#include "Dimension.h"
#include "support/convert.h"
+#include "support/debug.h"
#include "support/lassert.h"
#include "support/lyxlib.h"
-#include "support/debug.h"
+#include "support/textutils.h"
#define DISABLE_PMPROF
#include "support/pmprof.h"
#include <QByteArray>
#include <QRawFont>
#include <QtEndian>
+
+#if QT_VERSION >= 0x050100
#include <QtMath>
+#else
+#define qDegreesToRadians(degree) (degree * (M_PI / 180))
+#endif
using namespace std;
using namespace lyx::support;
#ifdef BIDI_USE_OVERRIDE
# define BIDI_OFFSET 1
+/* Unicode override characters enforce drawing direction
+ * Source: http://www.iamcal.com/understanding-bidirectional-text/
+ * Right-to-left override is 0x202e and left-to-right override is 0x202d.
+ */
+QChar const bidi_override[2] = {0x202d, 0x202e};
#else
# define BIDI_OFFSET 0
#endif
# error "Define at least one of BIDI_USE_OVERRIDE or BIDI_USE_FLAG"
#endif
+
+#if QT_VERSION < 0x050000
+inline uint qHash(double key)
+{
+ return qHash(QByteArray(reinterpret_cast<char const *>(&key), sizeof(key)));
+}
+#endif
+
+
namespace std {
/*
namespace frontend {
-/*
- * Limit (strwidth|breakat)_cache_ size to 512kB of string data.
- * Limit qtextlayout_cache_ size to 500 elements (we do not know the
- * size of the QTextLayout objects anyway).
- * Note that all these numbers are arbitrary.
- * Also, setting size to 0 is tantamount to disabling the cache.
- */
-int cache_metrics_width_size = 1 << 19;
-int cache_metrics_breakat_size = 1 << 19;
+namespace {
+// Maximal size/cost for various caches. See QCache documentation to
+// see what cost means.
+
+// Limit strwidth_cache_ total cost to 1MB of string data.
+int const strwidth_cache_max_cost = 1024 * 1024;
+// Limit breakstr_cache_ total cost to 10MB of string data.
+// This is useful for documents with very large insets.
+int const breakstr_cache_max_cost = 10 * 1024 * 1024;
// Qt 5.x already has its own caching of QTextLayout objects
// but it does not seem to work well on MacOS X.
#if (QT_VERSION < 0x050000) || defined(Q_OS_MAC)
-int cache_metrics_qtextlayout_size = 500;
+// Limit qtextlayout_cache_ size to 500 elements (we do not know the
+// size of the QTextLayout objects anyway).
+int const qtextlayout_cache_max_size = 500;
#else
-int cache_metrics_qtextlayout_size = 0;
+// Disable the cache
+int const qtextlayout_cache_max_size = 0;
#endif
-namespace {
/**
* Convert a UCS4 character into a QChar.
* This is a hack (it does only make sense for the common part of the UCS4
GuiFontMetrics::GuiFontMetrics(QFont const & font)
: font_(font), metrics_(font, 0),
- strwidth_cache_(cache_metrics_width_size),
- breakat_cache_(cache_metrics_breakat_size),
- qtextlayout_cache_(cache_metrics_qtextlayout_size)
+ strwidth_cache_(strwidth_cache_max_cost),
+ breakstr_cache_(breakstr_cache_max_cost),
+ qtextlayout_cache_(qtextlayout_cache_max_size)
{
// Determine italic slope
double const defaultSlope = tan(qDegreesToRadians(19.0));
{
// We add 1 as the value returned by QT is different than X
// See http://doc.trolltech.com/2.3/qfontmetrics.html#200b74
+ // FIXME: check this
return metrics_.descent() + 1;
}
int GuiFontMetrics::width(docstring const & s) const
{
PROFILE_THIS_BLOCK(width);
- if (strwidth_cache_.contains(s))
- return strwidth_cache_[s];
+ if (int * wid_p = strwidth_cache_.object_ptr(s))
+ return *wid_p;
PROFILE_CACHE_MISS(width);
/* Several problems have to be taken into account:
* * QFontMetrics::width does not returns a wrong value with Qt5 with
}
+uint qHash(TextLayoutKey const & key)
+{
+ double params = (2 * key.rtl - 1) * key.ws;
+ return std::qHash(key.s) ^ ::qHash(params);
+}
+
+
+namespace {
+
+// This holds a translation table between the original string and the
+// QString that we can use with QTextLayout.
+struct TextLayoutHelper
+{
+ /// Create the helper
+ /// \c s is the original string
+ /// \c isrtl is true if the string is right-to-left
+ /// \c naked is true to disable the insertion of zero width annotations
+ TextLayoutHelper(docstring const & s, bool isrtl, bool naked = false);
+
+ /// translate QString index to docstring index
+ docstring::size_type qpos2pos(int qpos) const
+ {
+ return lower_bound(pos2qpos_.begin(), pos2qpos_.end(), qpos) - pos2qpos_.begin();
+ }
+
+ /// Translate docstring index to QString index
+ int pos2qpos(docstring::size_type pos) const { return pos2qpos_[pos]; }
+
+ // The original string
+ docstring docstr;
+ // The mirror string
+ QString qstr;
+ // is string right-to-left?
+ bool rtl;
+
+private:
+ // This vector contains the QString pos for each string position
+ vector<int> pos2qpos_;
+};
+
+
+TextLayoutHelper::TextLayoutHelper(docstring const & s, bool isrtl, bool naked)
+ : docstr(s), rtl(isrtl)
+{
+ // Reserve memory for performance purpose
+ pos2qpos_.reserve(s.size());
+ qstr.reserve(2 * s.size());
+
+ /* Qt will not break at a leading or trailing space, and we need
+ * that sometimes, see http://www.lyx.org/trac/ticket/9921.
+ *
+ * To work around the problem, we enclose the string between
+ * word joiner characters so that the QTextLayout algorithm will
+ * agree to break the text at these extremal spaces.
+ */
+ // Unicode character WORD JOINER
+ QChar const word_joiner(0x2060);
+ if (!naked)
+ qstr += word_joiner;
+
+#ifdef BIDI_USE_OVERRIDE
+ if (!naked)
+ qstr += bidi_override[rtl];
+#endif
+
+ // Now translate the string character-by-character.
+ bool was_space = false;
+ for (char_type const c : s) {
+ // insert a word joiner character between consecutive spaces
+ bool const is_space = isSpace(c);
+ if (!naked && is_space && was_space)
+ qstr += word_joiner;
+ was_space = is_space;
+ // Remember the QString index at this point
+ pos2qpos_.push_back(qstr.size());
+ // Performance: UTF-16 characters are easier
+ if (is_utf16(c))
+ qstr += ucs4_to_qchar(c);
+ else
+ qstr += toqstr(c);
+ }
+
+ // Final word joiner (see above)
+ if (!naked)
+ qstr += word_joiner;
+
+ // Add virtual position at the end of the string
+ pos2qpos_.push_back(qstr.size());
+
+ //QString dump = qstr;
+ //LYXERR0("TLH: " << dump.replace(word_joiner, "|").toStdString());
+}
+
+}
+
shared_ptr<QTextLayout const>
GuiFontMetrics::getTextLayout(docstring const & s, bool const rtl,
double const wordspacing) const
{
PROFILE_THIS_BLOCK(getTextLayout);
- docstring const s_cache =
- s + (rtl ? "r" : "l") + convert<docstring>(wordspacing);
- if (auto ptl = qtextlayout_cache_[s_cache])
+ TextLayoutKey key{s, rtl, wordspacing};
+ if (auto ptl = qtextlayout_cache_[key])
return ptl;
PROFILE_CACHE_MISS(getTextLayout);
auto const ptl = make_shared<QTextLayout>();
#endif
#ifdef BIDI_USE_OVERRIDE
- /* Use unicode override characters to enforce drawing direction
- * Source: http://www.iamcal.com/understanding-bidirectional-text/
- */
- if (rtl)
- // Right-to-left override: forces to draw text right-to-left
- ptl->setText(QChar(0x202E) + toqstr(s));
- else
- // Left-to-right override: forces to draw text left-to-right
- ptl->setText(QChar(0x202D) + toqstr(s));
+ ptl->setText(bidi_override[rtl] + toqstr(s));
#else
ptl->setText(toqstr(s));
#endif
ptl->beginLayout();
ptl->createLine();
ptl->endLayout();
- qtextlayout_cache_.insert(s_cache, ptl);
+ qtextlayout_cache_.insert(key, ptl);
return ptl;
}
}
-int GuiFontMetrics::countExpanders(docstring const & str) const
+FontMetrics::Breaks
+GuiFontMetrics::breakString_helper(docstring const & s, int first_wid, int wid,
+ bool rtl, bool force) const
{
- // Numbers of characters that are expanded by inter-word spacing. These
- // characters are spaces, except for characters 09-0D which are treated
- // specially. (From a combination of testing with the notepad found in qt's
- // examples, and reading the source code.) In addition, consecutive spaces
- // only count as one expander.
- bool wasspace = false;
- int nexp = 0;
- for (char_type c : str)
- if (c > 0x0d && QChar(c).isSpace()) {
- if (!wasspace) {
- ++nexp;
- wasspace = true;
- }
- } else
- wasspace = false;
- return nexp;
-}
-
+ TextLayoutHelper const tlh(s, rtl);
-pair<int, int>
-GuiFontMetrics::breakAt_helper(docstring const & s, int const x,
- bool const rtl, bool const force) const
-{
QTextLayout tl;
- /* Qt will not break at a leading or trailing space, and we need
- * that sometimes, see http://www.lyx.org/trac/ticket/9921.
- *
- * To work around the problem, we enclose the string between
- * zero-width characters so that the QTextLayout algorithm will
- * agree to break the text at these extremal spaces.
- */
- // Unicode character ZERO WIDTH NO-BREAK SPACE
- QChar const zerow_nbsp(0xfeff);
- QString qs = zerow_nbsp + toqstr(s) + zerow_nbsp;
#ifdef BIDI_USE_FLAG
/* Use undocumented flag to enforce drawing direction
* FIXME: This does not work with Qt 5.11 (ticket #11284).
*/
tl.setFlags(rtl ? Qt::TextForceRightToLeft : Qt::TextForceLeftToRight);
#endif
-
-#ifdef BIDI_USE_OVERRIDE
- /* Use unicode override characters to enforce drawing direction
- * Source: http://www.iamcal.com/understanding-bidirectional-text/
- */
- if (rtl)
- // Right-to-left override: forces to draw text right-to-left
- qs = QChar(0x202E) + qs;
- else
- // Left-to-right override: forces to draw text left-to-right
- qs = QChar(0x202D) + qs;
-#endif
- int const offset = 1 + BIDI_OFFSET;
-
- tl.setText(qs);
+ tl.setText(tlh.qstr);
tl.setFont(font_);
QTextOption to;
+ /*
+ * Some Asian languages split lines anywhere (no notion of
+ * word). It seems that QTextLayout is not aware of this fact.
+ * See for reference:
+ * https://en.wikipedia.org/wiki/Line_breaking_rules_in_East_Asian_languages
+ *
+ * FIXME: Something shall be done about characters which are
+ * not allowed at the beginning or end of line.
+ */
to.setWrapMode(force ? QTextOption::WrapAtWordBoundaryOrAnywhere
: QTextOption::WordWrap);
tl.setTextOption(to);
+
+ bool first = true;
tl.beginLayout();
- QTextLine line = tl.createLine();
- line.setLineWidth(x);
- tl.createLine();
+ while(true) {
+ QTextLine line = tl.createLine();
+ if (!line.isValid())
+ break;
+ line.setLineWidth(first ? first_wid : wid);
+ first = false;
+ }
tl.endLayout();
- int const line_wid = iround(line.horizontalAdvance());
- if ((force && line.textLength() == offset) || line_wid > x)
- return {-1, -1};
- /* Since QString is UTF-16 and docstring is UCS-4, the offsets may
- * not be the same when there are high-plan unicode characters
- * (bug #10443).
- */
- // The variable `offset' is here to account for the extra leading characters.
- // The ending character zerow_nbsp has to be ignored if the line is complete.
- int const qlen = line.textLength() - offset - (line.textLength() == qs.length());
-#if QT_VERSION < 0x040801 || QT_VERSION >= 0x050100
- int len = qstring_to_ucs4(qs.mid(offset, qlen)).length();
+
+ Breaks breaks;
+ int pos = 0;
+ for (int i = 0 ; i < tl.lineCount() ; ++i) {
+ QTextLine const & line = tl.lineAt(i);
+ int const line_epos = line.textStart() + line.textLength();
+ int const epos = tlh.qpos2pos(line_epos);
+#if QT_VERSION >= 0x050000
+ // This does not take trailing spaces into account, except for the last line.
+ int const wid = iround(line.naturalTextWidth());
+ // If the line is not the last one, trailing space is always omitted.
+ int nspc_wid = wid;
+ // For the last line, compute the width without trailing space
+ if (i + 1 == tl.lineCount() && !s.empty() && isSpace(s.back())
+ && line.textStart() <= tlh.pos2qpos(s.size() - 1))
+ nspc_wid = iround(line.cursorToX(tlh.pos2qpos(s.size() - 1)));
#else
- /* Due to QTBUG-25536 in 4.8.1 <= Qt < 5.1.0, the string returned
- * by QString::toUcs4 (used by qstring_to_ucs4) may have wrong
- * length. We work around the problem by trying all docstring
- * positions until the right one is found. This is slow only if
- * there are many high-plane Unicode characters. It might be
- * worthwhile to implement a dichotomy search if this shows up
- * under a profiler.
- */
- int len = min(qlen, static_cast<int>(s.length()));
- while (len >= 0 && toqstr(s.substr(0, len)).length() != qlen)
- --len;
- LASSERT(len > 0 || qlen == 0, /**/);
+ // With some monospace fonts, the value of horizontalAdvance()
+ // can be wrong with Qt4. One hypothesis is that the invisible
+ // characters that we use are given a non-null width.
+ // FIXME: this is slower than it could be but we'll get rid of Qt4 anyway
+ docstring ss = s.substr(pos, epos - pos);
+ int const wid = width(ss);
+ if (!ss.empty() && isSpace(ss.back()))
+ ss.pop_back();
+ int const nspc_wid = i + 1 < tl.lineCount() ? width(ss) : wid;
+#endif
+ breaks.emplace_back(epos - pos, wid, nspc_wid);
+ pos = epos;
+#if 0
+ // FIXME: should it be kept in some form?
+ if ((force && line.textLength() == brkStrOffset) || line_wid > x)
+ return {-1, line_wid};
#endif
- return {len, line_wid};
+ }
+
+ return breaks;
}
-bool GuiFontMetrics::breakAt(docstring & s, int & x, bool const rtl, bool const force) const
+uint qHash(BreakStringKey const & key)
{
- PROFILE_THIS_BLOCK(breakAt);
- if (s.empty())
- return false;
+ // assume widths are less than 10000. This fits in 32 bits.
+ uint params = key.force + 2 * key.rtl + 4 * key.first_wid + 10000 * key.wid;
+ return std::qHash(key.s) ^ ::qHash(params);
+}
+
- docstring const s_cache =
- s + convert<docstring>(x) + (rtl ? "r" : "l") + (force ? "f" : "w");
- pair<int, int> pp;
+FontMetrics::Breaks GuiFontMetrics::breakString(docstring const & s, int first_wid, int wid,
+ bool rtl, bool force) const
+{
+ PROFILE_THIS_BLOCK(breakString);
+ if (s.empty())
+ return Breaks();
- if (breakat_cache_.contains(s_cache))
- pp = breakat_cache_[s_cache];
+ BreakStringKey key{s, first_wid, wid, rtl, force};
+ Breaks brks;
+ if (auto * brks_ptr = breakstr_cache_.object_ptr(key))
+ brks = *brks_ptr;
else {
- PROFILE_CACHE_MISS(breakAt);
- pp = breakAt_helper(s, x, rtl, force);
- breakat_cache_.insert(s_cache, pp, s_cache.size() * sizeof(char_type));
+ PROFILE_CACHE_MISS(breakString);
+ brks = breakString_helper(s, first_wid, wid, rtl, force);
+ breakstr_cache_.insert(key, brks, sizeof(key) + s.size() * sizeof(char_type));
}
- if (pp.first == -1)
- return false;
- s = s.substr(0, pp.first);
- x = pp.second;
- return true;
+ return brks;
}
if (value != outOfLimitMetric)
return value;
+#if QT_VERSION >= 0x050b00
+ if (is_utf16(c))
+ value = metrics_.horizontalAdvance(ucs4_to_qchar(c));
+ else
+ value = metrics_.horizontalAdvance(toqstr(docstring(1, c)));
+#else
if (is_utf16(c))
value = metrics_.width(ucs4_to_qchar(c));
else
value = metrics_.width(toqstr(docstring(1, c)));
+#endif
width_cache_.insert(c, value);