/*
* This file is part of OpenTTD.
* OpenTTD is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
* OpenTTD is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
* See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with OpenTTD. If not, see .
*/
/** @file gfx_layout_icu.cpp Handling of laying out with ICU / Harfbuzz. */
#include "stdafx.h"
#include "gfx_layout_icu.h"
#include "debug.h"
#include "strings_func.h"
#include "language.h"
#include "table/control_codes.h"
#include "zoom_func.h"
#include "3rdparty/icu/scriptrun.h"
#include
#include
#include
#include
#include "safeguards.h"
/** harfbuzz doesn't use floats, so we need a value to scale position with to get sub-pixel precision. */
constexpr float FONT_SCALE = 64.0;
/**
* Helper class to store the information of all the runs of a paragraph in.
*
* During itemization, more and more information is filled in.
*/
class ICURun {
public:
int start; ///< Start of the run in the buffer.
int length; ///< Length of the run in the buffer.
UBiDiLevel level; ///< Embedding level of the run.
UScriptCode script; ///< Script of the run.
Font *font; ///< Font of the run.
std::vector glyphs; ///< The glyphs of the run. Valid after Shape() is called.
std::vector advance; ///< The advance (width) of the glyphs. Valid after Shape() is called.
std::vector glyph_to_char; ///< The mapping from glyphs to characters. Valid after Shape() is called.
std::vector positions; ///< The positions of the glyphs. Valid after Shape() is called.
int total_advance = 0; ///< The total advance of the run. Valid after Shape() is called.
ICURun(int start, int length, UBiDiLevel level, UScriptCode script = USCRIPT_UNKNOWN, Font *font = nullptr) : start(start), length(length), level(level), script(script), font(font) {}
void Shape(UChar *buff, size_t length);
};
/**
* Wrapper for doing layouts with ICU.
*/
class ICUParagraphLayout : public ParagraphLayouter {
public:
/** Visual run contains data about the bit of text with the same font. */
class ICUVisualRun : public ParagraphLayouter::VisualRun {
private:
std::vector glyphs;
std::vector positions;
std::vector glyph_to_char;
int total_advance;
const Font *font;
public:
ICUVisualRun(const ICURun &run, int x);
const GlyphID *GetGlyphs() const override { return this->glyphs.data(); }
const float *GetPositions() const override { return this->positions.data(); }
const int *GetGlyphToCharMap() const override { return this->glyph_to_char.data(); }
const Font *GetFont() const override { return this->font; }
int GetLeading() const override { return this->font->fc->GetHeight(); }
int GetGlyphCount() const override { return this->glyphs.size(); }
int GetAdvance() const { return this->total_advance; }
};
/** A single line worth of VisualRuns. */
class ICULine : public std::vector, public ParagraphLayouter::Line {
public:
int GetLeading() const override;
int GetWidth() const override;
int CountRuns() const override { return (uint)this->size(); }
const VisualRun &GetVisualRun(int run) const override { return this->at(run); }
int GetInternalCharLength(char32_t c) const override
{
/* ICU uses UTF-16 internally which means we need to account for surrogate pairs. */
return c >= 0x010000U ? 2 : 1;
}
};
private:
std::vector runs;
UChar *buff;
size_t buff_length;
std::vector::iterator current_run;
int partial_offset;
public:
ICUParagraphLayout(std::vector &runs, UChar *buff, size_t buff_length) : runs(runs), buff(buff), buff_length(buff_length)
{
this->Reflow();
}
~ICUParagraphLayout() override { }
void Reflow() override
{
this->current_run = this->runs.begin();
this->partial_offset = 0;
}
std::unique_ptr NextLine(int max_width) override;
};
/**
* Constructor for a new ICUVisualRun.
*
* It bases all information on the ICURun, which should already be shaped.
*
* @param run The ICURun to base the visual run on.
* @param x The offset of the run on the line.
*/
ICUParagraphLayout::ICUVisualRun::ICUVisualRun(const ICURun &run, int x) :
glyphs(run.glyphs), glyph_to_char(run.glyph_to_char), total_advance(run.total_advance), font(run.font)
{
/* If there are no positions, the ICURun was not Shaped; that should never happen. */
assert(run.positions.size() != 0);
this->positions.reserve(run.positions.size());
/* "positions" is an array of x/y. So we need to alternate. */
bool is_x = true;
for (auto &position : run.positions) {
if (is_x) {
this->positions.push_back(position + x);
} else {
this->positions.push_back(position);
}
is_x = !is_x;
}
}
/**
* Shape a single run.
*
* @param buff The buffer of which a partial (depending on start/length of the run) will be shaped.
* @param length The length of the buffer.
*/
void ICURun::Shape(UChar *buff, size_t buff_length) {
auto hbfont = hb_ft_font_create_referenced(*(static_cast(font->fc->GetOSHandle())));
hb_font_set_scale(hbfont, this->font->fc->GetFontSize() * FONT_SCALE, this->font->fc->GetFontSize() * FONT_SCALE);
/* ICU buffer is in UTF-16. */
auto hbbuf = hb_buffer_create();
hb_buffer_add_utf16(hbbuf, reinterpret_cast(buff), buff_length, this->start, this->length);
/* Set all the properties of this segment. */
hb_buffer_set_direction(hbbuf, (this->level & 1) == 1 ? HB_DIRECTION_RTL : HB_DIRECTION_LTR);
hb_buffer_set_script(hbbuf, hb_script_from_string(uscript_getShortName(this->script), -1));
hb_buffer_set_language(hbbuf, hb_language_from_string(_current_language->isocode, -1));
hb_buffer_set_cluster_level(hbbuf, HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES);
/* Shape the segment. */
hb_shape(hbfont, hbbuf, nullptr, 0);
unsigned int glyph_count;
auto glyph_info = hb_buffer_get_glyph_infos(hbbuf, &glyph_count);
auto glyph_pos = hb_buffer_get_glyph_positions(hbbuf, &glyph_count);
/* Make sure any former run is lost. */
this->glyphs.clear();
this->glyph_to_char.clear();
this->positions.clear();
this->advance.clear();
/* Reserve space, as we already know the size. */
this->glyphs.reserve(glyph_count);
this->glyph_to_char.reserve(glyph_count);
this->positions.reserve(glyph_count * 2 + 2);
this->advance.reserve(glyph_count);
/* Prepare the glyphs/position. ICUVisualRun will give the position an offset if needed. */
hb_position_t advance = 0;
for (unsigned int i = 0; i < glyph_count; i++) {
int x_advance;
if (buff[glyph_info[i].cluster] >= SCC_SPRITE_START && buff[glyph_info[i].cluster] <= SCC_SPRITE_END) {
auto glyph = this->font->fc->MapCharToGlyph(buff[glyph_info[i].cluster]);
this->glyphs.push_back(glyph);
this->positions.push_back(advance);
this->positions.push_back((this->font->fc->GetHeight() - ScaleSpriteTrad(FontCache::GetDefaultFontHeight(this->font->fc->GetSize()))) / 2); // Align sprite font to centre
x_advance = this->font->fc->GetGlyphWidth(glyph);
} else {
this->glyphs.push_back(glyph_info[i].codepoint);
this->positions.push_back(glyph_pos[i].x_offset / FONT_SCALE + advance);
this->positions.push_back(glyph_pos[i].y_offset / FONT_SCALE);
x_advance = glyph_pos[i].x_advance / FONT_SCALE;
}
this->glyph_to_char.push_back(glyph_info[i].cluster);
this->advance.push_back(x_advance);
advance += x_advance;
}
/* Position has one more element to close off the array. */
this->positions.push_back(advance);
this->positions.push_back(0);
/* Track the total advancement we made. */
this->total_advance = advance;
hb_buffer_destroy(hbbuf);
hb_font_destroy(hbfont);
}
/**
* Get the height of the line.
* @return The maximum height of the line.
*/
int ICUParagraphLayout::ICULine::GetLeading() const
{
int leading = 0;
for (const auto &run : *this) {
leading = std::max(leading, run.GetLeading());
}
return leading;
}
/**
* Get the width of this line.
* @return The width of the line.
*/
int ICUParagraphLayout::ICULine::GetWidth() const
{
int length = 0;
for (const auto &run : *this) {
length += run.GetAdvance();
}
return length;
}
/**
* Itemize the string into runs per embedding level.
*
* Later on, based on the levels, we can deduce the order of a subset of runs.
*
* @param buff The string to itemize.
* @param length The length of the string.
* @return The runs.
*/
std::vector ItemizeBidi(UChar *buff, size_t length)
{
auto ubidi = ubidi_open();
auto parLevel = _current_text_dir == TD_RTL ? UBIDI_RTL : UBIDI_LTR;
UErrorCode err = U_ZERO_ERROR;
ubidi_setPara(ubidi, buff, length, parLevel, nullptr, &err);
if (U_FAILURE(err)) {
Debug(fontcache, 0, "Failed to set paragraph: %s", u_errorName(err));
ubidi_close(ubidi);
return std::vector();
}
int32_t count = ubidi_countRuns(ubidi, &err);
if (U_FAILURE(err)) {
Debug(fontcache, 0, "Failed to count runs: %s", u_errorName(err));
ubidi_close(ubidi);
return std::vector();
}
std::vector runs;
runs.reserve(count);
/* Find the breakpoints for the logical runs. So we get runs that say "from START to END". */
int32_t logical_pos = 0;
while (static_cast(logical_pos) < length) {
auto start_pos = logical_pos;
/* Fetch the embedding level, so we can order bidi correctly later on. */
UBiDiLevel level;
ubidi_getLogicalRun(ubidi, start_pos, &logical_pos, &level);
runs.emplace_back(ICURun(start_pos, logical_pos - start_pos, level));
}
assert(static_cast(count) == runs.size());
ubidi_close(ubidi);
return runs;
}
/**
* Itemize the string into runs per script, based on the previous created runs.
*
* Basically, this always returns the same or more runs than given.
*
* @param buff The string to itemize.
* @param length The length of the string.
* @param runs_current The current runs.
* @return The runs.
*/
std::vector ItemizeScript(UChar *buff, size_t length, std::vector &runs_current)
{
std::vector runs;
icu::ScriptRun script_itemizer(buff, length);
int cur_pos = 0;
auto cur_run = runs_current.begin();
while (true) {
while (cur_pos < script_itemizer.getScriptEnd() && cur_run != runs_current.end()) {
int stop_pos = std::min(script_itemizer.getScriptEnd(), cur_run->start + cur_run->length);
assert(stop_pos - cur_pos > 0);
runs.push_back(ICURun(cur_pos, stop_pos - cur_pos, cur_run->level, script_itemizer.getScriptCode()));
if (stop_pos == cur_run->start + cur_run->length) cur_run++;
cur_pos = stop_pos;
}
if (!script_itemizer.next()) break;
}
return runs;
}
/**
* Itemize the string into runs per style, based on the previous created runs.
*
* Basically, this always returns the same or more runs than given.
*
* @param runs_current The current runs.
* @param font_mapping The font mapping.
* @return The runs.
*/
std::vector ItemizeStyle(std::vector &runs_current, FontMap &font_mapping)
{
std::vector runs;
int cur_pos = 0;
auto cur_run = runs_current.begin();
for (auto const &font_map : font_mapping) {
while (cur_pos < font_map.first && cur_run != runs_current.end()) {
int stop_pos = std::min(font_map.first, cur_run->start + cur_run->length);
assert(stop_pos - cur_pos > 0);
runs.push_back(ICURun(cur_pos, stop_pos - cur_pos, cur_run->level, cur_run->script, font_map.second));
if (stop_pos == cur_run->start + cur_run->length) cur_run++;
cur_pos = stop_pos;
}
}
return runs;
}
/* static */ ParagraphLayouter *ICUParagraphLayoutFactory::GetParagraphLayout(UChar *buff, UChar *buff_end, FontMap &font_mapping)
{
size_t length = buff_end - buff;
/* Can't layout an empty string. */
if (length == 0) return nullptr;
/* Can't layout our in-built sprite fonts. */
for (auto const &pair : font_mapping) {
if (pair.second->fc->IsBuiltInFont()) return nullptr;
}
auto runs = ItemizeBidi(buff, length);
runs = ItemizeScript(buff, length, runs);
runs = ItemizeStyle(runs, font_mapping);
if (runs.size() == 0) return nullptr;
for (auto &run : runs) {
run.Shape(buff, length);
}
return new ICUParagraphLayout(runs, buff, length);
}
std::unique_ptr ICUParagraphLayout::NextLine(int max_width)
{
std::vector::iterator start_run = this->current_run;
std::vector::iterator last_run = this->current_run;
if (start_run == this->runs.end()) return nullptr;
int cur_width = 0;
/* Add remaining width of the first run if it is a broken run. */
if (this->partial_offset > 0) {
if ((start_run->level & 1) == 0) {
for (size_t i = this->partial_offset; i < start_run->advance.size(); i++) {
cur_width += start_run->advance[i];
}
} else {
for (int i = 0; i < this->partial_offset; i++) {
cur_width += start_run->advance[i];
}
}
last_run++;
}
/* Gather runs until the line is full. */
while (last_run != this->runs.end() && cur_width < max_width) {
cur_width += last_run->total_advance;
last_run++;
}
/* If the text does not fit into the available width, find a suitable breaking point. */
int new_partial_length = 0;
if (cur_width > max_width) {
auto locale = icu::Locale(_current_language->isocode);
/* Create a break-iterator to find a good place to break lines. */
UErrorCode err = U_ZERO_ERROR;
auto break_iterator = icu::BreakIterator::createLineInstance(locale, err);
break_iterator->setText(icu::UnicodeString(this->buff, this->buff_length));
auto overflow_run = last_run - 1;
/* Find the last glyph that fits. */
size_t index;
if ((overflow_run->level & 1) == 0) {
/* LTR */
for (index = overflow_run->glyphs.size(); index > 0; index--) {
cur_width -= overflow_run->advance[index - 1];
if (cur_width <= max_width) break;
}
index--;
} else {
/* RTL */
for (index = 0; index < overflow_run->glyphs.size(); index++) {
cur_width -= overflow_run->advance[index];
if (cur_width <= max_width) break;
}
}
/* Find the character that matches; this is the start of the cluster. */
auto char_pos = overflow_run->glyph_to_char[index];
/* See if there is a good breakpoint inside this run. */
int32_t break_pos = break_iterator->preceding(char_pos + 1);
if (break_pos != icu::BreakIterator::DONE && break_pos > overflow_run->start + this->partial_offset) {
/* There is a line-break inside this run that is suitable. */
new_partial_length = break_pos - overflow_run->start - this->partial_offset;
} else if (overflow_run != start_run) {
/* There is no suitable line-break in this run, but it is also not
* the only run on this line. So we remove the run. */
last_run--;
} else {
/* There is no suitable line-break and this is the only run on the
* line. So we break at the cluster. This is not pretty, but the
* best we can do. */
new_partial_length = char_pos - overflow_run->start - this->partial_offset;
}
}
/* Reorder the runs on this line for display. */
std::vector bidi_level;
for (auto run = start_run; run != last_run; run++) {
bidi_level.push_back(run->level);
}
std::vector vis_to_log(bidi_level.size());
ubidi_reorderVisual(bidi_level.data(), bidi_level.size(), vis_to_log.data());
/* Create line. */
std::unique_ptr line(new ICULine());
int cur_pos = 0;
for (auto &i : vis_to_log) {
auto i_run = start_run + i;
/* Copy the ICURun here, so we can modify it in case of a partial. */
ICURun run = *i_run;
if (i_run == last_run - 1 && new_partial_length > 0) {
if (i_run == start_run && this->partial_offset > 0) {
assert(run.length > this->partial_offset);
run.start += this->partial_offset;
run.length -= this->partial_offset;
}
assert(run.length > new_partial_length);
run.length = new_partial_length;
run.Shape(this->buff, this->buff_length);
} else if (i_run == start_run && this->partial_offset > 0) {
assert(run.length > this->partial_offset);
run.start += this->partial_offset;
run.length -= this->partial_offset;
run.Shape(this->buff, this->buff_length);
}
auto total_advance = run.total_advance;
line->emplace_back(std::move(run), cur_pos);
cur_pos += total_advance;
}
if (new_partial_length > 0) {
this->current_run = last_run - 1;
this->partial_offset += new_partial_length;
} else {
this->current_run = last_run;
this->partial_offset = 0;
}
return line;
}
/* static */ size_t ICUParagraphLayoutFactory::AppendToBuffer(UChar *buff, const UChar *buffer_last, char32_t c)
{
assert(buff < buffer_last);
/* Transform from UTF-32 to internal ICU format of UTF-16. */
int32_t length = 0;
UErrorCode err = U_ZERO_ERROR;
u_strFromUTF32(buff, buffer_last - buff, &length, (UChar32*)&c, 1, &err);
return length;
}