Make \relative { ... } interpret the first pitch as an absolute one

Make \relative { ... } interpret the first pitch as an absolute one

The previous default of \relative { ... } being \relative c' { ... }
made c' more special than warranted.  Having to write
\relative f { ... } for this functionality seems awkward since f
serves no musical function but just happens to be the middle of the
scale.  Iff the scale is the standard scale.  The new definition of
\relative takes care to determine the "middle of the scale" reference
point properly even in the case of non-standard scales.

The conversion rules catch the cases where
a) \relative was used without a pitch argument.  In this case the
pitch gets turned into the language-neutral #(ly:make-pitch 0 0).
b) \relative x''' { x ... is being used, with ''' being an arbitrary
octave and x an arbitrary notename.

This is an example of a conversion rule that should not be applied
more than once...


In a separate commit:
Run scripts/auxiliar/update-with-convert-ly.sh

[dak, 2013-03-07 00:59:07]

Catch the ugly ducklings, too.

[dak, 2013-03-07 01:27:27]

Use c' rather than $(ly:make-pitch 0 0) when no \include or \language threatens
a language change

[Keith, 2013-03-07 05:58:46]

In order from reasonable to ridiculous,
Converting 
 \relative{g b d}  -> \relative c' {g b d}
is reasonable,
 \relative{sol ti re}  -> \relative $(ly:make-pitch 0 0) {sold ti re}
mildly annoying.
Changing the default starting pitch of \relative down a fifth will annoy maybe
20 people who use \relative by habit, make maybe 5 people happy. 
Converting
 <<\relative c'{f a c} // \relative c'{c a f}>>
to <<\relative c'{f a c} // \relative {c' a f}>>
creates a mess.

[dak, 2013-03-07 06:35:00]

On 2013/03/07 05:58:46, Keith wrote:
> In order from reasonable to ridiculous,

Actually, you'll find me in almost full agreement.

> Converting 
>  \relative{g b d}  -> \relative c' {g b d}
> is reasonable,

Yes and no.  If the meaning of \relative { ... } remains unchanged,
this conversion is gratuitous.  It should not be in convert-ly, but we
should still apply it to the LilyPond tree to get our own
documentation consistent.

>  \relative{sol ti re}  -> \relative $(ly:make-pitch 0 0) {sold ti re}
> mildly annoying.

If the meaning of \relative { ... } remains unchanged, this conversion
is both gratuitous and ugly.  If the meaning changes, this conversion
is necessary and ugly.  Finding the right name for c'/do'... can
probably be done right with more complex heuristics in more cases.  Or
we just stipulate that every notename language knows c.

> Changing the default starting pitch of \relative down a fifth will
> annoy maybe 20 people who use \relative by habit, make maybe 5
> people happy.

Existing users: possibly.

> Converting
>  <<\relative c'{f a c} // \relative c'{c a f}>>
> to <<\relative c'{f a c} // \relative {c' a f}>>
> creates a mess.

It's a good start :)

If \relative c' { c } gets converted to \relative { c' } while
\relative c' { f } stays like it is: this is certainly a mess.  This
conversion covers about 40% of \relative uses in our
code/documentation base.  Being aware of the whole notename system
rather than just c would cover more and get more consistency.  I'll
prepare such a patch in a few hours and see how much coverage it
reaches.

The whole point of the exercise is to be able to give a satisfactorily
preferable choice to the question "which arbitrary pitch should we be
using for \relative by default".  And changing \relative behavior just
for the mentioned 20/5 people would be somewhat pointless.

Whether we go through with it is a different question.  Not everything
that would have been a good idea to start with is worth a large
change.  But it's nice getting a perspective of the possibility.

[janek, 2013-03-07 07:19:01]

Hi,

i like the idea of \relative interpreting the first pitch "in absolute" if
there's no explicit reference pitch.
However, i don't think we should promote this way of doing things: it's a
shorthand, and with LilyPond it's usually better to be explicit.
So, +1 for changing the behaviour of \relative, but -1 for using this everywhere
in docs.

best,
Janek

[dak, 2013-03-07 09:12:32]

On 2013/03/07 07:19:01, janek wrote:
> Hi,
> 
> i like the idea of \relative interpreting the first pitch "in absolute" if
> there's no explicit reference pitch.
> However, i don't think we should promote this way of doing things: it's a
> shorthand, and with LilyPond it's usually better to be explicit.

I disagree with the notion of a "shorthand" since it is not
_conceptually_ related to \relative f.  f is just the middle of the
standard scale by accident, and I made sure that the code works as
claimed even for non-standard scales.

We had a lot of discussions (and changes) about what pitch to use for
\relative if the first pitch in the music is going to be x'''.

And if the question is what actually _arbitrary_ base to pick in order
to arrive at x''' for the first real note, it is clear that we are
actually trying to find a solution for a self-inflicted problem.

If we want the first pitch to be x''', then the first pitch should be
x''' (and not c' and not f and not x) and not be in need of repeating
as x or x'' or x''' again.

It is _shorter_, but not a shorthand since it does not actually
replace any particular use of \relative.

> So, +1 for changing the behaviour of \relative, but -1 for using
> this everywhere in docs.

Shrug.  We have a "policy" about what \relative to use, this policy
changed over the time and it has not been applied consistently,
either.

I agree with Keith that using this for 40% of all cases in the docs
does not make an overwhelming amount of sense.

I am not convinced about the 100% case being bad, though.

[dak, 2013-03-07 09:25:11]

On 2013/03/07 07:19:01, janek wrote:
> Hi,
> 
> i like the idea of \relative interpreting the first pitch "in absolute" if
> there's no explicit reference pitch.

> However, i don't think we should promote this way of doing things: it's a
> shorthand, and with LilyPond it's usually better to be explicit.

It is no shorthand.  When discussing "policy" for \relative, the
standard question we are fuzzing about is "what pitch to pick for
\relative when the first pitch should be x'''".  That requires making
an arbitrary choice, and we had different policies at different times,
and none are used consistently.

So the proposal is not pick any pitch at all for \relative when the
first pitch should be x''' but rather just write x''' period.

That's not a "shorthand", much like \transpose c' c' { g } is not a
"shorthand" for { g }.

> So, +1 for changing the behaviour of \relative, but -1 for using
> this everywhere in docs.

I don't think using it everywhere would be bad.  Using it for 40% of
all cases would be bad: I agree with Keith on that.

[dak, 2013-03-07 09:27:10]

On 2013/03/07 09:25:11, dak wrote:

Sorry for the repetition.  My browser very much made it appear like it had lost
the previous comment, so I rewrote it from scratch.  I see no way to delete
comments from Rietveld (as opposed to Codeview).

[dak, 2013-03-07 12:44:22]

"Postpone \\include rule for our own code base, make convert-ly much cleverer
about pitches"

[dak, 2013-03-07 13:46:39]

Don't convert \relative c' c (Ugh)

[dak, 2013-03-07 15:02:43]

Typo in convert-ly rule tracked octave jumps only for \relative c...

[dak, 2013-03-07 15:27:43]

Include files with \language in conversion: our code base suffers that quite
well.

[Trevor Daniels, 2013-03-07 15:56:52]

Looking good.  Obviously the text which describes
\relative will need attention too, but this is now
looking like a change worth making.

Trevor

[dak, 2013-03-07 17:01:11]

Be more thorough about skipping through prefatory matter before the first pitch

[dak, 2013-03-07 17:31:08]

On 2013/03/07 15:56:52, Trevor Daniels wrote:
> Looking good.  Obviously the text which describes
> \relative will need attention too,

Not just that.  So far all humanly written changes are constrained to
python/convertrules.ly and ly/music-functions-init.ly (the actual
change of the definition).  All the rest is done by convert-ly so far.

If my grep-foo is correct, there are still more than 1500 occurences
of \relative followed by a pitch.  Now this is spread across
translations, but it is still clear that we are not talking about a
complete job.  But then the \relative conventions used previously had
been far from consistent, too.

One can probably still improve the convert-ly rule and reduce quite a
bit more.

> but this is now looking like a change worth making.

Well, let's say that it is now looking like something quite more than
a hoax.  I've put about a day of work into it now, and I think the
current state is at least serious enough for getting more opinions
from both developers and users.

Unfortunately, the patch is now large enough that git cl uploads it in
parts to Rietveld and Patchy does not get along with that.

And obviously, this needs to get a documentation testing as well as
the regtests.

[pkx166h, 2013-03-07 18:10:46]

On 2013/03/07 17:31:08, dak wrote:
> On 2013/03/07 15:56:52, Trevor Daniels wrote:
> > Looking good.  Obviously the text which describes
> > \relative will need attention too,
> 
> Not just that.  So far all humanly written changes are constrained to
> python/convertrules.ly and ly/music-functions-init.ly (the actual
> change of the definition).  All the rest is done by convert-ly so far.
> 
> If my grep-foo is correct, there are still more than 1500 occurences
> of \relative followed by a pitch.  Now this is spread across
> translations, but it is still clear that we are not talking about a
> complete job.  But then the \relative conventions used previously had
> been far from consistent, too.
> 
> One can probably still improve the convert-ly rule and reduce quite a
> bit more.
> 
> > but this is now looking like a change worth making.
> 
> Well, let's say that it is now looking like something quite more than
> a hoax.  I've put about a day of work into it now, and I think the
> current state is at least serious enough for getting more opinions
> from both developers and users.
> 
> Unfortunately, the patch is now large enough that git cl uploads it in
> parts to Rietveld and Patchy does not get along with that.
> 
> And obviously, this needs to get a documentation testing as well as
> the regtests.

Well if you want to give me a git formatted patch I'd be happy to test it
manually.

James

[dak, 2013-03-07 19:07:08]

On 2013/03/07 18:10:46, J_lowe wrote:
> On 2013/03/07 17:31:08, dak wrote:
> > On 2013/03/07 15:56:52, Trevor Daniels wrote:
> > > Looking good.  Obviously the text which describes
> > > \relative will need attention too,
> > 
> > Not just that.  So far all humanly written changes are constrained to
> > python/convertrules.ly and ly/music-functions-init.ly (the actual
> > change of the definition).  All the rest is done by convert-ly so far.
> > 
> > If my grep-foo is correct, there are still more than 1500 occurences
> > of \relative followed by a pitch.  Now this is spread across
> > translations, but it is still clear that we are not talking about a
> > complete job.  But then the \relative conventions used previously had
> > been far from consistent, too.
> > 
> > One can probably still improve the convert-ly rule and reduce quite a
> > bit more.
> > 
> > > but this is now looking like a change worth making.
> > 
> > Well, let's say that it is now looking like something quite more than
> > a hoax.  I've put about a day of work into it now, and I think the
> > current state is at least serious enough for getting more opinions
> > from both developers and users.
> > 
> > Unfortunately, the patch is now large enough that git cl uploads it in
> > parts to Rietveld and Patchy does not get along with that.
> > 
> > And obviously, this needs to get a documentation testing as well as
> > the regtests.
> 
> Well if you want to give me a git formatted patch I'd be happy to test it
> manually.
> 
> James

On the way.