Parser: discriminate quoted and non-quoted strings

Parser: discriminate quoted and non-quoted strings

Checking quoted strings for syntactic correctness seems spurious, as does
splitting them into multiple symbols.  Now "x.y" is treated as a list
of two symbols without the quotation marks, but as a single string or
symbol when given quotation marks.

This allows using quotation marks for creating arbitrary symbols, and
consequently specifying key lists using quoted strings without the
strings being split in other components or being rejected because
of syntactical reasons.


Also contains commit:

Reorganize Lily_lexer::scan_bare_word

This also redefines LilyPond's manners of converting simple
expressions into music, most notably checking drum types for being
defined before accepting them as note values.

[lemzwerg, 2017-04-02 16:47:52]

Honestly, I don't understand the description of your change.   Can you give a
second example?  And is there something to add to the documentation?

[Dan Eble, 2017-04-02 16:57:12]

FWIW, the description sounds like a good idea to me.

[dak, 2017-04-02 17:21:39]

On 2017/04/02 16:47:52, lemzwerg wrote:
> Honestly, I don't understand the description of your change.   Can you give a
> second example?  And is there something to add to the documentation?

Words are split into lists at . and , : this is what makes

\lyrics { \override Lyrics.LyricText = ... }

work in the first place since in lyricmode Lyrics.LyricText is a single word. 
But when I write "Lyrics.LyricText" with quotemarks around, splitting it seems
like a bad idea in most contexts.

If I have a function argument of type key-list? and I give it

    1,"soprano2","alto.3"

it seems like complaining about bad symbol syntax for soprano2 and/or splitting
"alto" and "3" into two separate arguments and/or complaining about them will
just not be expected behavior.

The concrete case that I want this behavior for looks like

\version "2.19.59"

\header {
  texidoc = "The @code{\\voicify} command can be used for continuing voices
and changing the order of @code{\\voiceOne}@dots{}@code{\\voiceFour} style
overrides."
}

\layout { ragged-right = ##t }

{
  \time 2/4
  \new Voice = "sop" {
    a'2~ |
    \voicify 1,"sop" << e''2 \\ { \voiceTwo a'2~ \oneVoice } >>
    a'2
  }
  \voicify 1,3,4,2 << a''2 \\ e''2 \\ c''2 \\ a'2 >>
}

Now I don't really want to have to distinguish between using strings/symbols for
key lists on the LilyPond side of things, and obviously something like
"soprano.2" or "tenors, at least purportedly" should end up being one rather
than two list entries.

I am not overly enthused that everything will be turned into symbols in this
application, but as long as strings are being used for identification, interning
them into symbols might actually make sense in the long haul internally even
though currently voice ids are still strings.

\voicify will have a separate review, but most of the tradeoffs taken are
actually integrated in this issue here.  The keypoint is that I don't want
material in quote marks to be further syntax-analyzed or split, even if that
material is intended to end up as a symbol.  And for that the lexer has to tell
the parser whether or not quote marks were involved in its forming of a
word/string.  And I don't want to have the user worry too much about the
string/symbol distinction while sticking with LilyPond syntax (and even a bit of
Scheme).