gecko-dev/layout/generic/nsTextFrameUtils.cpp
Emilio Cobos Álvarez e07d5eb564 Bug 1657437 - Don't treat lone CRs as segment breaks. r=jfkthame
That prevents preceding whitespace from getting collapsed.

When there's a single lone CR (so `a\rb`) our behavior here diverges
from Chrome's but matches Safari's. We treat it as ZWSP.

That matches the initial resolution of [1], but then there have been
various doing and undoings of that resolution, so it's not totally clear
to me what the correct behavior per spec should be. I think "treat it as
other control character"? But I haven't dug into what that implies, so
for now I've just kept behavior there as-is.

[1]: https://github.com/w3c/csswg-drafts/issues/855

Differential Revision: https://phabricator.services.mozilla.com/D86188
2020-08-08 16:21:35 +00:00

403 lines
14 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "nsTextFrameUtils.h"
#include "mozilla/dom/Text.h"
#include "nsBidiUtils.h"
#include "nsCharTraits.h"
#include "nsIContent.h"
#include "nsStyleStruct.h"
#include "nsTextFragment.h"
#include "nsUnicharUtils.h"
#include "nsUnicodeProperties.h"
#include <algorithm>
using namespace mozilla;
using namespace mozilla::dom;
// static
bool nsTextFrameUtils::IsSpaceCombiningSequenceTail(const char16_t* aChars,
int32_t aLength) {
return aLength > 0 &&
(mozilla::unicode::IsClusterExtender(aChars[0]) ||
(IsBidiControl(aChars[0]) &&
IsSpaceCombiningSequenceTail(aChars + 1, aLength - 1)));
}
static bool IsDiscardable(char16_t ch, nsTextFrameUtils::Flags* aFlags) {
// Unlike IS_DISCARDABLE, we don't discard \r. \r will be ignored by
// gfxTextRun and discarding it would force us to copy text in many cases of
// preformatted text containing \r\n.
if (ch == CH_SHY) {
*aFlags |= nsTextFrameUtils::Flags::HasShy;
return true;
}
return IsBidiControl(ch);
}
static bool IsDiscardable(uint8_t ch, nsTextFrameUtils::Flags* aFlags) {
if (ch == CH_SHY) {
*aFlags |= nsTextFrameUtils::Flags::HasShy;
return true;
}
return false;
}
static bool IsSegmentBreak(char16_t aCh) { return aCh == '\n'; }
static bool IsSpaceOrTab(char16_t aCh) { return aCh == ' ' || aCh == '\t'; }
static bool IsSpaceOrTabOrSegmentBreak(char16_t aCh) {
return IsSpaceOrTab(aCh) || IsSegmentBreak(aCh);
}
template <typename CharT>
/* static */
bool nsTextFrameUtils::IsSkippableCharacterForTransformText(CharT aChar) {
return aChar == ' ' || aChar == '\t' || aChar == '\n' || aChar == CH_SHY ||
(aChar > 0xFF && IsBidiControl(aChar));
}
#ifdef DEBUG
template <typename CharT>
static void AssertSkippedExpectedChars(const CharT* aText,
const gfxSkipChars& aSkipChars,
int32_t aSkipCharsOffset) {
gfxSkipCharsIterator it(aSkipChars);
it.AdvanceOriginal(aSkipCharsOffset);
while (it.GetOriginalOffset() < it.GetOriginalEnd()) {
CharT ch = aText[it.GetOriginalOffset() - aSkipCharsOffset];
MOZ_ASSERT(!it.IsOriginalCharSkipped() ||
nsTextFrameUtils::IsSkippableCharacterForTransformText(ch),
"skipped unexpected character; need to update "
"IsSkippableCharacterForTransformText?");
it.AdvanceOriginal(1);
}
}
#endif
template <class CharT>
static CharT* TransformWhiteSpaces(
const CharT* aText, uint32_t aLength, uint32_t aBegin, uint32_t aEnd,
bool aHasSegmentBreak, bool& aInWhitespace, CharT* aOutput,
nsTextFrameUtils::Flags& aFlags,
nsTextFrameUtils::CompressionMode aCompression, gfxSkipChars* aSkipChars) {
MOZ_ASSERT(aCompression == nsTextFrameUtils::COMPRESS_WHITESPACE ||
aCompression == nsTextFrameUtils::COMPRESS_WHITESPACE_NEWLINE,
"whitespaces should be skippable!!");
// Get the context preceding/following this white space range.
// For 8-bit text (sizeof CharT == 1), the checks here should get optimized
// out, and isSegmentBreakSkippable should be initialized to be 'false'.
bool isSegmentBreakSkippable =
sizeof(CharT) > 1 &&
((aBegin > 0 && IS_ZERO_WIDTH_SPACE(aText[aBegin - 1])) ||
(aEnd < aLength && IS_ZERO_WIDTH_SPACE(aText[aEnd])));
if (sizeof(CharT) > 1 && !isSegmentBreakSkippable && aBegin > 0 &&
aEnd < aLength) {
uint32_t ucs4before;
uint32_t ucs4after;
if (aBegin > 1 &&
NS_IS_SURROGATE_PAIR(aText[aBegin - 2], aText[aBegin - 1])) {
ucs4before = SURROGATE_TO_UCS4(aText[aBegin - 2], aText[aBegin - 1]);
} else {
ucs4before = aText[aBegin - 1];
}
if (aEnd + 1 < aLength &&
NS_IS_SURROGATE_PAIR(aText[aEnd], aText[aEnd + 1])) {
ucs4after = SURROGATE_TO_UCS4(aText[aEnd], aText[aEnd + 1]);
} else {
ucs4after = aText[aEnd];
}
// Discard newlines between characters that have F, W, or H
// EastAsianWidth property and neither side is Hangul.
isSegmentBreakSkippable =
IsSegmentBreakSkipChar(ucs4before) && IsSegmentBreakSkipChar(ucs4after);
}
for (uint32_t i = aBegin; i < aEnd; ++i) {
CharT ch = aText[i];
bool keepChar = false;
bool keepTransformedWhiteSpace = false;
if (IsDiscardable(ch, &aFlags)) {
aSkipChars->SkipChar();
continue;
}
if (IsSpaceOrTab(ch)) {
if (aHasSegmentBreak) {
// If white-space is set to normal, nowrap, or pre-line, white space
// characters are considered collapsible and all spaces and tabs
// immediately preceding or following a segment break are removed.
aSkipChars->SkipChar();
continue;
}
if (aInWhitespace) {
aSkipChars->SkipChar();
continue;
} else {
keepTransformedWhiteSpace = true;
}
} else {
// Apply Segment Break Transformation Rules (CSS Text 3 - 4.1.2) for
// segment break characters.
if (aCompression == nsTextFrameUtils::COMPRESS_WHITESPACE ||
// XXX: According to CSS Text 3, a lone CR should not always be
// kept, but still go through the Segment Break Transformation
// Rules. However, this is what current modern browser engines
// (webkit/blink/edge) do. So, once we can get some clarity
// from the specification issue, we should either remove the
// lone CR condition here, or leave it here with this comment
// being rephrased.
// Please see https://github.com/w3c/csswg-drafts/issues/855.
ch == '\r') {
keepChar = true;
} else {
// aCompression == COMPRESS_WHITESPACE_NEWLINE
// Any collapsible segment break immediately following another
// collapsible segment break is removed. Then the remaining segment
// break is either transformed into a space (U+0020) or removed
// depending on the context before and after the break.
if (isSegmentBreakSkippable || aInWhitespace) {
aSkipChars->SkipChar();
continue;
}
isSegmentBreakSkippable = true;
keepTransformedWhiteSpace = true;
}
}
if (keepChar) {
*aOutput++ = ch;
aSkipChars->KeepChar();
aInWhitespace = IsSpaceOrTab(ch);
} else if (keepTransformedWhiteSpace) {
*aOutput++ = ' ';
aSkipChars->KeepChar();
aInWhitespace = true;
} else {
MOZ_ASSERT_UNREACHABLE("Should've skipped the character!!");
}
}
return aOutput;
}
template <class CharT>
CharT* nsTextFrameUtils::TransformText(const CharT* aText, uint32_t aLength,
CharT* aOutput,
CompressionMode aCompression,
uint8_t* aIncomingFlags,
gfxSkipChars* aSkipChars,
Flags* aAnalysisFlags) {
Flags flags = Flags();
#ifdef DEBUG
int32_t skipCharsOffset = aSkipChars->GetOriginalCharCount();
#endif
bool lastCharArabic = false;
if (aCompression == COMPRESS_NONE ||
aCompression == COMPRESS_NONE_TRANSFORM_TO_SPACE) {
// Skip discardables.
uint32_t i;
for (i = 0; i < aLength; ++i) {
CharT ch = aText[i];
if (IsDiscardable(ch, &flags)) {
aSkipChars->SkipChar();
} else {
aSkipChars->KeepChar();
if (ch > ' ') {
lastCharArabic = IS_ARABIC_CHAR(ch);
} else if (aCompression == COMPRESS_NONE_TRANSFORM_TO_SPACE) {
if (ch == '\t' || ch == '\n') {
ch = ' ';
}
} else {
// aCompression == COMPRESS_NONE
if (ch == '\t') {
flags |= Flags::HasTab;
}
}
*aOutput++ = ch;
}
}
if (lastCharArabic) {
*aIncomingFlags |= INCOMING_ARABICCHAR;
} else {
*aIncomingFlags &= ~INCOMING_ARABICCHAR;
}
*aIncomingFlags &= ~INCOMING_WHITESPACE;
} else {
bool inWhitespace = (*aIncomingFlags & INCOMING_WHITESPACE) != 0;
uint32_t i;
for (i = 0; i < aLength; ++i) {
CharT ch = aText[i];
// CSS Text 3 - 4.1. The White Space Processing Rules
// White space processing in CSS affects only the document white space
// characters: spaces (U+0020), tabs (U+0009), and segment breaks.
// Since we need the context of segment breaks and their surrounding
// white spaces to proceed the white space processing, a consecutive run
// of spaces/tabs/segment breaks is collected in a first pass loop, then
// we apply the collapsing and transformation rules to this run in a
// second pass loop.
if (IsSpaceOrTabOrSegmentBreak(ch)) {
bool keepLastSpace = false;
bool hasSegmentBreak = IsSegmentBreak(ch);
uint32_t countTrailingDiscardables = 0;
uint32_t j;
for (j = i + 1; j < aLength && (IsSpaceOrTabOrSegmentBreak(aText[j]) ||
IsDiscardable(aText[j], &flags));
j++) {
if (IsSegmentBreak(aText[j])) {
hasSegmentBreak = true;
}
}
// Exclude trailing discardables before checking space combining
// sequence tail.
for (; IsDiscardable(aText[j - 1], &flags); j--) {
countTrailingDiscardables++;
}
// If the last white space is followed by a combining sequence tail,
// exclude it from the range of TransformWhiteSpaces.
if (sizeof(CharT) > 1 && aText[j - 1] == ' ' && j < aLength &&
IsSpaceCombiningSequenceTail(&aText[j], aLength - j)) {
keepLastSpace = true;
j--;
}
if (j > i) {
aOutput = TransformWhiteSpaces(aText, aLength, i, j, hasSegmentBreak,
inWhitespace, aOutput, flags,
aCompression, aSkipChars);
}
// We need to keep KeepChar()/SkipChar() in order, so process the
// last white space first, then process the trailing discardables.
if (keepLastSpace) {
keepLastSpace = false;
*aOutput++ = ' ';
aSkipChars->KeepChar();
lastCharArabic = false;
j++;
}
for (; countTrailingDiscardables > 0; countTrailingDiscardables--) {
aSkipChars->SkipChar();
j++;
}
i = j - 1;
continue;
}
// Process characters other than the document white space characters.
if (IsDiscardable(ch, &flags)) {
aSkipChars->SkipChar();
} else {
*aOutput++ = ch;
aSkipChars->KeepChar();
}
lastCharArabic = IS_ARABIC_CHAR(ch);
inWhitespace = false;
}
if (lastCharArabic) {
*aIncomingFlags |= INCOMING_ARABICCHAR;
} else {
*aIncomingFlags &= ~INCOMING_ARABICCHAR;
}
if (inWhitespace) {
*aIncomingFlags |= INCOMING_WHITESPACE;
} else {
*aIncomingFlags &= ~INCOMING_WHITESPACE;
}
}
*aAnalysisFlags = flags;
#ifdef DEBUG
AssertSkippedExpectedChars(aText, *aSkipChars, skipCharsOffset);
#endif
return aOutput;
}
/*
* NOTE: The TransformText and IsSkippableCharacterForTransformText template
* functions are part of the public API of nsTextFrameUtils, while
* their function bodies are not available in the header. They may stop working
* (fail to resolve symbol in link time) once their callsites are moved to a
* different translation unit (e.g. a different unified source file).
* Explicit instantiating this function template with `uint8_t` and `char16_t`
* could prevent us from the potential risk.
*/
template uint8_t* nsTextFrameUtils::TransformText(
const uint8_t* aText, uint32_t aLength, uint8_t* aOutput,
CompressionMode aCompression, uint8_t* aIncomingFlags,
gfxSkipChars* aSkipChars, Flags* aAnalysisFlags);
template char16_t* nsTextFrameUtils::TransformText(
const char16_t* aText, uint32_t aLength, char16_t* aOutput,
CompressionMode aCompression, uint8_t* aIncomingFlags,
gfxSkipChars* aSkipChars, Flags* aAnalysisFlags);
template bool nsTextFrameUtils::IsSkippableCharacterForTransformText(
uint8_t aChar);
template bool nsTextFrameUtils::IsSkippableCharacterForTransformText(
char16_t aChar);
uint32_t nsTextFrameUtils::ComputeApproximateLengthWithWhitespaceCompression(
Text* aText, const nsStyleText* aStyleText) {
const nsTextFragment* frag = &aText->TextFragment();
// This is an approximation so we don't really need anything
// too fancy here.
uint32_t len;
if (aStyleText->WhiteSpaceIsSignificant()) {
len = frag->GetLength();
} else {
bool is2b = frag->Is2b();
union {
const char* s1b;
const char16_t* s2b;
} u;
if (is2b) {
u.s2b = frag->Get2b();
} else {
u.s1b = frag->Get1b();
}
bool prevWS = true; // more important to ignore blocks with
// only whitespace than get inline boundaries
// exactly right
len = 0;
for (uint32_t i = 0, i_end = frag->GetLength(); i < i_end; ++i) {
char16_t c = is2b ? u.s2b[i] : u.s1b[i];
if (c == ' ' || c == '\n' || c == '\t' || c == '\r') {
if (!prevWS) {
++len;
}
prevWS = true;
} else {
++len;
prevWS = false;
}
}
}
return len;
}
bool nsSkipCharsRunIterator::NextRun() {
do {
if (mRunLength) {
mIterator.AdvanceOriginal(mRunLength);
NS_ASSERTION(mRunLength > 0,
"No characters in run (initial length too large?)");
if (!mSkipped || mLengthIncludesSkipped) {
mRemainingLength -= mRunLength;
}
}
if (!mRemainingLength) {
return false;
}
int32_t length;
mSkipped = mIterator.IsOriginalCharSkipped(&length);
mRunLength = std::min(length, mRemainingLength);
} while (!mVisitSkipped && mSkipped);
return true;
}