lyx_mirror/00README_STR_METRICS_BRANCH
Jean-Marc Lasgouttes 7734270163 Speed-up drawing when text is not justified.
Do not cut strings at separators in RowPainter when text is not
justified. This speeds-up painting by reducing the number of strings
to draw.

Do also a modest cleanup of paintChar (remove dubious optimization).
2014-07-08 21:24:01 +02:00

104 lines
3.9 KiB
Plaintext

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
PLEASE DO NOT DO WORK ON TOP OF THIS BRANCH.
I INTEND TO REWRITE HISTORY LATER!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
This branch is where I (jmarc) try to implement string-wise metrics
computation. The goal is to have both good metrics computation (and
font with proper kerning and ligatures) and better performance than
what we have with force_paint_single_char. Moreover there has been
some code factorization in TextMetrics, where the same row-breaking
algorithm was basically implemented 3 times.
Currently everything is supposed to work for both LTR and RTL text.
When KEEP_OLD_METRICS_CODE is defined in TextMetrics.cpp, the new code
is tested against the old one in getPosNearX and cursorX. This can be
helpful when looking for discrepancies between the algorithms. Note
that this only makes sense when force_paint_single_char=true, since
this enforces char-by-char metrics computation.
What is done:
* Make TextMetrics methods operate on Row objects: breakRow and
setRowHeight instead of rowBreakPoint and rowHeight.
* Change breakRow operation to operate at strings level to compute
metrics The list of elements is stored in the row object in visual
ordering, not logical. This will eventually allow to get rid of the
Bidi class.
* rename getColumnNearX to getPosNearX (and change code accordingly).
It does not make sense to return a position relative to the start of
row, since nobody needs this.
* Re-implement cursorX and getPosNearX using row elements.
* Implement proper string metrics computation (with cache), when
lyxrc.force_paint_single_char is false. In this case, remove also
useless workarounds which disable kerning and ligatures.
* when lyxrc.force_paint_single_char is false, draw also RtL text
string-wise. This both speeds-up drawing and prepares for code
removal, since we now rely on Qt to do things we use to do by
ourselves (see isArabic* and isHebrew* code in Encodings.cpp).
* Do not cut strings at separators in RowPainter when text is not
justified. This speeds-up painting by reducing the number of strings
to draw.
Next steps:
* get lots of testing.
* Get rid of old code in cursorX and getPosNearX; it has been
kept for comparison purpose, guarded with KEEP_OLD_METRICS_CODE in
order to check computations.
* Profile and see how performance can be improved.
Steps for later (aka out of the scope of this branch):
* Re-implement row painting using row elements. This is not difficult
in principle, but the code is intricate and needs some careful
analysis. The first thing that needs to be done is to break row
elements with the same criteria. Currently breakRow does not
consider on-the-fly spell-checking and selection changes, but it is
not clear to me that it is required. Moreover, this thing would only
work if we are sure that the Row object is up-to-date when drawing
happens. This depends on the update machinery.
This would allow to get rid of the Bidi code.
Known bugs:
* When selecting text, the display seems to move around. This is
because partly selected words are drawn in two parts, and in case
like "ef|fort" or "V|AN", there are some ligature or kerning effects
that change the display. I am not sure yet how to fix that.
* there are other differences in what breaks words. In particular,
RowPainter breaks strings at spell-checking extremities. This seems
to be harmless.
* when clicking in the right margin, getPosNearX does not return
the same value as before. I am not sure whether this is important.
Other differences in behavior (aka bug fixes):
* end of paragraph markers metrics are computed with the font of the
actual text, not default font.
* in RtL paragraphs, the end-of-paragraph marker does not move the row
to the right anymore (ticket #9040).
* When cursor is after a LtR separator just before a RtL chunk, the
cursor position is computed better with the new code.