Skip to content

Latest commit

 

History

History
224 lines (145 loc) · 7.94 KB

identifier-escapes.md

File metadata and controls

224 lines (145 loc) · 7.94 KB

Escapes in Identifiers: Draft 3

(Issue, Changelog)

This proposal adjusts how escaped code points are handled by Sass outside of string contexts. It's intended to bring Sass's semantics more in line with how CSS handles escapes.

Table of Contents

Background

This section is non-normative.

At time of writing, while Sass recognizes escaped code points in identifiers and other names, it doesn't resolve them into the code points they represent. This means that, for example, Sass considers the selector .\!foo and the selector .\21 foo to be distinct. This is contrary to the CSS Syntax Level 3, which says that the value of an escaped code point should be included in the name rather than the syntax of the escape.

However, the current behavior works well for unquoted strings in SassScript. These strings need to distinguish between escaped code points and the literal characters they represent, because unquoted strings can represent more than just identifiers. For example, the SassScript expression unquote("@x") should be rendered to CSS as @x, whereas the expression \@x should be rendered as \@x (or \40 x). Any proposal for parsing escapes properly should preserve this distinction.

Summary

This section is non-normative.

As identifiers are parsed, escapes will be normalized into a canonical form. This preserves the benefits of the existing behavior, where \@x and unquote("@x") are different SassScript expressions, while ensuring that .\!foo and .\21 foo are considered the same selector.

The canonical form of a code point is:

  • The literal code point if it's a valid identifier character; or

  • a backslash followed by the code point's lowercase hex code followed by a space if it's not printable or a newline; or

  • a backslash followed by the code point's lowercase hex code followed by a space if it's a digit at the beginning of an identifier; or

  • a backslash followed by the literal code point.

For example, in SassScript:

  • ax, \61x, and \61 x all parse to the unquoted string ax;
  • \7f x, \7fx, and \7Fx all parse to the unquoted string \7f x; and
  • \31 x and \31x parse to the unquoted string \31 x; and
  • \@x, \40x, and \0040x all parse to the unquoted string \@x.

Compatibility

The proposed change affects existing observable behavior. It's theoretically possible that an existing user is, for example, using \@x and \40 x as distinct map keys; or that they're relying on length(\40 x) returning 5 rather than 3. However, the chances of this seem extremely low, and it would be very difficult to produce actionable deprecation warnings without compromising efficiency.

Given that, and given that this is arguably a bug fix (in that we're moving towards interpreting plain CSS text following the CSS spec, which we hadn't been before), I propose that we don't consider this a breaking change and release it with only a minor version bump.

Syntax

This proposal defines a new algorithm for consuming an identifier and an interpolated identifier. These are intended to replace the existing algorithms.

Other than modifying the way escaped code points are handled, these algorithm are designed to accurately capture the current behavior of all Sass implementations.

Consuming an Identifier

This algorithm consumes input from a stream of code points and returns a string.

This production has the same grammar as <ident-token>.

  • Let string be an empty string.

  • If the stream starts with --, consume it and append it to string.

  • Otherwise:

    • If the stream starts with -, consume it and append it to string.

    • If the stream starts with \, consume an escaped code point with the start flag set and append it to string.

    • Otherwise, if the stream starts with a name-start code point, consume it and append it to string.

    • Otherwise, throw an error.

  • Consume a name and append it to string.

  • Return string.

Consuming an Interpolated Identifier

This algorithm consumes input from a stream of code points and returns a sequence of strings and/or expressions.

The grammar for this production is:

InterpolatedIdentifier ::= (<ident-token> | '-'? Interpolation) (Name | Interpolation)*

No whitespace is allowed between components of an InterpolatedIdentifier.

  • Let components be an empty list of strings and/or expressions.

  • If the input starts with -#{, consume a single code point and add "-" to components.

  • If the input starts with #{, consume an interpolation and add its expression to components.

  • Otherwise, consume an identifier and add its string to components.

  • While the input starts with #{, a name code point, or \:

    • If the input starts with #{, consume an interpolation and add its expression to components.

    • Otherwise, consume a name and add its string to components.

  • Return components.

Consuming a Name

This algorithm consumes input from a stream of code points and returns a string.

The grammar for this production is:

Name ::= (name code point | escape)+

Consuming an Escaped Code Point

This algorithm consumes input from a stream of code points. It takes an optional boolean flag, start, which indicates whether it's at the beginning of an identifier and defaults to false. It returns a string.

This production has the same grammar as escape in CSS Syntax Level 3.

  • If the stream doesn't start with a valid escape, throw an error.

  • Let codepoint be the result of consuming an escaped code point.

  • Let character be the string containing only codepoint.

  • If codepoint is a name-start code point, return character.

  • Otherwise, if codepoint is a name code point and the start flag is not set, return character.

  • Otherwise, if codepoint is a non-printable code point, U+000A LINE FEED, U+000D CARRIAGE RETURN, or U+000C FORM FEED; or if codepoint is a digit and the start flag is set:

    • Let code be the lowercase hexadecimal representation of codepoint, with no leading 0s.

    • Return "\" + code + " ".

  • Otherwise, return "\" + character.