Yehowshua/ZIm
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

added parsing support for <|S| |E|> spans

Browse source
This commit is contained in:
KCaverly 2023-10-30 13:32:47 -04:00
parent 67e590202a
commit 08e9b2e848
2 changed files with 96 additions and 32 deletions
Download patch file Download diff file Expand all files Collapse all files

110
crates/assistant/src/codegen.rs
View file

@ -117,7 +117,7 @@ impl Codegen {
let (mut hunks_tx, mut hunks_rx) = mpsc::channel(1);
let diff = cx.background().spawn(async move {
let chunks = strip_markdown_codeblock(response.await?);
let chunks = strip_invalid_spans_from_codeblock(response.await?);
futures::pin_mut!(chunks);
let mut diff = StreamingDiff::new(selected_text.to_string());
@ -278,12 +278,13 @@ impl Codegen {
}
}
fn strip_markdown_codeblock(
fn strip_invalid_spans_from_codeblock(
stream: impl Stream<Item = Result<String>>,
) -> impl Stream<Item = Result<String>> {
let mut first_line = true;
let mut buffer = String::new();
let mut starts_with_fenced_code_block = false;
let mut starts_with_markdown_codeblock = false;
let mut includes_start_or_end_span = false;
stream.filter_map(move |chunk| {
let chunk = match chunk {
Ok(chunk) => chunk,
@ -291,11 +292,31 @@ fn strip_markdown_codeblock(
};
buffer.push_str(&chunk);
if buffer.len() > "<|S|".len() && buffer.starts_with("<|S|") {
includes_start_or_end_span = true;
buffer = buffer
.strip_prefix("<|S|>")
.or_else(|| buffer.strip_prefix("<|S|"))
.unwrap_or(&buffer)
.to_string();
} else if buffer.ends_with("|E|>") {
includes_start_or_end_span = true;
} else if buffer.starts_with("<|")
|| buffer.starts_with("<|S")
|| buffer.starts_with("<|S|")
|| buffer.ends_with("|")
|| buffer.ends_with("|E")
|| buffer.ends_with("|E|")
{
return future::ready(None);
}
if first_line {
if buffer == "" || buffer == "`" || buffer == "``" {
return future::ready(None);
} else if buffer.starts_with("```") {
starts_with_fenced_code_block = true;
starts_with_markdown_codeblock = true;
if let Some(newline_ix) = buffer.find('\n') {
buffer.replace_range(..newline_ix + 1, "");
first_line = false;
@ -305,16 +326,26 @@ fn strip_markdown_codeblock(
}
}
let text = if starts_with_fenced_code_block {
buffer
let mut text = buffer.to_string();
if starts_with_markdown_codeblock {
text = text
.strip_suffix("\n```\n")
.or_else(|| buffer.strip_suffix("\n```"))
.or_else(|| buffer.strip_suffix("\n``"))
.or_else(|| buffer.strip_suffix("\n`"))
.or_else(|| buffer.strip_suffix('\n'))
.unwrap_or(&buffer)
} else {
&buffer
.or_else(|| text.strip_suffix("\n```"))
.or_else(|| text.strip_suffix("\n``"))
.or_else(|| text.strip_suffix("\n`"))
.or_else(|| text.strip_suffix('\n'))
.unwrap_or(&text)
.to_string();
}
if includes_start_or_end_span {
text = text
.strip_suffix("|E|>")
.or_else(|| text.strip_suffix("E|>"))
.or_else(|| text.strip_prefix("|>"))
.or_else(|| text.strip_prefix(">"))
.unwrap_or(&text)
.to_string();
};
if text.contains('\n') {
@ -327,6 +358,7 @@ fn strip_markdown_codeblock(
} else {
Some(Ok(buffer.clone()))
};
buffer = remainder;
future::ready(result)
})
@ -537,50 +569,82 @@ mod tests {
}
#[gpui::test]
async fn test_strip_markdown_codeblock() {
async fn test_strip_invalid_spans_from_codeblock() {
assert_eq!(
strip_markdown_codeblock(chunks("Lorem ipsum dolor", 2))
strip_invalid_spans_from_codeblock(chunks("Lorem ipsum dolor", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum dolor"
);
assert_eq!(
strip_markdown_codeblock(chunks("```\nLorem ipsum dolor", 2))
strip_invalid_spans_from_codeblock(chunks("```\nLorem ipsum dolor", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum dolor"
);
assert_eq!(
strip_markdown_codeblock(chunks("```\nLorem ipsum dolor\n```", 2))
strip_invalid_spans_from_codeblock(chunks("```\nLorem ipsum dolor\n```", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum dolor"
);
assert_eq!(
strip_markdown_codeblock(chunks("```\nLorem ipsum dolor\n```\n", 2))
strip_invalid_spans_from_codeblock(chunks("```\nLorem ipsum dolor\n```\n", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum dolor"
);
assert_eq!(
strip_markdown_codeblock(chunks("```html\n```js\nLorem ipsum dolor\n```\n```", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
strip_invalid_spans_from_codeblock(chunks(
"```html\n```js\nLorem ipsum dolor\n```\n```",
2
))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"```js\nLorem ipsum dolor\n```"
);
assert_eq!(
strip_markdown_codeblock(chunks("``\nLorem ipsum dolor\n```", 2))
strip_invalid_spans_from_codeblock(chunks("``\nLorem ipsum dolor\n```", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"``\nLorem ipsum dolor\n```"
);
assert_eq!(
strip_invalid_spans_from_codeblock(chunks("<|S|Lorem ipsum|E|>", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum"
);
assert_eq!(
strip_invalid_spans_from_codeblock(chunks("<|S|>Lorem ipsum", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum"
);
assert_eq!(
strip_invalid_spans_from_codeblock(chunks("```\n<|S|>Lorem ipsum\n```", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum"
);
assert_eq!(
strip_invalid_spans_from_codeblock(chunks("```\n<|S|Lorem ipsum|E|>\n```", 2))
.map(|chunk| chunk.unwrap())
.collect::<String>()
.await,
"Lorem ipsum"
);
fn chunks(text: &str, size: usize) -> impl Stream<Item = Result<String>> {
stream::iter(
text.chars()