diff options
author | Titus Wormer <tituswormer@gmail.com> | 2022-06-15 18:17:01 +0200 |
---|---|---|
committer | Titus Wormer <tituswormer@gmail.com> | 2022-06-15 18:17:01 +0200 |
commit | acc35758778bfda5cb01951533868eb8baa2e2d2 (patch) | |
tree | 82ded1c1f88d04a1c37e40822ad7e5a6bba8a717 /src/compiler.rs | |
parent | 75dcb48f78a8a798fde525d2d39e20cffec48e50 (diff) | |
download | markdown-rs-acc35758778bfda5cb01951533868eb8baa2e2d2.tar.gz markdown-rs-acc35758778bfda5cb01951533868eb8baa2e2d2.tar.bz2 markdown-rs-acc35758778bfda5cb01951533868eb8baa2e2d2.zip |
Add code (text)
Diffstat (limited to '')
-rw-r--r-- | src/compiler.rs | 42 |
1 files changed, 35 insertions, 7 deletions
diff --git a/src/compiler.rs b/src/compiler.rs index 6f0215c..d3d935b 100644 --- a/src/compiler.rs +++ b/src/compiler.rs @@ -96,8 +96,6 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St // let mut slurp_all_line_endings = false; - println!("events: {:#?}", events); - while index < events.len() { let event = &events[index]; let token_type = &event.token_type; @@ -133,6 +131,10 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St ignore_encode = true; } } + TokenType::CodeText => { + buf_tail_mut(buffers).push("<code>".to_string()); + buffer(buffers); + } TokenType::Content | TokenType::AtxHeading | TokenType::AtxHeadingSequence @@ -152,6 +154,9 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St | TokenType::CodeFencedFence | TokenType::CodeFencedFenceSequence | TokenType::CodeFencedFenceWhitespace + | TokenType::CodeTextSequence + | TokenType::CodeTextData + | TokenType::CodeTextLineEnding | TokenType::Data | TokenType::CharacterEscape | TokenType::CharacterEscapeMarker @@ -181,6 +186,7 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St | TokenType::Whitespace | TokenType::CodeFencedFenceSequence | TokenType::CodeFencedFenceWhitespace + | TokenType::CodeTextSequence | TokenType::CharacterEscape | TokenType::CharacterEscapeMarker | TokenType::CharacterReference @@ -264,6 +270,7 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St false, ))); } + // `AtxHeadingWhitespace` is ignored after the opening sequence, // before the closing sequence, and after the closing sequence. // But it is used around intermediate sequences. @@ -290,7 +297,6 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St } } TokenType::AtxHeadingText => { - println!("text: {:?}", atx_heading_buffer); if let Some(ref buf) = atx_heading_buffer { if !buf.is_empty() { buf_tail_mut(buffers).push(encode(buf)); @@ -301,7 +307,6 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St } let slice = encode(&serialize(codes, &from_exit_event(events, index), false)); - println!("slice: {:?}", slice); buf_tail_mut(buffers).push(slice); } TokenType::AtxHeading => { @@ -340,8 +345,6 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St // } else if slurp_one_line_ending { slurp_one_line_ending = false; - // } else if code_text_inside { - // buf_tail_mut(buffers).push(" ".to_string()); } else { buf_tail_mut(buffers).push(encode(&serialize( codes, @@ -378,8 +381,33 @@ pub fn compile(events: &[Event], codes: &[Code], options: &CompileOptions) -> St character_reference_kind = None; } + TokenType::CodeText => { + let result = resume(buffers); + let mut chars = result.chars(); + let mut trim = false; + + if Some(' ') == chars.next() && Some(' ') == chars.next_back() { + let mut next = chars.next(); + while next != None && !trim { + if Some(' ') != next { + trim = true; + } + next = chars.next(); + } + } + + buf_tail_mut(buffers).push(if trim { + result[1..(result.len() - 1)].to_string() + } else { + result + }); + buf_tail_mut(buffers).push("</code>".to_string()); + } + TokenType::CodeTextLineEnding => { + buf_tail_mut(buffers).push(" ".to_string()); + } // This branch below currently acts as the resulting `data` tokens. - TokenType::Data | TokenType::CharacterEscapeValue => { + TokenType::CodeTextData | TokenType::Data | TokenType::CharacterEscapeValue => { // last_was_tag = false; buf_tail_mut(buffers).push(encode(&serialize( codes, |