|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
*/ |
|
|
|
/* |
|
* |
|
* (C) Copyright IBM Corp. 2003 - All Rights Reserved |
|
*/ |
|
|
|
package sun.font; |
|
|
|
|
|
|
|
*/ |
|
public final class FontRunIterator { |
|
CompositeFont font; |
|
char[] text; |
|
int start; |
|
int limit; |
|
|
|
CompositeGlyphMapper mapper; |
|
|
|
int slot = -1; |
|
int pos; |
|
|
|
public void init(CompositeFont font, char[] text, int start, int limit) { |
|
if (font == null || text == null || start < 0 || limit < start || limit > text.length) { |
|
throw new IllegalArgumentException(); |
|
} |
|
|
|
this.font = font; |
|
this.text = text; |
|
this.start = start; |
|
this.limit = limit; |
|
|
|
this.mapper = (CompositeGlyphMapper)font.getMapper(); |
|
this.slot = -1; |
|
this.pos = start; |
|
} |
|
|
|
public PhysicalFont getFont() { |
|
return slot == -1 ? null : font.getSlotFont(slot); |
|
} |
|
|
|
public int getGlyphMask() { |
|
return slot << 24; |
|
} |
|
|
|
public int getPos() { |
|
return pos; |
|
} |
|
|
|
/* |
|
* characters that are in the 'common' script become part of the |
|
* surrounding script run. we want to fetch these from the same font |
|
* used to get surrounding characters, where possible. but we don't |
|
* want to force non-common characters to come from other than their |
|
* standard font. |
|
* |
|
* what we really want to do is this: |
|
* 1) fetch a code point from the text. |
|
* 2) get its 'native' script code |
|
* 3) determine its 'resolved' script code |
|
* 4) if its native script is COMMON, and its resolved script is the same as the previous |
|
* code point's, then see if the previous font supports this code point. if so, use it. |
|
* 5) otherwise resolve the font as usual |
|
* 6) break the run when either the physical font or the resolved script changes. |
|
* |
|
* problems: we optimize latin-1 and cjk text assuming a fixed |
|
* width for each character. since latin-1 digits and punctuation |
|
* are common, following this algorithm they will change to match |
|
* the fonts used for the preceding text, and potentially change metrics. |
|
* |
|
* this also seems to have the potential for changing arbitrary runs of text, e.g. |
|
* any number of digits and spaces can change depending on the preceding (or following!) |
|
* non-COMMON character's font assignment. this is not good. |
|
* |
|
* since the goal is to enable layout to be performed using as few physical fonts as |
|
* possible, and the primary cause of switching fonts is to handle spaces, perhaps |
|
* we should just special-case spaces and assign them from the current font, whatever |
|
* it may be. |
|
* |
|
* One could also argue that the job of the composite font is to assign physical fonts |
|
* to text runs, however it wishes. we don't necessarily have to provide script info |
|
* to let it do this. it can determine based on whatever. so having a special 'next' |
|
* function that takes script (and limit) is redundant. It can fetch the script again |
|
* if need be. |
|
* |
|
* both this and the script iterator are turning char sequences into code point |
|
* sequences. maybe it would be better to feed a single code point into each iterator-- push |
|
* the data instead of pull it? |
|
*/ |
|
|
|
public boolean next(int script, int lim) { |
|
if (pos == lim) { |
|
return false; |
|
} |
|
|
|
int ch = nextCodePoint(lim); |
|
int sl = mapper.charToGlyph(ch) & CompositeGlyphMapper.SLOTMASK; |
|
slot = sl >>> 24; |
|
while ((ch = nextCodePoint(lim)) != DONE && (mapper.charToGlyph(ch) & CompositeGlyphMapper.SLOTMASK) == sl); |
|
pushback(ch); |
|
|
|
return true; |
|
} |
|
|
|
public boolean next() { |
|
return next(Script.COMMON, limit); |
|
} |
|
|
|
static final int SURROGATE_START = 0x10000; |
|
static final int LEAD_START = 0xd800; |
|
static final int LEAD_LIMIT = 0xdc00; |
|
static final int TAIL_START = 0xdc00; |
|
static final int TAIL_LIMIT = 0xe000; |
|
static final int LEAD_SURROGATE_SHIFT = 10; |
|
static final int SURROGATE_OFFSET = SURROGATE_START - (LEAD_START << LEAD_SURROGATE_SHIFT) - TAIL_START; |
|
|
|
static final int DONE = -1; |
|
|
|
final int nextCodePoint() { |
|
return nextCodePoint(limit); |
|
} |
|
|
|
final int nextCodePoint(int lim) { |
|
if (pos >= lim) { |
|
return DONE; |
|
} |
|
int ch = text[pos++]; |
|
if (ch >= LEAD_START && ch < LEAD_LIMIT && pos < lim) { |
|
int nch = text[pos]; |
|
if (nch >= TAIL_START && nch < TAIL_LIMIT) { |
|
++pos; |
|
ch = (ch << LEAD_SURROGATE_SHIFT) + nch + SURROGATE_OFFSET; |
|
} |
|
} |
|
return ch; |
|
} |
|
|
|
final void pushback(int ch) { |
|
if (ch >= 0) { |
|
if (ch >= 0x10000) { |
|
pos -= 2; |
|
} else { |
|
pos -= 1; |
|
} |
|
} |
|
} |
|
} |