Re: Sketch for broken beams with consistent slopes (issue 4961041)

[mtsolo]

Thanks for the comments!
Just a technical note - this issue was closed after the patch was pushed
to current master, but I don't mind opening it again to continue a
discussion about what this patch (if Colin has no objections).

The code I'm dealing with in beam-quanting and beam is difficult to
decipher to begin with: it took me at least a year to get my head fully
around what it's doing and to figure out where the gaps were.  I know
there is a lot of reorganization, but I wouldn't do it unless I felt
that it was necessary to make the code perform more consistently (i.e.
to standardize the measuring of X extent).

Cheers,
MS


http://codereview.appspot.com/4961041/diff/60001/lily/beam-quanting.cc
File lily/beam-quanting.cc (right):

http://codereview.appspot.com/4961041/diff/60001/lily/beam-quanting.cc#newcode202
lily/beam-quanting.cc:202: for (vsize i = 0; i < beams.size (); i++)
On 2011/10/26 18:57:38, Keith wrote:
> When there is more than one beam in 'beams' what do you intend the
members
> Beam_scoring_problem to represent ?  The beam as if it were on one
line, but
> allowing the extra horizontal span due to the line-break ?  Should the
members
> like segments_ and is_knee_ include both halves of the beam ?

All of Beam_scoring_problem happens in an artificial space as if there
was no line break (which is why this function is so long - you'll see
x_span_ and x_left being used quite a bit to bring elements to their
positions on this imaginary line).  So, Beam_scoring_problem does not
know what line breaking is, and it is only at the very end that the
normalized endpoints are used to take care of the breaks.

One of the reasons for my most recent patch is that the normalized
endpoints of the beam don't sync up with its actual extent, which is
only reliably derived from the segments.

http://codereview.appspot.com/4961041/diff/60001/lily/beam-quanting.cc#newcode947
lily/beam-quanting.cc:947: final_positions[RIGHT] -= (1 -
normalized_endpoints[RIGHT]) * y_length;
On 2011/10/26 18:57:38, Keith wrote:
> At what horizontal locations are these final_positions? First and last
stems or
> the end of the beam including overhang or x_span_ ?

Including overhang.  This is improved in the most recent patch set.

http://codereview.appspot.com/4961041/