT_DUMP" with current (unhacked) CVS  
sources and try "make bootstrap", I eventually get this error:

Wrote /Users/raeburn/dev/e/fsf/lisp/progmodes/cfengine.elc
Compiling /Users/raeburn/dev/e/fsf/lisp/progmodes/cmacexp.el
Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
23.0.94/site-lisp' does not exist.
Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
site-lisp' does not exist.
Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
23.0.94/lisp' does not exist.
Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
23.0.94/leim' does not exist.
Loading loadup.el (source)...
Using load-path (/Users/raeburn/dev/e/fsf/lisp)
Loading emacs-lisp/byte-run...
[....]
Loading tooltip...
((66226 . 4511) (13575 . 0) (649 . 67) 96380 876313 (122 . 7) (22 . 8)  
(6632 . 5779))
Finding pointers to doc strings...
Finding pointers to doc strings...done
Loading cc-langs...
Eval error in the `c-lang-defconst' for `c-cpp-matchers' in c-mode:
Eval error in the `c-lang-defconst' for `c-matchers-1' in c-mode:

In toplevel form:
../../lisp/progmodes/cmacexp.el:97:1:Error: `c-opt-cpp-prefix' got no  
(prior) value in c-mode (might be a cyclic reference)
make[2]: *** [/Users/raeburn/dev/e/fsf/lisp/progmodes/cmacexp.elc]  
Error 1

Without CANNOT_DUMP, it bootstraps just fine; I'm not sure why there  
should be a difference.

Ken

Re: bootstrap error in c-mode with CANNOT_DUMP

[Stefan Monnier]

> I'm doing some experimenting with some changes that require linking in a new
> library and calling some code early in the Emacs process.  Since  the
> library isn't currently unexec-friendly, I'm trying some builds  without
> using unexec.  I find that I can't do a bootstrap.

This is a known problem since it plagued the GNUStep port as well (when
it couldn't dump).  Right now the use of CANNOT_DUMP leads to all kinds
of problems (IIRC even things like ~/.emacs don't get processed
correctly).  But it's not very high on the todo-list, so if you can find
a fix for it, it would probably be welcome, but otherwise don't hold
your breath.  A better path might be to try and find a way to do the
unexec.


        Stefan

Re: bootstrap error in c-mode with CANNOT_DUMP

[Eli Zaretskii]

> From: Ken Raeburn <raeburn@raeburn.org>
> Date: Fri, 12 Jun 2009 02:25:39 -0400
> 
> Compiling /Users/raeburn/dev/e/fsf/lisp/progmodes/cmacexp.el
> Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
> 23.0.94/site-lisp' does not exist.
> Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
> site-lisp' does not exist.
> Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
> 23.0.94/lisp' does not exist.
> Warning: Lisp directory `/Users/raeburn/dev/e/fsf/bld/Inst/share/emacs/ 
> 23.0.94/leim' does not exist.
> Loading loadup.el (source)...
> Using load-path (/Users/raeburn/dev/e/fsf/lisp)
> Loading emacs-lisp/byte-run...
> [....]
> Loading tooltip...
> ((66226 . 4511) (13575 . 0) (649 . 67) 96380 876313 (122 . 7) (22 . 8)  
> (6632 . 5779))
> Finding pointers to doc strings...
> Finding pointers to doc strings...done
> Loading cc-langs...
> Eval error in the `c-lang-defconst' for `c-cpp-matchers' in c-mode:
> Eval error in the `c-lang-defconst' for `c-matchers-1' in c-mode:
> 
> In toplevel form:
> ../../lisp/progmodes/cmacexp.el:97:1:Error: `c-opt-cpp-prefix' got no  
> (prior) value in c-mode (might be a cyclic reference)
> make[2]: *** [/Users/raeburn/dev/e/fsf/lisp/progmodes/cmacexp.elc]  
> Error 1

It looks for Lisp files in the wrong directory, right?  If so, one way
solving that is to explicitly set loadpath as part of the command run
by Make that errors out.

Re: bootstrap error in c-mode with CANNOT_DUMP

[Ken Raeburn]

On Jun 12, 2009, at 17:37, Stefan Monnier wrote:
> This is a known problem since it plagued the GNUStep port as well  
> (when
> it couldn't dump).  Right now the use of CANNOT_DUMP leads to all  
> kinds
> of problems (IIRC even things like ~/.emacs don't get processed
> correctly).  But it's not very high on the todo-list, so if you can  
> find
> a fix for it, it would probably be welcome, but otherwise don't hold
> your breath.  A better path might be to try and find a way to do the
> unexec.


Working on it, but Guile's initialization is a little complicated in  
places.  In the (much) longer term, I think finding some solution for  
Emacs other than unexec would be better, if we want to permit  
extensions in general, because unexec-ability seems like a pretty  
onerous demand to make of all libraries one might want to use.   
(Luckily, I suspect most such libraries don't need to have *any* code  
run before the unexec step, so loading fresh versions in the new  
process isn't a problem; that just doesn't work in my case.)  But  
that's a whole other matter...

I was kind of hoping CANNOT_DUMP might give me a shortcut to something  
I could actually run. :-(
Actually, it kind of does, if I avoid bootstrapping; I just have to  
figure out if each bug is in my work or in the Emacs code base I'm  
starting with....

Ken

Re: bootstrap error in c-mode with CANNOT_DUMP

[Ken Raeburn]

On Jun 12, 2009, at 22:41, Eli Zaretskii wrote:
> It looks for Lisp files in the wrong directory, right?  If so, one way
> solving that is to explicitly set loadpath as part of the command run
> by Make that errors out.

There are other Emacs invocations that fail that way (e.g., list-load- 
path-shadows), but the compilation commands do set EMACSLOADPATH;  
Emacs just warns about the install directories missing anyways.  And  
it does manage to load the lisp files; it doesn't fail until it's  
actually compiling stuff.

Ken

Re: bootstrap error in c-mode with CANNOT_DUMP

[Eli Zaretskii]

> From: Stefan Monnier <monnier@iro.umontreal.ca>
> Date: Fri, 12 Jun 2009 17:37:46 -0400
> Cc: Emacs Development <emacs-devel@gnu.org>
> 
> Right now the use of CANNOT_DUMP leads to all kinds of problems
> (IIRC even things like ~/.emacs don't get processed correctly).

Perhaps we should mention in etc/PROBLEMS that CANNOT_DUMP is broken.

Re: bootstrap error in c-mode with CANNOT_DUMP

[Richard Stallman]

If CANNOT_DUMP does not work, does that mean Emacs is broken on some
platforms?  If so, it seems to me that fixing this is urgent.  A
regression that entirely breaks certain platforms is as bad as it
gets.

Re: bootstrap error in c-mode with CANNOT_DUMP

[Ken Raeburn]

On Jun 14, 2009, at 00:56, Richard Stallman wrote:
> If CANNOT_DUMP does not work, does that mean Emacs is broken on some
> platforms?  If so, it seems to me that fixing this is urgent.  A
> regression that entirely breaks certain platforms is as bad as it
> gets.

There are no platforms currently using that option.  But it might mean  
porting to new platforms or other environments (e.g., using libguile)  
has a higher initial cost than necessary, if it requires making unexec  
work.

Ken

Re: bootstrap error in c-mode with CANNOT_DUMP

[Richard Stallman]

    There are no platforms currently using that option.

That's a good thing, overall.

      But it might mean  
    porting to new platforms or other environments (e.g., using libguile)  
    has a higher initial cost than necessary, if it requires making unexec  
    work.

True, but new platforms are rare these days, so I suppose it is not
worth fixing CANNOT_DUMP in anticipation.