summaryrefslogtreecommitdiffstats
path: root/servo/components/style/values/specified/text.rs
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--servo/components/style/values/specified/text.rs1154
1 files changed, 1154 insertions, 0 deletions
diff --git a/servo/components/style/values/specified/text.rs b/servo/components/style/values/specified/text.rs
new file mode 100644
index 0000000000..64a896cb27
--- /dev/null
+++ b/servo/components/style/values/specified/text.rs
@@ -0,0 +1,1154 @@
+/* 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 https://mozilla.org/MPL/2.0/. */
+
+//! Specified types for text properties.
+
+use crate::parser::{Parse, ParserContext};
+use crate::properties::longhands::writing_mode::computed_value::T as SpecifiedWritingMode;
+use crate::values::computed::text::LineHeight as ComputedLineHeight;
+use crate::values::computed::text::TextEmphasisStyle as ComputedTextEmphasisStyle;
+use crate::values::computed::text::TextOverflow as ComputedTextOverflow;
+use crate::values::computed::{Context, ToComputedValue};
+use crate::values::generics::text::InitialLetter as GenericInitialLetter;
+use crate::values::generics::text::LineHeight as GenericLineHeight;
+use crate::values::generics::text::{GenericTextDecorationLength, Spacing};
+use crate::values::specified::length::NonNegativeLengthPercentage;
+use crate::values::specified::length::{FontRelativeLength, Length};
+use crate::values::specified::length::{LengthPercentage, NoCalcLength};
+use crate::values::specified::{AllowQuirks, Integer, NonNegativeNumber, Number};
+use cssparser::{Parser, Token};
+use selectors::parser::SelectorParseErrorKind;
+use std::fmt::{self, Write};
+use style_traits::values::SequenceWriter;
+use style_traits::{CssWriter, ParseError, StyleParseErrorKind, ToCss};
+use style_traits::{KeywordsCollectFn, SpecifiedValueInfo};
+use unicode_segmentation::UnicodeSegmentation;
+
+/// A specified type for the `initial-letter` property.
+pub type InitialLetter = GenericInitialLetter<Number, Integer>;
+
+/// A specified value for the `letter-spacing` property.
+pub type LetterSpacing = Spacing<Length>;
+
+/// A specified value for the `word-spacing` property.
+pub type WordSpacing = Spacing<LengthPercentage>;
+
+/// A specified value for the `line-height` property.
+pub type LineHeight = GenericLineHeight<NonNegativeNumber, NonNegativeLengthPercentage>;
+
+/// A value for the `hyphenate-character` property.
+#[derive(
+ Clone,
+ Debug,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(C, u8)]
+pub enum HyphenateCharacter {
+ /// `auto`
+ Auto,
+ /// `<string>`
+ String(crate::OwnedStr),
+}
+
+impl Parse for InitialLetter {
+ fn parse<'i, 't>(
+ context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<Self, ParseError<'i>> {
+ if input
+ .try_parse(|i| i.expect_ident_matching("normal"))
+ .is_ok()
+ {
+ return Ok(GenericInitialLetter::Normal);
+ }
+ let size = Number::parse_at_least_one(context, input)?;
+ let sink = input
+ .try_parse(|i| Integer::parse_positive(context, i))
+ .ok();
+ Ok(GenericInitialLetter::Specified(size, sink))
+ }
+}
+
+impl Parse for LetterSpacing {
+ fn parse<'i, 't>(
+ context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<Self, ParseError<'i>> {
+ Spacing::parse_with(context, input, |c, i| {
+ Length::parse_quirky(c, i, AllowQuirks::Yes)
+ })
+ }
+}
+
+impl Parse for WordSpacing {
+ fn parse<'i, 't>(
+ context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<Self, ParseError<'i>> {
+ Spacing::parse_with(context, input, |c, i| {
+ LengthPercentage::parse_quirky(c, i, AllowQuirks::Yes)
+ })
+ }
+}
+
+impl ToComputedValue for LineHeight {
+ type ComputedValue = ComputedLineHeight;
+
+ #[inline]
+ fn to_computed_value(&self, context: &Context) -> Self::ComputedValue {
+ use crate::values::specified::length::FontBaseSize;
+ match *self {
+ GenericLineHeight::Normal => GenericLineHeight::Normal,
+ #[cfg(feature = "gecko")]
+ GenericLineHeight::MozBlockHeight => GenericLineHeight::MozBlockHeight,
+ GenericLineHeight::Number(number) => {
+ GenericLineHeight::Number(number.to_computed_value(context))
+ },
+ GenericLineHeight::Length(ref non_negative_lp) => {
+ let result = match non_negative_lp.0 {
+ LengthPercentage::Length(NoCalcLength::Absolute(ref abs)) => {
+ context.maybe_zoom_text(abs.to_computed_value(context))
+ },
+ LengthPercentage::Length(ref length) => length.to_computed_value(context),
+ LengthPercentage::Percentage(ref p) => FontRelativeLength::Em(p.0)
+ .to_computed_value(context, FontBaseSize::CurrentStyle),
+ LengthPercentage::Calc(ref calc) => {
+ let computed_calc =
+ calc.to_computed_value_zoomed(context, FontBaseSize::CurrentStyle);
+ let base = context.style().get_font().clone_font_size().computed_size();
+ computed_calc.resolve(base)
+ },
+ };
+ GenericLineHeight::Length(result.into())
+ },
+ }
+ }
+
+ #[inline]
+ fn from_computed_value(computed: &Self::ComputedValue) -> Self {
+ match *computed {
+ GenericLineHeight::Normal => GenericLineHeight::Normal,
+ #[cfg(feature = "gecko")]
+ GenericLineHeight::MozBlockHeight => GenericLineHeight::MozBlockHeight,
+ GenericLineHeight::Number(ref number) => {
+ GenericLineHeight::Number(NonNegativeNumber::from_computed_value(number))
+ },
+ GenericLineHeight::Length(ref length) => {
+ GenericLineHeight::Length(NoCalcLength::from_computed_value(&length.0).into())
+ },
+ }
+ }
+}
+
+/// A generic value for the `text-overflow` property.
+#[derive(
+ Clone,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ PartialEq,
+ Parse,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(C, u8)]
+pub enum TextOverflowSide {
+ /// Clip inline content.
+ Clip,
+ /// Render ellipsis to represent clipped inline content.
+ Ellipsis,
+ /// Render a given string to represent clipped inline content.
+ String(crate::OwnedStr),
+}
+
+#[derive(Clone, Debug, Eq, MallocSizeOf, PartialEq, SpecifiedValueInfo, ToCss, ToShmem)]
+/// text-overflow. Specifies rendering when inline content overflows its line box edge.
+pub struct TextOverflow {
+ /// First value. Applies to end line box edge if no second is supplied; line-left edge otherwise.
+ pub first: TextOverflowSide,
+ /// Second value. Applies to the line-right edge if supplied.
+ pub second: Option<TextOverflowSide>,
+}
+
+impl Parse for TextOverflow {
+ fn parse<'i, 't>(
+ context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<TextOverflow, ParseError<'i>> {
+ let first = TextOverflowSide::parse(context, input)?;
+ let second = input
+ .try_parse(|input| TextOverflowSide::parse(context, input))
+ .ok();
+ Ok(TextOverflow { first, second })
+ }
+}
+
+impl ToComputedValue for TextOverflow {
+ type ComputedValue = ComputedTextOverflow;
+
+ #[inline]
+ fn to_computed_value(&self, _context: &Context) -> Self::ComputedValue {
+ if let Some(ref second) = self.second {
+ Self::ComputedValue {
+ first: self.first.clone(),
+ second: second.clone(),
+ sides_are_logical: false,
+ }
+ } else {
+ Self::ComputedValue {
+ first: TextOverflowSide::Clip,
+ second: self.first.clone(),
+ sides_are_logical: true,
+ }
+ }
+ }
+
+ #[inline]
+ fn from_computed_value(computed: &Self::ComputedValue) -> Self {
+ if computed.sides_are_logical {
+ assert_eq!(computed.first, TextOverflowSide::Clip);
+ TextOverflow {
+ first: computed.second.clone(),
+ second: None,
+ }
+ } else {
+ TextOverflow {
+ first: computed.first.clone(),
+ second: Some(computed.second.clone()),
+ }
+ }
+ }
+}
+
+bitflags! {
+ #[derive(MallocSizeOf, Parse, Serialize, SpecifiedValueInfo, ToCss, ToComputedValue, ToResolvedValue, ToShmem)]
+ #[css(bitflags(single = "none", mixed = "underline,overline,line-through,blink"))]
+ #[repr(C)]
+ /// Specified keyword values for the text-decoration-line property.
+ pub struct TextDecorationLine: u8 {
+ /// No text decoration line is specified.
+ const NONE = 0;
+ /// underline
+ const UNDERLINE = 1 << 0;
+ /// overline
+ const OVERLINE = 1 << 1;
+ /// line-through
+ const LINE_THROUGH = 1 << 2;
+ /// blink
+ const BLINK = 1 << 3;
+ /// Only set by presentation attributes
+ ///
+ /// Setting this will mean that text-decorations use the color
+ /// specified by `color` in quirks mode.
+ ///
+ /// For example, this gives <a href=foo><font color="red">text</font></a>
+ /// a red text decoration
+ #[cfg(feature = "gecko")]
+ const COLOR_OVERRIDE = 0x10;
+ }
+}
+
+impl Default for TextDecorationLine {
+ fn default() -> Self {
+ TextDecorationLine::NONE
+ }
+}
+
+impl TextDecorationLine {
+ #[inline]
+ /// Returns the initial value of text-decoration-line
+ pub fn none() -> Self {
+ TextDecorationLine::NONE
+ }
+}
+
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(C)]
+/// Specified value of the text-transform property, stored in two parts:
+/// the case-related transforms (mutually exclusive, only one may be in effect), and others (non-exclusive).
+pub struct TextTransform {
+ /// Case transform, if any.
+ pub case_: TextTransformCase,
+ /// Non-case transforms.
+ pub other_: TextTransformOther,
+}
+
+impl TextTransform {
+ #[inline]
+ /// Returns the initial value of text-transform
+ pub fn none() -> Self {
+ TextTransform {
+ case_: TextTransformCase::None,
+ other_: TextTransformOther::empty(),
+ }
+ }
+ #[inline]
+ /// Returns whether the value is 'none'
+ pub fn is_none(&self) -> bool {
+ self.case_ == TextTransformCase::None && self.other_.is_empty()
+ }
+}
+
+// TODO: This can be simplified by deriving it.
+impl Parse for TextTransform {
+ fn parse<'i, 't>(
+ _context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<Self, ParseError<'i>> {
+ let mut result = TextTransform::none();
+
+ // Case keywords are mutually exclusive; other transforms may co-occur.
+ loop {
+ let location = input.current_source_location();
+ let ident = match input.next() {
+ Ok(&Token::Ident(ref ident)) => ident,
+ Ok(other) => return Err(location.new_unexpected_token_error(other.clone())),
+ Err(..) => break,
+ };
+
+ match_ignore_ascii_case! { ident,
+ "none" if result.is_none() => {
+ return Ok(result);
+ },
+ "uppercase" if result.case_ == TextTransformCase::None => {
+ result.case_ = TextTransformCase::Uppercase
+ },
+ "lowercase" if result.case_ == TextTransformCase::None => {
+ result.case_ = TextTransformCase::Lowercase
+ },
+ "capitalize" if result.case_ == TextTransformCase::None => {
+ result.case_ = TextTransformCase::Capitalize
+ },
+ "full-width" if !result.other_.intersects(TextTransformOther::FULL_WIDTH) => {
+ result.other_.insert(TextTransformOther::FULL_WIDTH)
+ },
+ "full-size-kana" if !result.other_.intersects(TextTransformOther::FULL_SIZE_KANA) => {
+ result.other_.insert(TextTransformOther::FULL_SIZE_KANA)
+ },
+ _ => return Err(location.new_custom_error(
+ SelectorParseErrorKind::UnexpectedIdent(ident.clone())
+ )),
+ }
+ }
+
+ if result.is_none() {
+ Err(input.new_custom_error(StyleParseErrorKind::UnspecifiedError))
+ } else {
+ Ok(result)
+ }
+ }
+}
+
+impl ToCss for TextTransform {
+ fn to_css<W>(&self, dest: &mut CssWriter<W>) -> fmt::Result
+ where
+ W: Write,
+ {
+ if self.is_none() {
+ return dest.write_str("none");
+ }
+
+ if self.case_ != TextTransformCase::None {
+ self.case_.to_css(dest)?;
+ if !self.other_.is_empty() {
+ dest.write_str(" ")?;
+ }
+ }
+
+ self.other_.to_css(dest)
+ }
+}
+
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(C)]
+/// Specified keyword values for case transforms in the text-transform property. (These are exclusive.)
+pub enum TextTransformCase {
+ /// No case transform.
+ None,
+ /// All uppercase.
+ Uppercase,
+ /// All lowercase.
+ Lowercase,
+ /// Capitalize each word.
+ Capitalize,
+}
+
+bitflags! {
+ #[derive(MallocSizeOf, SpecifiedValueInfo, ToComputedValue, ToResolvedValue, ToShmem)]
+ #[value_info(other_values = "none,full-width,full-size-kana")]
+ #[repr(C)]
+ /// Specified keyword values for non-case transforms in the text-transform property. (Non-exclusive.)
+ pub struct TextTransformOther: u8 {
+ /// full-width
+ const FULL_WIDTH = 1 << 0;
+ /// full-size-kana
+ const FULL_SIZE_KANA = 1 << 1;
+ }
+}
+
+impl ToCss for TextTransformOther {
+ fn to_css<W>(&self, dest: &mut CssWriter<W>) -> fmt::Result
+ where
+ W: Write,
+ {
+ let mut writer = SequenceWriter::new(dest, " ");
+ let mut any = false;
+ macro_rules! maybe_write {
+ ($ident:ident => $str:expr) => {
+ if self.contains(TextTransformOther::$ident) {
+ writer.raw_item($str)?;
+ any = true;
+ }
+ };
+ }
+
+ maybe_write!(FULL_WIDTH => "full-width");
+ maybe_write!(FULL_SIZE_KANA => "full-size-kana");
+
+ debug_assert!(any || self.is_empty());
+
+ Ok(())
+ }
+}
+
+/// Specified and computed value of text-align-last.
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ FromPrimitive,
+ Hash,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+#[repr(u8)]
+pub enum TextAlignLast {
+ Auto,
+ Start,
+ End,
+ Left,
+ Right,
+ Center,
+ Justify,
+}
+
+/// Specified value of text-align keyword value.
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ FromPrimitive,
+ Hash,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+#[repr(u8)]
+pub enum TextAlignKeyword {
+ Start,
+ Left,
+ Right,
+ Center,
+ #[cfg(any(feature = "gecko", feature = "servo-layout-2013"))]
+ Justify,
+ #[css(skip)]
+ #[cfg(feature = "gecko")]
+ Char,
+ End,
+ #[cfg(feature = "gecko")]
+ MozCenter,
+ #[cfg(feature = "gecko")]
+ MozLeft,
+ #[cfg(feature = "gecko")]
+ MozRight,
+ #[cfg(feature = "servo-layout-2013")]
+ ServoCenter,
+ #[cfg(feature = "servo-layout-2013")]
+ ServoLeft,
+ #[cfg(feature = "servo-layout-2013")]
+ ServoRight,
+}
+
+/// Specified value of text-align property.
+#[derive(
+ Clone, Copy, Debug, Eq, Hash, MallocSizeOf, Parse, PartialEq, SpecifiedValueInfo, ToCss, ToShmem,
+)]
+pub enum TextAlign {
+ /// Keyword value of text-align property.
+ Keyword(TextAlignKeyword),
+ /// `match-parent` value of text-align property. It has a different handling
+ /// unlike other keywords.
+ #[cfg(feature = "gecko")]
+ MatchParent,
+ /// This is how we implement the following HTML behavior from
+ /// https://html.spec.whatwg.org/#tables-2:
+ ///
+ /// User agents are expected to have a rule in their user agent style sheet
+ /// that matches th elements that have a parent node whose computed value
+ /// for the 'text-align' property is its initial value, whose declaration
+ /// block consists of just a single declaration that sets the 'text-align'
+ /// property to the value 'center'.
+ ///
+ /// Since selectors can't depend on the ancestor styles, we implement it with a
+ /// magic value that computes to the right thing. Since this is an
+ /// implementation detail, it shouldn't be exposed to web content.
+ #[cfg(feature = "gecko")]
+ #[parse(condition = "ParserContext::in_ua_or_chrome_sheet")]
+ MozCenterOrInherit,
+}
+
+impl ToComputedValue for TextAlign {
+ type ComputedValue = TextAlignKeyword;
+
+ #[inline]
+ fn to_computed_value(&self, _context: &Context) -> Self::ComputedValue {
+ match *self {
+ TextAlign::Keyword(key) => key,
+ #[cfg(feature = "gecko")]
+ TextAlign::MatchParent => {
+ // on the root <html> element we should still respect the dir
+ // but the parent dir of that element is LTR even if it's <html dir=rtl>
+ // and will only be RTL if certain prefs have been set.
+ // In that case, the default behavior here will set it to left,
+ // but we want to set it to right -- instead set it to the default (`start`),
+ // which will do the right thing in this case (but not the general case)
+ if _context.builder.is_root_element {
+ return TextAlignKeyword::Start;
+ }
+ let parent = _context
+ .builder
+ .get_parent_inherited_text()
+ .clone_text_align();
+ let ltr = _context.builder.inherited_writing_mode().is_bidi_ltr();
+ match (parent, ltr) {
+ (TextAlignKeyword::Start, true) => TextAlignKeyword::Left,
+ (TextAlignKeyword::Start, false) => TextAlignKeyword::Right,
+ (TextAlignKeyword::End, true) => TextAlignKeyword::Right,
+ (TextAlignKeyword::End, false) => TextAlignKeyword::Left,
+ _ => parent,
+ }
+ },
+ #[cfg(feature = "gecko")]
+ TextAlign::MozCenterOrInherit => {
+ let parent = _context
+ .builder
+ .get_parent_inherited_text()
+ .clone_text_align();
+ if parent == TextAlignKeyword::Start {
+ TextAlignKeyword::Center
+ } else {
+ parent
+ }
+ },
+ }
+ }
+
+ #[inline]
+ fn from_computed_value(computed: &Self::ComputedValue) -> Self {
+ TextAlign::Keyword(*computed)
+ }
+}
+
+fn fill_mode_is_default_and_shape_exists(
+ fill: &TextEmphasisFillMode,
+ shape: &Option<TextEmphasisShapeKeyword>,
+) -> bool {
+ shape.is_some() && fill.is_filled()
+}
+
+/// Specified value of text-emphasis-style property.
+///
+/// https://drafts.csswg.org/css-text-decor/#propdef-text-emphasis-style
+#[derive(Clone, Debug, MallocSizeOf, PartialEq, SpecifiedValueInfo, ToCss, ToShmem)]
+#[allow(missing_docs)]
+pub enum TextEmphasisStyle {
+ /// [ <fill> || <shape> ]
+ Keyword {
+ #[css(contextual_skip_if = "fill_mode_is_default_and_shape_exists")]
+ fill: TextEmphasisFillMode,
+ shape: Option<TextEmphasisShapeKeyword>,
+ },
+ /// `none`
+ None,
+ /// `<string>` (of which only the first grapheme cluster will be used).
+ String(crate::OwnedStr),
+}
+
+/// Fill mode for the text-emphasis-style property
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToCss,
+ ToComputedValue,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(u8)]
+pub enum TextEmphasisFillMode {
+ /// `filled`
+ Filled,
+ /// `open`
+ Open,
+}
+
+impl TextEmphasisFillMode {
+ /// Whether the value is `filled`.
+ #[inline]
+ pub fn is_filled(&self) -> bool {
+ matches!(*self, TextEmphasisFillMode::Filled)
+ }
+}
+
+/// Shape keyword for the text-emphasis-style property
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToCss,
+ ToComputedValue,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[repr(u8)]
+pub enum TextEmphasisShapeKeyword {
+ /// `dot`
+ Dot,
+ /// `circle`
+ Circle,
+ /// `double-circle`
+ DoubleCircle,
+ /// `triangle`
+ Triangle,
+ /// `sesame`
+ Sesame,
+}
+
+impl ToComputedValue for TextEmphasisStyle {
+ type ComputedValue = ComputedTextEmphasisStyle;
+
+ #[inline]
+ fn to_computed_value(&self, context: &Context) -> Self::ComputedValue {
+ match *self {
+ TextEmphasisStyle::Keyword { fill, shape } => {
+ let shape = shape.unwrap_or_else(|| {
+ // FIXME(emilio, bug 1572958): This should set the
+ // rule_cache_conditions properly.
+ //
+ // Also should probably use WritingMode::is_vertical rather
+ // than the computed value of the `writing-mode` property.
+ if context.style().get_inherited_box().clone_writing_mode() ==
+ SpecifiedWritingMode::HorizontalTb
+ {
+ TextEmphasisShapeKeyword::Circle
+ } else {
+ TextEmphasisShapeKeyword::Sesame
+ }
+ });
+ ComputedTextEmphasisStyle::Keyword { fill, shape }
+ },
+ TextEmphasisStyle::None => ComputedTextEmphasisStyle::None,
+ TextEmphasisStyle::String(ref s) => {
+ // Passing `true` to iterate over extended grapheme clusters, following
+ // recommendation at http://www.unicode.org/reports/tr29/#Grapheme_Cluster_Boundaries
+ //
+ // FIXME(emilio): Doing this at computed value time seems wrong.
+ // The spec doesn't say that this should be a computed-value
+ // time operation. This is observable from getComputedStyle().
+ let s = s.graphemes(true).next().unwrap_or("").to_string();
+ ComputedTextEmphasisStyle::String(s.into())
+ },
+ }
+ }
+
+ #[inline]
+ fn from_computed_value(computed: &Self::ComputedValue) -> Self {
+ match *computed {
+ ComputedTextEmphasisStyle::Keyword { fill, shape } => TextEmphasisStyle::Keyword {
+ fill,
+ shape: Some(shape),
+ },
+ ComputedTextEmphasisStyle::None => TextEmphasisStyle::None,
+ ComputedTextEmphasisStyle::String(ref string) => {
+ TextEmphasisStyle::String(string.clone())
+ },
+ }
+ }
+}
+
+impl Parse for TextEmphasisStyle {
+ fn parse<'i, 't>(
+ _context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<Self, ParseError<'i>> {
+ if input
+ .try_parse(|input| input.expect_ident_matching("none"))
+ .is_ok()
+ {
+ return Ok(TextEmphasisStyle::None);
+ }
+
+ if let Ok(s) = input.try_parse(|i| i.expect_string().map(|s| s.as_ref().to_owned())) {
+ // Handle <string>
+ return Ok(TextEmphasisStyle::String(s.into()));
+ }
+
+ // Handle a pair of keywords
+ let mut shape = input.try_parse(TextEmphasisShapeKeyword::parse).ok();
+ let fill = input.try_parse(TextEmphasisFillMode::parse).ok();
+ if shape.is_none() {
+ shape = input.try_parse(TextEmphasisShapeKeyword::parse).ok();
+ }
+
+ if shape.is_none() && fill.is_none() {
+ return Err(input.new_custom_error(StyleParseErrorKind::UnspecifiedError));
+ }
+
+ // If a shape keyword is specified but neither filled nor open is
+ // specified, filled is assumed.
+ let fill = fill.unwrap_or(TextEmphasisFillMode::Filled);
+
+ // We cannot do the same because the default `<shape>` depends on the
+ // computed writing-mode.
+ Ok(TextEmphasisStyle::Keyword { fill, shape })
+ }
+}
+
+bitflags! {
+ #[derive(MallocSizeOf, SpecifiedValueInfo, ToComputedValue, ToResolvedValue, ToShmem, Parse, ToCss)]
+ #[repr(C)]
+ #[css(bitflags(mixed="over,under,left,right", validate_mixed="Self::validate_and_simplify"))]
+ /// Values for text-emphasis-position:
+ /// <https://drafts.csswg.org/css-text-decor/#text-emphasis-position-property>
+ pub struct TextEmphasisPosition: u8 {
+ /// Draws marks to the right of the text in vertical writing mode.
+ const OVER = 1 << 0;
+ /// Draw marks under the text in horizontal writing mode.
+ const UNDER = 1 << 1;
+ /// Draw marks to the left of the text in vertical writing mode.
+ const LEFT = 1 << 2;
+ /// Draws marks to the right of the text in vertical writing mode.
+ const RIGHT = 1 << 3;
+ }
+}
+
+impl TextEmphasisPosition {
+ fn validate_and_simplify(&mut self) -> bool {
+ if self.intersects(Self::OVER) == self.intersects(Self::UNDER) {
+ return false;
+ }
+
+ if self.intersects(Self::LEFT) {
+ return !self.intersects(Self::RIGHT);
+ }
+
+ self.remove(Self::RIGHT); // Right is the default
+ true
+ }
+}
+
+/// Values for the `word-break` property.
+#[repr(u8)]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum WordBreak {
+ Normal,
+ BreakAll,
+ KeepAll,
+ /// The break-word value, needed for compat.
+ ///
+ /// Specifying `word-break: break-word` makes `overflow-wrap` behave as
+ /// `anywhere`, and `word-break` behave like `normal`.
+ #[cfg(feature = "gecko")]
+ BreakWord,
+}
+
+/// Values for the `text-justify` CSS property.
+#[repr(u8)]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum TextJustify {
+ Auto,
+ None,
+ InterWord,
+ // See https://drafts.csswg.org/css-text-3/#valdef-text-justify-distribute
+ // and https://github.com/w3c/csswg-drafts/issues/6156 for the alias.
+ #[parse(aliases = "distribute")]
+ InterCharacter,
+}
+
+/// Values for the `-moz-control-character-visibility` CSS property.
+#[repr(u8)]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum MozControlCharacterVisibility {
+ Hidden,
+ Visible,
+}
+
+impl Default for MozControlCharacterVisibility {
+ fn default() -> Self {
+ if static_prefs::pref!("layout.css.control-characters.visible") {
+ Self::Visible
+ } else {
+ Self::Hidden
+ }
+ }
+}
+
+/// Values for the `line-break` property.
+#[repr(u8)]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum LineBreak {
+ Auto,
+ Loose,
+ Normal,
+ Strict,
+ Anywhere,
+}
+
+/// Values for the `overflow-wrap` property.
+#[repr(u8)]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum OverflowWrap {
+ Normal,
+ BreakWord,
+ Anywhere,
+}
+
+/// Implements text-decoration-skip-ink which takes the keywords auto | none | all
+///
+/// https://drafts.csswg.org/css-text-decor-4/#text-decoration-skip-ink-property
+#[repr(u8)]
+#[cfg_attr(feature = "servo", derive(Deserialize, Serialize))]
+#[derive(
+ Clone,
+ Copy,
+ Debug,
+ Eq,
+ MallocSizeOf,
+ Parse,
+ PartialEq,
+ SpecifiedValueInfo,
+ ToComputedValue,
+ ToCss,
+ ToResolvedValue,
+ ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum TextDecorationSkipInk {
+ Auto,
+ None,
+ All,
+}
+
+/// Implements type for `text-decoration-thickness` property
+pub type TextDecorationLength = GenericTextDecorationLength<LengthPercentage>;
+
+impl TextDecorationLength {
+ /// `Auto` value.
+ #[inline]
+ pub fn auto() -> Self {
+ GenericTextDecorationLength::Auto
+ }
+
+ /// Whether this is the `Auto` value.
+ #[inline]
+ pub fn is_auto(&self) -> bool {
+ matches!(*self, GenericTextDecorationLength::Auto)
+ }
+}
+
+bitflags! {
+ #[derive(MallocSizeOf, SpecifiedValueInfo, ToComputedValue, ToResolvedValue, ToShmem)]
+ #[value_info(other_values = "auto,from-font,under,left,right")]
+ #[repr(C)]
+ /// Specified keyword values for the text-underline-position property.
+ /// (Non-exclusive, but not all combinations are allowed: the spec grammar gives
+ /// `auto | [ from-font | under ] || [ left | right ]`.)
+ /// https://drafts.csswg.org/css-text-decor-4/#text-underline-position-property
+ pub struct TextUnderlinePosition: u8 {
+ /// Use automatic positioning below the alphabetic baseline.
+ const AUTO = 0;
+ /// Use underline position from the first available font.
+ const FROM_FONT = 1 << 0;
+ /// Below the glyph box.
+ const UNDER = 1 << 1;
+ /// In vertical mode, place to the left of the text.
+ const LEFT = 1 << 2;
+ /// In vertical mode, place to the right of the text.
+ const RIGHT = 1 << 3;
+ }
+}
+
+// TODO: This can be derived with some care.
+impl Parse for TextUnderlinePosition {
+ fn parse<'i, 't>(
+ _context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<TextUnderlinePosition, ParseError<'i>> {
+ let mut result = TextUnderlinePosition::empty();
+
+ loop {
+ let location = input.current_source_location();
+ let ident = match input.next() {
+ Ok(&Token::Ident(ref ident)) => ident,
+ Ok(other) => return Err(location.new_unexpected_token_error(other.clone())),
+ Err(..) => break,
+ };
+
+ match_ignore_ascii_case! { ident,
+ "auto" if result.is_empty() => {
+ return Ok(result);
+ },
+ "from-font" if !result.intersects(TextUnderlinePosition::FROM_FONT |
+ TextUnderlinePosition::UNDER) => {
+ result.insert(TextUnderlinePosition::FROM_FONT);
+ },
+ "under" if !result.intersects(TextUnderlinePosition::FROM_FONT |
+ TextUnderlinePosition::UNDER) => {
+ result.insert(TextUnderlinePosition::UNDER);
+ },
+ "left" if !result.intersects(TextUnderlinePosition::LEFT |
+ TextUnderlinePosition::RIGHT) => {
+ result.insert(TextUnderlinePosition::LEFT);
+ },
+ "right" if !result.intersects(TextUnderlinePosition::LEFT |
+ TextUnderlinePosition::RIGHT) => {
+ result.insert(TextUnderlinePosition::RIGHT);
+ },
+ _ => return Err(location.new_custom_error(
+ SelectorParseErrorKind::UnexpectedIdent(ident.clone())
+ )),
+ }
+ }
+
+ if !result.is_empty() {
+ Ok(result)
+ } else {
+ Err(input.new_custom_error(StyleParseErrorKind::UnspecifiedError))
+ }
+ }
+}
+
+impl ToCss for TextUnderlinePosition {
+ fn to_css<W>(&self, dest: &mut CssWriter<W>) -> fmt::Result
+ where
+ W: Write,
+ {
+ if self.is_empty() {
+ return dest.write_str("auto");
+ }
+
+ let mut writer = SequenceWriter::new(dest, " ");
+ let mut any = false;
+
+ macro_rules! maybe_write {
+ ($ident:ident => $str:expr) => {
+ if self.contains(TextUnderlinePosition::$ident) {
+ any = true;
+ writer.raw_item($str)?;
+ }
+ };
+ }
+
+ maybe_write!(FROM_FONT => "from-font");
+ maybe_write!(UNDER => "under");
+ maybe_write!(LEFT => "left");
+ maybe_write!(RIGHT => "right");
+
+ debug_assert!(any);
+
+ Ok(())
+ }
+}
+
+/// Values for `ruby-position` property
+#[repr(u8)]
+#[derive(
+ Clone, Copy, Debug, Eq, MallocSizeOf, PartialEq, ToComputedValue, ToResolvedValue, ToShmem,
+)]
+#[allow(missing_docs)]
+pub enum RubyPosition {
+ AlternateOver,
+ AlternateUnder,
+ Over,
+ Under,
+}
+
+impl Parse for RubyPosition {
+ fn parse<'i, 't>(
+ _context: &ParserContext,
+ input: &mut Parser<'i, 't>,
+ ) -> Result<RubyPosition, ParseError<'i>> {
+ // Parse alternate before
+ let alternate = input
+ .try_parse(|i| i.expect_ident_matching("alternate"))
+ .is_ok();
+ if alternate && input.is_exhausted() {
+ return Ok(RubyPosition::AlternateOver);
+ }
+ // Parse over / under
+ let over = try_match_ident_ignore_ascii_case! { input,
+ "over" => true,
+ "under" => false,
+ };
+ // Parse alternate after
+ let alternate = alternate ||
+ input
+ .try_parse(|i| i.expect_ident_matching("alternate"))
+ .is_ok();
+
+ Ok(match (over, alternate) {
+ (true, true) => RubyPosition::AlternateOver,
+ (false, true) => RubyPosition::AlternateUnder,
+ (true, false) => RubyPosition::Over,
+ (false, false) => RubyPosition::Under,
+ })
+ }
+}
+
+impl ToCss for RubyPosition {
+ fn to_css<W>(&self, dest: &mut CssWriter<W>) -> fmt::Result
+ where
+ W: Write,
+ {
+ dest.write_str(match self {
+ RubyPosition::AlternateOver => "alternate",
+ RubyPosition::AlternateUnder => "alternate under",
+ RubyPosition::Over => "over",
+ RubyPosition::Under => "under",
+ })
+ }
+}
+
+impl SpecifiedValueInfo for RubyPosition {
+ fn collect_completion_keywords(f: KeywordsCollectFn) {
+ f(&["alternate", "over", "under"])
+ }
+}