diff options
Diffstat (limited to 'helix-term/src/ui')
-rw-r--r-- | helix-term/src/ui/markdown.rs | 77 |
1 files changed, 61 insertions, 16 deletions
diff --git a/helix-term/src/ui/markdown.rs b/helix-term/src/ui/markdown.rs index 45273002..037e2f13 100644 --- a/helix-term/src/ui/markdown.rs +++ b/helix-term/src/ui/markdown.rs @@ -135,6 +135,7 @@ impl Markdown { "markup.heading.5", "markup.heading.6", ]; + const INDENT: &'static str = " "; pub fn new(contents: String, config_loader: Arc<syntax::Loader>) -> Self { Self { @@ -144,8 +145,12 @@ impl Markdown { } fn parse(&self, theme: Option<&Theme>) -> tui::text::Text<'_> { - // // also 2021-03-04T16:33:58.553 helix_lsp::transport [INFO] <- {"contents":{"kind":"markdown","value":"\n```rust\ncore::num\n```\n\n```rust\npub const fn saturating_sub(self, rhs:Self) ->Self\n```\n\n---\n\n```rust\n```"},"range":{"end":{"character":61,"line":101},"start":{"character":47,"line":101}}} - // let text = "\n```rust\ncore::iter::traits::iterator::Iterator\n```\n\n```rust\nfn collect<B: FromIterator<Self::Item>>(self) -> B\nwhere\n Self: Sized,\n```\n\n---\n\nTransforms an iterator into a collection.\n\n`collect()` can take anything iterable, and turn it into a relevant\ncollection. This is one of the more powerful methods in the standard\nlibrary, used in a variety of contexts.\n\nThe most basic pattern in which `collect()` is used is to turn one\ncollection into another. You take a collection, call [`iter`](https://doc.rust-lang.org/nightly/core/iter/traits/iterator/trait.Iterator.html) on it,\ndo a bunch of transformations, and then `collect()` at the end.\n\n`collect()` can also create instances of types that are not typical\ncollections. For example, a [`String`](https://doc.rust-lang.org/nightly/core/iter/std/string/struct.String.html) can be built from [`char`](type@char)s,\nand an iterator of [`Result<T, E>`](https://doc.rust-lang.org/nightly/core/result/enum.Result.html) items can be collected\ninto `Result<Collection<T>, E>`. See the examples below for more.\n\nBecause `collect()` is so general, it can cause problems with type\ninference. As such, `collect()` is one of the few times you'll see\nthe syntax affectionately known as the 'turbofish': `::<>`. This\nhelps the inference algorithm understand specifically which collection\nyou're trying to collect into.\n\n# Examples\n\nBasic usage:\n\n```rust\nlet a = [1, 2, 3];\n\nlet doubled: Vec<i32> = a.iter()\n .map(|&x| x * 2)\n .collect();\n\nassert_eq!(vec![2, 4, 6], doubled);\n```\n\nNote that we needed the `: Vec<i32>` on the left-hand side. This is because\nwe could collect into, for example, a [`VecDeque<T>`](https://doc.rust-lang.org/nightly/core/iter/std/collections/struct.VecDeque.html) instead:\n\n```rust\nuse std::collections::VecDeque;\n\nlet a = [1, 2, 3];\n\nlet doubled: VecDeque<i32> = a.iter().map(|&x| x * 2).collect();\n\nassert_eq!(2, doubled[0]);\nassert_eq!(4, doubled[1]);\nassert_eq!(6, doubled[2]);\n```\n\nUsing the 'turbofish' instead of annotating `doubled`:\n\n```rust\nlet a = [1, 2, 3];\n\nlet doubled = a.iter().map(|x| x * 2).collect::<Vec<i32>>();\n\nassert_eq!(vec![2, 4, 6], doubled);\n```\n\nBecause `collect()` only cares about what you're collecting into, you can\nstill use a partial type hint, `_`, with the turbofish:\n\n```rust\nlet a = [1, 2, 3];\n\nlet doubled = a.iter().map(|x| x * 2).collect::<Vec<_>>();\n\nassert_eq!(vec![2, 4, 6], doubled);\n```\n\nUsing `collect()` to make a [`String`](https://doc.rust-lang.org/nightly/core/iter/std/string/struct.String.html):\n\n```rust\nlet chars = ['g', 'd', 'k', 'k', 'n'];\n\nlet hello: String = chars.iter()\n .map(|&x| x as u8)\n .map(|x| (x + 1) as char)\n .collect();\n\nassert_eq!(\"hello\", hello);\n```\n\nIf you have a list of [`Result<T, E>`](https://doc.rust-lang.org/nightly/core/result/enum.Result.html)s, you can use `collect()` to\nsee if any of them failed:\n\n```rust\nlet results = [Ok(1), Err(\"nope\"), Ok(3), Err(\"bad\")];\n\nlet result: Result<Vec<_>, &str> = results.iter().cloned().collect();\n\n// gives us the first error\nassert_eq!(Err(\"nope\"), result);\n\nlet results = [Ok(1), Ok(3)];\n\nlet result: Result<Vec<_>, &str> = results.iter().cloned().collect();\n\n// gives us the list of answers\nassert_eq!(Ok(vec![1, 3]), result);\n```"; + fn push_line<'a>(spans: &mut Vec<Span<'a>>, lines: &mut Vec<Spans<'a>>) { + let spans = std::mem::take(spans); + if !spans.is_empty() { + lines.push(Spans::from(spans)); + } + } let mut options = Options::empty(); options.insert(Options::ENABLE_STRIKETHROUGH); @@ -155,6 +160,15 @@ impl Markdown { let mut tags = Vec::new(); let mut spans = Vec::new(); let mut lines = Vec::new(); + let mut list_stack = Vec::new(); + + let get_indent = |level: usize| { + if level < 1 { + String::new() + } else { + Self::INDENT.repeat(level - 1) + } + }; let get_theme = |key: &str| -> Style { theme.map(|t| t.get(key)).unwrap_or_default() }; let text_style = get_theme(Self::TEXT_STYLE); @@ -164,22 +178,53 @@ impl Markdown { .map(|key| get_theme(key)) .collect(); - let mut list_stack = Vec::new(); - for event in parser { match event { - Event::Start(Tag::List(list)) => list_stack.push(list), + Event::Start(Tag::List(list)) => { + // if the list stack is not empty this is a sub list, in that + // case we need to push the current line before proceeding + if !list_stack.is_empty() { + push_line(&mut spans, &mut lines); + } + + list_stack.push(list); + } Event::End(Tag::List(_)) => { list_stack.pop(); - // whenever list closes, new line - lines.push(Spans::default()); + + // whenever top-level list closes, empty line + if list_stack.is_empty() { + lines.push(Spans::default()); + } } Event::Start(Tag::Item) => { + if list_stack.is_empty() { + log::warn!("markdown parsing error, list item without list"); + } + tags.push(Tag::Item); - spans.push(Span::from("- ")); + + // get the approriate bullet for the current list + let bullet = list_stack + .last() + .unwrap_or(&None) // use the '- ' bullet in case the list stack would be empty + .map_or(String::from("- "), |number| format!("{}. ", number)); + + // increment the current list number if there is one + if let Some(v) = list_stack.last_mut().unwrap_or(&mut None).as_mut() { + *v += 1; + } + + let prefix = get_indent(list_stack.len()) + bullet.as_str(); + spans.push(Span::from(prefix)); } Event::Start(tag) => { tags.push(tag); + if spans.is_empty() && !list_stack.is_empty() { + // TODO: could push indent + 2 or 3 spaces to align with + // the rest of the list. + spans.push(Span::from(get_indent(list_stack.len()))); + } } Event::End(tag) => { tags.pop(); @@ -188,15 +233,12 @@ impl Markdown { | Tag::Paragraph | Tag::CodeBlock(CodeBlockKind::Fenced(_)) | Tag::Item => { - let spans = std::mem::take(&mut spans); - if !spans.is_empty() { - lines.push(Spans::from(spans)); - } + push_line(&mut spans, &mut lines); } _ => (), } - // whenever heading, code block or paragraph closes, new line + // whenever heading, code block or paragraph closes, empty line match tag { Tag::Heading(_, _, _) | Tag::Paragraph @@ -237,9 +279,12 @@ impl Markdown { spans.push(Span::styled(text, code_style)); } Event::SoftBreak | Event::HardBreak => { - // let spans = std::mem::replace(&mut spans, Vec::new()); - // lines.push(Spans::from(spans)); - spans.push(Span::raw(" ")); + push_line(&mut spans, &mut lines); + if !list_stack.is_empty() { + // TODO: could push indent + 2 or 3 spaces to align with + // the rest of the list. + spans.push(Span::from(get_indent(list_stack.len()))); + } } Event::Rule => { lines.push(Spans::from(Span::styled("---", code_style))); |