refactor: move html5lib test to own crate to fix `cargo test`

Previously `cargo test` failed because it ran the test_html5lib
integration test, which depends on the integration-tests feature
(so you always had to run `cargo test` with
`--features integration-tests` or `--all-features`, which was annoying).

This commit moves the integration tests to another crate,
so that the dependency on the feature can be properly defined
in a way so that `cargo test` just works and runs the test.

3 files changed, 405 insertions, 0 deletions

diff --git a/integration_tests/Cargo.toml b/integration_tests/Cargo.toml
new file mode 100644
index 0000000..1e68a0b
--- /dev/null
+++ b/integration_tests/Cargo.toml
@@ -0,0 +1,22 @@
+# The html5lib integration test lives in a separate crate because
+# we want `cargo test` to run these tests despite their dependency
+# on the `integration-tests` feature from the html5tokenizer crate
+# and cargo doesn't support features to be automatically enabled for
+# integration tests in a single crate. (required-features under [[test]]
+# just results in the test being skipped if the feature isn't enabled).
+# See https://github.com/rust-lang/cargo/issues/2911#issuecomment-524652568.
+
+[package]
+name = "integration_tests"
+publish = false
+version = "0.0.0"
+edition = "2021"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dev-dependencies]
+glob = "0.3.1"
+html5tokenizer = { path = "..", features = ["integration-tests"] }
+pretty_assertions = "1.0.0"
+serde = { version = "1.0.130", features = ["derive"] }
+serde_json = "1.0.71"
diff --git a/integration_tests/html5lib-tests b/integration_tests/html5lib-tests
new file mode 160000
+Subproject 6030cb6e40a0cf68ae38bf0001bb85b727b80a2
diff --git a/integration_tests/tests/test_html5lib.rs b/integration_tests/tests/test_html5lib.rs
new file mode 100644
index 0000000..cf95bb6
--- /dev/null
+++ b/integration_tests/tests/test_html5lib.rs
@@ -0,0 +1,383 @@
+use html5tokenizer::{
+    Attribute, Doctype, EndTag, Error, InternalState as State, Reader, StartTag, Token, Tokenizer,
+};
+use pretty_assertions::assert_eq;
+use serde::{de::Error as _, Deserialize};
+use std::{collections::BTreeMap, fs::File, io::BufReader, path::Path};
+
+struct ExpectedOutputTokens(Vec<Token<()>>);
+
+impl<'de> Deserialize<'de> for ExpectedOutputTokens {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        // this macro is a horrible way to define a type that deserializes only from a particular
+        // string. Together with serde(untagged) this gives us really flexible enum tagging with really
+        // terrible error messages.
+        macro_rules! def_const {
+            ($str:expr, $ty:ident) => {
+                #[derive(Deserialize)]
+                enum $ty {
+                    #[serde(rename = $str)]
+                    $ty,
+                }
+            };
+        }
+
+        def_const!("DOCTYPE", DoctypeConst);
+        def_const!("StartTag", StartTagConst);
+        def_const!("EndTag", EndTagConst);
+        def_const!("Comment", CommentConst);
+        def_const!("Character", CharacterConst);
+
+        type Attributes = BTreeMap<String, String>;
+
+        #[derive(Deserialize)]
+        #[serde(untagged)]
+        enum OutputToken {
+            // "DOCTYPE", name, public_id, system_id, correctness
+            Doctype(
+                DoctypeConst,
+                Option<String>,
+                Option<String>,
+                Option<String>,
+                bool,
+            ),
+            // "StartTag", name, attributes, self_closing
+            StartTag(StartTagConst, String, Attributes),
+            StartTag2(StartTagConst, String, Attributes, bool),
+            // "EndTag", name
+            EndTag(EndTagConst, String),
+            // "Comment", data
+            Comment(CommentConst, String),
+            // "Character", data
+            Character(CharacterConst, String),
+        }
+
+        Ok(ExpectedOutputTokens(
+            Vec::deserialize(deserializer)?
+                .into_iter()
+                .map(|output_token| match output_token {
+                    OutputToken::Doctype(
+                        _,
+                        name,
+                        public_identifier,
+                        system_identifier,
+                        correctness,
+                    ) => Token::Doctype(Doctype {
+                        name: name.unwrap_or_default(),
+                        public_identifier,
+                        system_identifier,
+                        force_quirks: !correctness,
+                    }),
+                    OutputToken::StartTag(_, name, attributes) => Token::StartTag(StartTag {
+                        self_closing: false,
+                        name,
+                        attributes: attributes
+                            .into_iter()
+                            .map(|(k, v)| {
+                                (
+                                    k,
+                                    Attribute {
+                                        value: v,
+                                        ..Default::default()
+                                    },
+                                )
+                            })
+                            .collect(),
+                        name_span: (),
+                    }),
+                    OutputToken::StartTag2(_, name, attributes, self_closing) => {
+                        Token::StartTag(StartTag {
+                            self_closing,
+                            name,
+                            attributes: attributes
+                                .into_iter()
+                                .map(|(k, v)| {
+                                    (
+                                        k,
+                                        Attribute {
+                                            value: v,
+                                            ..Default::default()
+                                        },
+                                    )
+                                })
+                                .collect(),
+                            name_span: (),
+                        })
+                    }
+                    OutputToken::EndTag(_, name) => Token::EndTag(EndTag {
+                        name,
+                        name_span: (),
+                    }),
+                    OutputToken::Comment(_, data) => Token::Comment(data),
+                    OutputToken::Character(_, data) => Token::String(data),
+                })
+                .collect::<Vec<Token<()>>>(),
+        ))
+    }
+}
+
+struct InitialState(State);
+
+impl<'de> Deserialize<'de> for InitialState {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        #[derive(Deserialize)]
+        enum RawInitialState {
+            #[serde(rename = "Data state")]
+            Data,
+            #[serde(rename = "PLAINTEXT state")]
+            PlainText,
+            #[serde(rename = "RCDATA state")]
+            RcData,
+            #[serde(rename = "RAWTEXT state")]
+            RawText,
+            #[serde(rename = "Script data state")]
+            ScriptData,
+            #[serde(rename = "CDATA section state")]
+            CdataSection,
+        }
+
+        Ok(Self(match RawInitialState::deserialize(deserializer)? {
+            RawInitialState::Data => State::Data,
+            RawInitialState::PlainText => State::PlainText,
+            RawInitialState::RcData => State::RcData,
+            RawInitialState::RawText => State::RawText,
+            RawInitialState::ScriptData => State::ScriptData,
+            RawInitialState::CdataSection => State::CdataSection,
+        }))
+    }
+}
+
+fn initial_states_default() -> Vec<InitialState> {
+    vec![InitialState(State::Data)]
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+struct Test {
+    description: String,
+    input: String,
+    output: ExpectedOutputTokens,
+    #[serde(default = "initial_states_default")]
+    initial_states: Vec<InitialState>,
+    #[serde(default)]
+    double_escaped: bool,
+    #[serde(default)]
+    last_start_tag: Option<String>,
+    #[serde(default)]
+    errors: Vec<ParseError>,
+}
+
+#[derive(Debug, Eq, PartialEq)]
+struct ParseErrorInner(Error);
+
+impl<'de> Deserialize<'de> for ParseErrorInner {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        let str_err = String::deserialize(deserializer)?;
+        let err: Error = str_err
+            .parse()
+            .map_err(|_| D::Error::custom(&format!("failed to deserialize error: {}", str_err)))?;
+        Ok(ParseErrorInner(err))
+    }
+}
+
+#[derive(Deserialize, Debug, Eq, PartialEq)]
+#[serde(rename_all = "camelCase")]
+struct ParseError {
+    code: ParseErrorInner,
+    // TODO: lineno and column?
+}
+
+#[derive(Deserialize)]
+struct Tests {
+    tests: Vec<Test>,
+}
+
+/// Path to a local checkout of [html5lib-tests], relative to the
+/// directory containing the `Cargo.toml` file of the current crate.
+///
+/// [html5lib-tests]: https://github.com/html5lib/html5lib-tests
+const HTML5LIB_TESTS_PATH: &str = "html5lib-tests";
+
+// FUTURE: it would be nice to assert that HTML5LIB_TESTS_PATH matches the path defined in .gitmodules
+// but this is currently blocked by:
+// * Cargo not setting CARGO_WORKSPACE_DIR (see https://github.com/rust-lang/cargo/issues/3946)
+// * gix-config having more dependencies than I'd want to add for this
+
+#[test]
+fn tokenizer() {
+    // TODO: use a custom test harness with e.g. libtest-mimic
+    let test_dir = format!("{HTML5LIB_TESTS_PATH}/tokenizer");
+
+    let mut test_paths = glob::glob(&format!("{test_dir}/*.test"))
+        .unwrap()
+        .peekable();
+
+    if test_paths.peek().is_none() {
+        panic!(
+            "could not find any .test files in {}, maybe try `git submodule update --init`",
+            test_dir
+        );
+    }
+
+    for test_path in test_paths {
+        let test_path = test_path.unwrap();
+
+        test_tokenizer_file(&test_path);
+    }
+}
+
+fn test_tokenizer_file(path: &Path) {
+    let fname = path.file_name().unwrap().to_str().unwrap();
+
+    if matches!(
+        fname,
+        // We don't implement "Coercing an HTML DOM into an infoset" section
+        "xmlViolation.test" |
+        // Our parser does not operate on bytes, the input isn't valid Rust &str
+        "unicodeCharsProblematic.test"
+    ) {
+        return;
+    }
+
+    let f = File::open(path).unwrap();
+    let bf = BufReader::new(f);
+    let tests: Tests = serde_json::from_reader(bf).unwrap();
+
+    for (i, test) in tests.tests.into_iter().enumerate() {
+        run_test(fname, i, test);
+    }
+}
+
+fn run_test(fname: &str, test_i: usize, mut test: Test) {
+    test.input = if test.double_escaped {
+        unescape(&test.input)
+    } else {
+        test.input
+    };
+
+    test.output = if test.double_escaped {
+        ExpectedOutputTokens(
+            test.output
+                .0
+                .into_iter()
+                .map(|token| match token {
+                    Token::String(x) => Token::String(unescape(&x)),
+                    Token::Comment(x) => Token::Comment(unescape(&x)),
+                    token => token,
+                })
+                .collect(),
+        )
+    } else {
+        ExpectedOutputTokens(test.output.0)
+    };
+
+    for state in &test.initial_states {
+        run_test_inner(
+            fname,
+            test_i,
+            &test,
+            state.0,
+            Tokenizer::new(&test.input),
+            "string",
+        );
+
+        run_test_inner(
+            fname,
+            test_i,
+            &test,
+            state.0,
+            Tokenizer::new(BufReader::new(test.input.as_bytes())),
+            "bufread",
+        );
+    }
+}
+
+fn run_test_inner<R: Reader>(
+    fname: &str,
+    test_i: usize,
+    test: &Test,
+    state: State,
+    mut tokenizer: Tokenizer<R>,
+    tokenizer_info: &str,
+) {
+    println!(
+        "==== FILE {}, TEST {}, STATE {:?}, TOKENIZER {} ====",
+        fname, test_i, state, tokenizer_info,
+    );
+    println!("description: {}", test.description);
+    tokenizer.set_internal_state(state);
+    tokenizer.set_last_start_tag(test.last_start_tag.as_ref().map(String::as_str));
+
+    let mut actual_tokens = Vec::new();
+    let mut actual_errors = Vec::new();
+
+    for token in tokenizer {
+        let token = token.unwrap();
+
+        if let Token::Error { error, .. } = token {
+            actual_errors.push(ParseError {
+                code: ParseErrorInner(error),
+            });
+        } else {
+            actual_tokens.push(token);
+        }
+    }
+
+    assert_eq!(test.output.0, actual_tokens);
+
+    if !matches!(
+        (fname, test_i),
+        // TODO: html5lib-tests bug?
+        ("test3.test", 79)
+    ) {
+        assert_eq!(test.errors, actual_errors);
+    }
+}
+
+/// Implements the escape sequences described in the tokenizer tests of html5lib-tests (and nothing
+/// more)
+fn unescape(data: &str) -> String {
+    let mut stream = data.chars();
+    let mut rv = String::new();
+
+    loop {
+        match stream.next() {
+            Some('\\') => (),
+            Some(x) => {
+                rv.push(x);
+                continue;
+            }
+            None => break,
+        }
+
+        match stream.next() {
+            Some('u') => (),
+            x => panic!("unexpected escape: {:?}", x),
+        }
+
+        let orig_len = rv.len();
+
+        for _ in 0..4 {
+            rv.push(match stream.next() {
+                Some(x) => x,
+                None => panic!("unexpected eof after \\u"),
+            });
+        }
+
+        let c = u32::from_str_radix(&rv[orig_len..], 16).expect("failed to parse as hex");
+        let c = char::from_u32(c).expect("bad character");
+        rv.truncate(orig_len);
+        rv.push(c);
+    }
+
+    rv
+}