Issue 2507: Stop the entanglement of context properties and grob property internals

Issue 2507: Stop the entanglement of context properties and grob property internals

This introduces a semi-opaque structure Grob_properties meeting the
predicate ly:grob-properties? that is algorithmically handled at the C++
level via a wrapper structure Grob_property_info.

Encapsulating grob properties in that manner reduces the potential for
clashes and makes it easier to change algorithms and/or internal
representation.

While the principal distinction between context properties (one value
per context) and context-based grob property templates (one stack per
context) remains, at least the separation of the handling is more
pronounced.

[dak, 2014-08-14 20:29:24]

Fix the internals-meddling scheme-text-spanner regtest

[dak, 2014-08-14 23:41:45]

Fix incredibly reckless context/grob property mixup in event-listener

[lemzwerg, 2014-08-15 06:11:31]

> Fix incredibly reckless context/grob property mixup in event-listener

I would rather call this clueless :-)

[dak, 2014-08-15 07:44:08]

On 2014/08/15 06:11:31, lemzwerg wrote:
> > Fix incredibly reckless context/grob property mixup in event-listener
> 
> I would rather call this clueless :-)

Ill-advised would be an understatement since this is not just poking into
internal data structures in a way that happens to work by chance, but it is
doing so in a way that tends to deliver stale data for any properties set
outside of the bottom context.

At any rate, discovering this problem is actually a good argument for this patch
in the first place.

[dak, 2014-08-15 07:58:26]

At any rate, putting Graham in the Cc since event-listener.ly or equivalent code
is purportedly used in Vivi.  The change in event-listener.ly is independent of
this issue itself and should be applied to any similar code in order to avoid
getting flaky results.  ly:context-grob-definition has been available in working
state since 2004:

commit 5762e8bb079db16dd31166fa0c4b2a85464b0b8b
Author: Nicolas Sceaux <nicolas.sceaux@free.fr>
Date:   Sun Aug 29 13:58:05 2004 +0000

    (LY_DEFINE): ly:context-grob-definition takes two arguments.

[dak, 2014-08-15 09:42:15]

When Grob_property_info::pop removes a Grob_properties, drop the reference

[Graham Percival, 2014-08-15 15:07:07]

On Fri, Aug 15, 2014 at 07:58:26AM +0000, dak@gnu.org wrote:
> At any rate, putting Graham in the Cc since event-listener.ly or
> equivalent code is purportedly used in Vivi.  The change in
> event-listener.ly is independent of this issue itself and should be
> applied to any similar code in order to avoid getting flaky results.

Thanks for the heads-up!  Vivi turned out to be a dead end (the
machine learning technique I used was really inappropriate for the
problem), but I'm still doing things with lilypond output.

Cheers,
- Graham

[Keith, 2014-08-18 05:51:22]

Big change, looks good, just one question.

https://codereview.appspot.com/126280043/diff/40001/lily/context-property.cc
File lily/context-property.cc (right):

https://codereview.appspot.com/126280043/diff/40001/lily/context-property.cc#...
lily/context-property.cc:152: Don't mess with MIDI.
I am confused by this comment.  MIDI output does not create any grobs, so I
figured the code would return false after finding no definition of symbol_, same
as if I type \override Bogus.color =#red
What is the special case for MIDI ?

[dak, 2014-08-18 08:30:22]

On 2014/08/18 05:51:22, Keith wrote:
> Big change, looks good, just one question.
> 
> https://codereview.appspot.com/126280043/diff/40001/lily/context-property.cc
> File lily/context-property.cc (right):
> 
>
https://codereview.appspot.com/126280043/diff/40001/lily/context-property.cc#...
> lily/context-property.cc:152: Don't mess with MIDI.
> I am confused by this comment.  MIDI output does not create any grobs, so I
> figured the code would return false after finding no definition of symbol_,
same
> as if I type \override Bogus.color =#red
> What is the special case for MIDI ?

The special case for MIDI is that the \override is actually accepted in the
input (rather than an error getting flagged).  Consequently the code is able to
arrive here in the normal course of events.  I don't think that \override
Bogus.color = #red even makes it into the music.  You probably have to revert to
Scheme-created music in order to smuggle it in.

This comment is actually taken unchanged from the original code.