Don't depend on obsolete Guile features.

Don't depend on obsolete Guile features.

* Replace scm_listify with scm_list_*. It's been deprecated for a long time, and is gone in Guile 2.1.
* Use scm_protects only in Guile 1.x. It's internal in Guile 2.0 (and obsolete as far as Lilypond is concerned).

This makes Lilypond build with Guile 2.1.

[dak, 2014-06-26 20:24:03]

https://codereview.appspot.com/109230043/diff/1/lily/general-scheme.cc
File lily/general-scheme.cc (right):

https://codereview.appspot.com/109230043/diff/1/lily/general-scheme.cc#newcod...
lily/general-scheme.cc:269: protects are no longer needed in Guile 2.
What does "are no longer needed" mean?  What happens with the gc dumps in
lily.scm?  hashtable->alist is called on its result, and I get
Parsing...
guile> (hash-table->alist #f)
ERROR: In procedure hash-fold:
ERROR: Wrong type argument in position 3: #f
ABORT: (wrong-type-arg)

in the Scheme sandbox.  So this does not appear to do the right thing.

https://codereview.appspot.com/109230043/diff/1/lily/include/lily-guile.hh
File lily/include/lily-guile.hh (right):

https://codereview.appspot.com/109230043/diff/1/lily/include/lily-guile.hh#ne...
lily/include/lily-guile.hh:124: return scm_append (scm_list_2 (x1, x2));
Good changes...

[Devon Schudy, 2014-06-27 02:30:13]

dak@gnu.org wrote:
>
https://codereview.appspot.com/109230043/diff/1/lily/general-scheme.cc#newcod...
> lily/general-scheme.cc:269: protects are no longer needed in Guile 2.
> What does "are no longer needed" mean?

I thought Guile 2's GC would look for pointers in the heap, so
protects wouldn't be necessary — but after checking the manual it
seems it only does that for memory allocated with scm_gc_malloc. So
they're still needed after all. (Although the stack is scanned
automatically, which may make some protects unnecessary.)

> What happens with the gc dumps in lily.scm?

AFAICT there's no way to get the set of protected objects in Guile 2,
so there's no way to make this work. Should ly:protects give a "not
supported in Guile 2" error instead of returning #f? (And maybe
dump-gc-protects should be disabled under Guile 2, since it's
useless.)

[dak, 2014-06-27 06:50:13]

On 2014/06/27 02:30:13, Devon Schudy wrote:
> mailto:dak@gnu.org wrote:
> >
>
https://codereview.appspot.com/109230043/diff/1/lily/general-scheme.cc#newcod...
> > lily/general-scheme.cc:269: protects are no longer needed in Guile 2.
> > What does "are no longer needed" mean?
> 
> I thought Guile 2's GC would look for pointers in the heap, so
> protects wouldn't be necessary — but after checking the manual it
> seems it only does that for memory allocated with scm_gc_malloc. So
> they're still needed after all. (Although the stack is scanned
> automatically, which may make some protects unnecessary.)

The stack is scanned automatically in GUILEv1 already.  That's extensively used.

> > What happens with the gc dumps in lily.scm?
> 
> AFAICT there's no way to get the set of protected objects in Guile 2,
> so there's no way to make this work. Should ly:protects give a "not
> supported in Guile 2" error instead of returning #f? (And maybe
> dump-gc-protects should be disabled under Guile 2, since it's
> useless.)

No idea, have not looked closed yet.  To be honest, I am currently more worried
about getting LilyPond to run with Guile 2.0 rather than Guile 2.1.  For Guile
2.1, it might make sense to enquire on the Guile developer list what one is
supposed to use instead.

[Devon Schudy, 2014-07-03 15:15:15]

dak@gnu.org wrote:
> The stack is scanned automatically in GUILEv1 already.  That's
> extensively used.

Huh, so it is. I thought the conservative GC was new in Guile 2, but
apparently the change was not that dramatic.

> For Guile 2.1, it might make sense to enquire on the Guile
> developer list what one is supposed to use instead.

I asked, and Ludovic Courtès suggested simply implementing our own
version of protects. In Guile 2, there's no longer anything special
about the built-in protects; it's just a convenience, and it can be
recreated in less than a page.

Are the memory-leak debugging tools important enough to bother?

[Devon Schudy, 2014-07-03 15:48:15]

Fix incorrect comment. Support scm_protects in Guile 2.0. Give an error instead
of #f when it's not available.