1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
//! What a plan says, read as CommonMark: the text a review binds to.
//!
//! A plan's sha used to be taken over its bytes, so a formatter that only
//! added blank lines before nested lists (website-builder's prettier hook,
//! 2026-10-06) made a reviewed plan look edited. [`words`] reads the plan
//! the way a Markdown renderer does and keeps what a reader would see:
//!
//! * `w:` a prose word, after escapes and emphasis are resolved and line
//! breaks are read as spaces, so reflow and `*x*`/`_x_` do not count;
//! * `s:` a code span and `c:` a code block, exactly as parsed: blank lines,
//! indentation inside the code and spaces inside a span all count;
//! * `u:` a link or image destination;
//! * `k:` the kind of a block: a list item, a quote, a heading, a table
//! row or cell, an ordered list's start number, a code block's info
//! string. A plain paragraph or list boundary is a bare `k:`, absorbed by
//! any neighbouring `k:` item, so a tight list and the same list made
//! loose by blank lines read the same, while `- deny` and `deny` do not.
//!
//! Structure decides through the parser, never by a token's place on a line:
//! `a * b` keeps its operator, and `a` followed by a line `* b` is a
//! paragraph and a list item. A line rule cannot tell those apart; that is
//! why this needs a parser (the person's review of the plan, 2026-10-06).
use pulldown_cmark::{CodeBlockKind, Event, Options, Parser, Tag, TagEnd};
/// A block boundary that says nothing about its kind.
const BARE: &str = "k:";
/// The plan's items, one per line. Deterministic for a given parser version,
/// which is why `pulldown-cmark` is pinned exactly.
pub fn words(markdown: &str) -> String {
// Smart punctuation stays off: it would rewrite quotes inside prose.
let options =
Options::ENABLE_TABLES | Options::ENABLE_STRIKETHROUGH | Options::ENABLE_TASKLISTS;
let mut out = Items::default();
let mut code: Option<String> = None;
for event in Parser::new_ext(markdown, options) {
// A code block's text arrives in as many events as the parser likes;
// it is collected into one item so that the split never counts.
if let Some(block) = code.as_mut() {
if let Event::End(TagEnd::CodeBlock) = event {
out.push(format!("c:{block}"));
code = None;
out.kind(BARE);
} else if let Event::Text(t) = event {
block.push_str(&t);
}
// CommonMark gives a code block only text; anything else inside
// one would be a parser change, which the fixture sha catches.
continue;
}
match event {
Event::Start(Tag::CodeBlock(kind)) => {
let info = match kind {
CodeBlockKind::Fenced(info) => info.trim().to_string(),
CodeBlockKind::Indented => String::new(),
};
out.kind(&format!("k:code {info}"));
code = Some(String::new());
}
Event::Text(t) => out.prose.push_str(&t),
Event::Code(c) | Event::InlineMath(c) | Event::DisplayMath(c) => {
out.push(format!("s:{c}"));
}
Event::FootnoteReference(f) => out.push(format!("w:[^{f}]")),
Event::SoftBreak | Event::HardBreak => out.prose.push(' '),
Event::Html(h) | Event::InlineHtml(h) => out.prose.push_str(&h),
Event::Start(Tag::Link { dest_url, .. } | Tag::Image { dest_url, .. }) => {
out.push(format!("u:{dest_url}"));
}
Event::Start(Tag::Emphasis | Tag::Strong)
| Event::End(TagEnd::Emphasis | TagEnd::Strong | TagEnd::Link | TagEnd::Image) => {}
Event::Start(Tag::Strikethrough) | Event::End(TagEnd::Strikethrough) => {
out.prose.push_str(" ~~ ");
}
Event::TaskListMarker(done) => out.prose.push_str(if done { " [x] " } else { " [ ] " }),
Event::Start(Tag::List(Some(start))) => out.kind(&format!("k:list {start}")),
Event::Start(Tag::Item) => out.kind("k:item"),
Event::Start(Tag::BlockQuote(_)) => out.kind("k:quote"),
Event::Start(Tag::Heading { .. }) => out.kind("k:heading"),
Event::Start(Tag::TableRow | Tag::TableHead) => out.kind("k:row"),
Event::Start(Tag::TableCell) => out.kind("k:cell"),
// Every other block, and every end: a boundary of no kind. A tag
// a parser bump adds lands here too, as a boundary, never as
// nothing; the `Event` match itself is exhaustive.
Event::Start(_) | Event::End(_) | Event::Rule => out.kind(BARE),
}
}
out.flush();
out.items.join("\n")
}
#[derive(Default)]
struct Items {
items: Vec<String>,
/// Prose not yet split into words: a word may span several text events
/// (`a\_b` arrives as `a`, `_b`).
prose: String,
}
impl Items {
fn flush(&mut self) {
for w in self.prose.split_whitespace() {
self.items.push(format!("w:{w}"));
}
self.prose.clear();
}
fn push(&mut self, item: String) {
self.flush();
self.items.push(item);
}
/// A block boundary. A bare one next to any other boundary adds nothing,
/// and a kind replaces a bare one before it.
fn kind(&mut self, kind: &str) {
self.flush();
let last_is_kind = self.items.last().is_some_and(|s| s.starts_with(BARE));
if kind == BARE {
if !last_is_kind {
self.items.push(BARE.to_string());
}
return;
}
if self.items.last().is_some_and(|s| s == BARE) {
self.items.pop();
}
self.items.push(kind.to_string());
}
}
#[cfg(test)]
mod tests {
use super::words;
/// Pairs a formatter produces, which must read the same.
const SAME: &[(&str, &str, &str)] = &[
(
"blank lines",
"# T\nOne.\nTwo.\n",
"# T\n\nOne.\nTwo.\n\n\n",
),
(
"blank line before a nested list",
"- a\n - b\n",
"- a\n\n - b\n",
),
(
"loose nested list",
"- a\n - b\n- c\n",
"- a\n\n - b\n\n- c\n",
),
("reflow", "one two\nthree\n", "one\ntwo three\n"),
(
"table padding",
"|a|b|\n|-|-|\n|1|2|\n",
"| a | b |\n| --- | --- |\n| 1 | 2 |\n",
),
("emphasis style", "x *y* z\n", "x _y_ z\n"),
("quoted emphasis", "*\"x\"*\n", "_\"x\"_\n"),
("strong style", "**x**\n", "__x__\n"),
("escape", "a\\_b\n", "a_b\n"),
("list marker", "* a\n* b\n", "- a\n- b\n"),
(
"fence re-indented with its item",
"- x\n\n ```py\n if a:\n go()\n ```\n",
"- x\n\n ```py\n if a:\n go()\n ```\n",
),
];
/// Edits that change what the plan says, which must read differently.
const DIFFERENT: &[(&str, &str, &str)] = &[
("a word", "Body.\n", "Body!\n"),
(
"operator before a break",
"Calculate a * b\n",
"Calculate a\n* b\n",
),
(
"list item vs words",
"Calculate a b\n",
"Calculate a\n* b\n",
),
("plus before a break", "a + b\n", "a\n+ b\n"),
("quote before a break", "x > y\n", "x\n> y\n"),
("operator", "a > b\n", "a >= b\n"),
("operator dropped", "a * b\n", "a b\n"),
("identifier", "MAX_BLOBS\n", "MAXBLOBS\n"),
("leading underscore", "_private x\n", "private x\n"),
("spaces in a span", "`\"a b\"`\n", "`\"a b\"`\n"),
("tab in a span", "`a\tb`\n", "`a\t\tb`\n"),
(
"blank line in a string literal",
"```py\ns = \"\"\"a\n\nb\"\"\"\n```\n",
"```py\ns = \"\"\"a\nb\"\"\"\n```\n",
),
(
"indent inside a fence",
"```py\nif a:\n go()\n```\n",
"```py\nif a:\n go()\n```\n",
),
("fence language", "```py\nx\n```\n", "```sh\nx\n```\n"),
("number", "5 ms\n", "50 ms\n"),
("swapped words", "allow then deny\n", "deny then allow\n"),
("link destination", "[x](a.md)\n", "[x](b.md)\n"),
("paragraph vs item", "x\n\n- deny\n", "x\n\ndeny\n"),
("paragraph vs quote", "> deny\n", "deny\n"),
("ordered start", "5. a\n", "50. a\n"),
("span vs word", "`deny`\n", "deny\n"),
(
"table rows",
"|a|b|\n|-|-|\n|1|2|\n",
"|a|b|\n|-|-|\n|1|\n|2|\n",
),
];
#[test]
fn formatting_alone_reads_the_same() {
for (why, a, b) in SAME {
assert_eq!(words(a), words(b), "{why}");
}
}
#[test]
fn a_change_of_meaning_reads_differently() {
for (why, a, b) in DIFFERENT {
assert_ne!(words(a), words(b), "{why}");
}
}
}