JavaScript正則中g標誌

緣起

有一天在思否社群看到有個問題，大致描述如下

const list = ['a', 'b', '-', 'c', 'd'];
const reg = /[a-z]/g;
const letters = list.filter(i => reg.test(i));

// letters === ['a', 'c'];
// 如果正則不使用`g`標誌可以得到所有的字母
// 為什麼加入`g`之後就不可以了

對問題而言，遍歷中的i就是一個字元，不需要用到g。

但是就我對正則的理解（過於淺薄）感覺上有沒有g（只是全域性搜尋，不會匹配到就停下來）應該不影響，激發了我的好奇心。

上面題的建議寫法如下

const reg = /[a-z]/g;
reg.test('a'); // => true
reg.test('a'); // => false
reg.test('a'); // => true
reg.test('a'); // => false
reg.test('a'); // => true

解密過程

首先可以確定的表現一定是g導致的

搜尋引擎

開啟 MDN 仔細檢視g標誌的作用，得到結論和我的理解無二。

我猜想應該就是g可能啟用了某種快取，又因為reg相對過濾器是全域性變數，我將程式碼改為：

const list = ['a', 'b', '-', 'c', 'd'];
const letters = list.filter(i => /[a-z]/g.test(i));

// letters === ['a', 'b', 'c', 'd'];

將正則宣告到每一次遍歷，得到結論就是正確的，驗證了我的猜想。也得到了，快取就是正則中的某個地方

下面我找到對應的原始碼來檢視問題的原因

原始碼層面

由於最近在看 Rust，所以使用 Rust 編寫的原始碼檢視 https://github/boa-dev/boa

開啟專案後，點選.進入 vscode 模式，command+p 搜尋 regexp 關鍵詞

進入test.rs檔案，command+f 搜尋/g可以找到在 90 行有個last_index()的測試

#[test]
fn last_index() {
    let mut context = Context::default();
    let init = r#"
        var regex = /[0-9]+(\.[0-9]+)?/g;
        "#;
    // forward 的作用：更改 context，並返回結果的字串。
    eprintln!("{}", forward(&mut context, init));
    assert_eq!(forward(&mut context, "regex.lastIndex"), "0");
    assert_eq!(forward(&mut context, "regex.test('1.0foo')"), "true");
    assert_eq!(forward(&mut context, "regex.lastIndex"), "3");
    assert_eq!(forward(&mut context, "regex.test('1.0foo')"), "false");
    assert_eq!(forward(&mut context, "regex.lastIndex"), "0");
}

看到了有lastIndex關鍵字，這裡再已經大致猜到問題的原因了，g 標誌存在匹配後的最後一個下標，導致出現問題。

我們將視線移入到mod.rs檔案中，搜尋test

在 631 行看到了fn test()方法

pub(crate) fn test(
    this: &JsValue,
    args: &[JsValue],
    context: &mut Context,
) -> JsResult<JsValue> {
    // 1. Let R be the this value.
    // 2. If Type(R) is not Object, throw a TypeError exception.
    let this = this.as_object().ok_or_else(|| {
        context
            .construct_type_error("RegExp.prototype.test method called on incompatible value")
    })?;

    // 3. Let string be ? ToString(S).
    let arg_str = args
        .get(0)
        .cloned()
        .unwrap_or_default()
        .to_string(context)?;

    // 4. Let match be ? RegExpExec(R, string).
    let m = Self::abstract_exec(this, arg_str, context)?;

    // 5. If match is not null, return true; else return false.
    if m.is_some() {
        Ok(JsValue::new(true))
    } else {
        Ok(JsValue::new(false))
    }
}

在test()方法中找到了Self::abstract_exec()方法

pub(crate) fn abstract_exec(
    this: &JsObject,
    input: JsString,
    context: &mut Context,
) -> JsResult<Option<JsObject>> {
    // 1. Assert: Type(R) is Object.
    // 2. Assert: Type(S) is String.

    // 3. Let exec be ? Get(R, "exec").
    let exec = this.get("exec", context)?;

    // 4. If IsCallable(exec) is true, then
    if let Some(exec) = exec.as_callable() {
        // a. Let result be ? Call(exec, R, « S »).
        let result = exec.call(&this.clone().into(), &[input.into()], context)?;

        // b. If Type(result) is neither Object nor Null, throw a TypeError exception.
        if !result.is_object() && !result.is_null() {
            return context.throw_type_error("regexp exec returned neither object nor null");
        }

        // c. Return result.
        return Ok(result.as_object().cloned());
    }

    // 5. Perform ? RequireInternalSlot(R, [[RegExpMatcher]]).
    if !this.is_regexp() {
        return context.throw_type_error("RegExpExec called with invalid value");
    }

    // 6. Return ? RegExpBuiltinExec(R, S).
    Self::abstract_builtin_exec(this, &input, context)
}

又在Self::abstract_exec()方法中找到了Self::abstract_builtin_exec()方法

pub(crate) fn abstract_builtin_exec(
    this: &JsObject,
    input: &JsString,
    context: &mut Context,
) -> JsResult<Option<JsObject>> {
    // 1. Assert: R is an initialized RegExp instance.
    let rx = {
        let obj = this.borrow();
        if let Some(rx) = obj.as_regexp() {
            rx.clone()
        } else {
            return context.throw_type_error("RegExpBuiltinExec called with invalid value");
        }
    };

    // 2. Assert: Type(S) is String.

    // 3. Let length be the number of code units in S.
    let length = input.encode_utf16().count();

    // 4. Let lastIndex be ℝ(? ToLength(? Get(R, "lastIndex"))).
    let mut last_index = this.get("lastIndex", context)?.to_length(context)?;

    // 5. Let flags be R.[[OriginalFlags]].
    let flags = &rx.original_flags;

    // 6. If flags contains "g", let global be true; else let global be false.
    let global = flags.contains('g');

    // 7. If flags contains "y", let sticky be true; else let sticky be false.
    let sticky = flags.contains('y');

    // 8. If global is false and sticky is false, set lastIndex to 0.
    if !global && !sticky {
        last_index = 0;
    }

    // 9. Let matcher be R.[[RegExpMatcher]].
    let matcher = &rx.matcher;

    // 10. If flags contains "u", let fullUnicode be true; else let fullUnicode be false.
    let unicode = flags.contains('u');

    // 11. Let matchSucceeded be false.
    // 12. Repeat, while matchSucceeded is false,
    let match_value = loop {
        // a. If lastIndex > length, then
        if last_index > length {
            // i. If global is true or sticky is true, then
            if global || sticky {
                // 1. Perform ? Set(R, "lastIndex", +0?, true).
                this.set("lastIndex", 0, true, context)?;
            }

            // ii. Return null.
            return Ok(None);
        }

        // b. Let r be matcher(S, lastIndex).
        // Check if last_index is a valid utf8 index into input.
        let last_byte_index = match String::from_utf16(
            &input.encode_utf16().take(last_index).collect::<Vec<u16>>(),
        ) {
            Ok(s) => s.len(),
            Err(_) => {
                return context
                    .throw_type_error("Failed to get byte index from utf16 encoded string")
            }
        };
        let r = matcher.find_from(input, last_byte_index).next();

        match r {
            // c. If r is failure, then
            None => {
                // i. If sticky is true, then
                if sticky {
                    // 1. Perform ? Set(R, "lastIndex", +0?, true).
                    this.set("lastIndex", 0, true, context)?;

                    // 2. Return null.
                    return Ok(None);
                }

                // ii. Set lastIndex to AdvanceStringIndex(S, lastIndex, fullUnicode).
                last_index = advance_string_index(input, last_index, unicode);
            }

            Some(m) => {
                // c. If r is failure, then
                #[allow(clippy::if_not_else)]
                if m.start() != last_index {
                    // i. If sticky is true, then
                    if sticky {
                        // 1. Perform ? Set(R, "lastIndex", +0?, true).
                        this.set("lastIndex", 0, true, context)?;

                        // 2. Return null.
                        return Ok(None);
                    }

                    // ii. Set lastIndex to AdvanceStringIndex(S, lastIndex, fullUnicode).
                    last_index = advance_string_index(input, last_index, unicode);
                // d. Else,
                } else {
                    //i. Assert: r is a State.
                    //ii. Set matchSucceeded to true.
                    break m;
                }
            }
        }
    };

    // 13. Let e be r's endIndex value.
    let mut e = match_value.end();

    // 14. If fullUnicode is true, then
    if unicode {
        // e is an index into the Input character list, derived from S, matched by matcher.
        // Let eUTF be the smallest index into S that corresponds to the character at element e of Input.
        // If e is greater than or equal to the number of elements in Input, then eUTF is the number of code units in S.
        // b. Set e to eUTF.
        e = input.split_at(e).0.encode_utf16().count();
    }

    // 15. If global is true or sticky is true, then
    if global || sticky {
        // a. Perform ? Set(R, "lastIndex", ?(e), true).
        this.set("lastIndex", e, true, context)?;
    }

    // 16. Let n be the number of elements in r's captures List. (This is the same value as 22.2.2.1's NcapturingParens.)
    let n = match_value.captures.len();
    // 17. Assert: n < 23^2 - 1.
    debug_assert!(n < 23usize.pow(2) - 1);

    // 18. Let A be ! ArrayCreate(n + 1).
    // 19. Assert: The mathematical value of A's "length" property is n + 1.
    let a = Array::array_create(n + 1, None, context)?;

    // 20. Perform ! CreateDataPropertyOrThrow(A, "index", ?(lastIndex)).
    a.create_data_property_or_throw("index", match_value.start(), context)
        .expect("this CreateDataPropertyOrThrow call must not fail");

    // 21. Perform ! CreateDataPropertyOrThrow(A, "input", S).
    a.create_data_property_or_throw("input", input.clone(), context)
        .expect("this CreateDataPropertyOrThrow call must not fail");

    // 22. Let matchedSubstr be the substring of S from lastIndex to e.
    let matched_substr = if let Some(s) = input.get(match_value.range()) {
        s
    } else {
        ""
    };

    // 23. Perform ! CreateDataPropertyOrThrow(A, "0", matchedSubstr).
    a.create_data_property_or_throw(0, matched_substr, context)
        .expect("this CreateDataPropertyOrThrow call must not fail");

    // 24. If R contains any GroupName, then
    // 25. Else,
    let named_groups = match_value.named_groups();
    let groups = if named_groups.clone().count() > 0 {
        // a. Let groups be ! OrdinaryObjectCreate(null).
        let groups = JsValue::from(JsObject::empty());

        // Perform 27.f here
        // f. If the ith capture of R was defined with a GroupName, then
        // i. Let s be the CapturingGroupName of the corresponding RegExpIdentifierName.
        // ii. Perform ! CreateDataPropertyOrThrow(groups, s, capturedValue).
        for (name, range) in named_groups {
            if let Some(range) = range {
                let value = if let Some(s) = input.get(range.clone()) {
                    s
                } else {
                    ""
                };

                groups
                    .to_object(context)?
                    .create_data_property_or_throw(name, value, context)
                    .expect("this CreateDataPropertyOrThrow call must not fail");
            }
        }
        groups
    } else {
        // a. Let groups be undefined.
        JsValue::undefined()
    };

    // 26. Perform ! CreateDataPropertyOrThrow(A, "groups", groups).
    a.create_data_property_or_throw("groups", groups, context)
        .expect("this CreateDataPropertyOrThrow call must not fail");

    // 27. For each integer i such that i ≥ 1 and i ≤ n, in ascending order, do
    for i in 1..=n {
        // a. Let captureI be ith element of r's captures List.
        let capture = match_value.group(i);

        let captured_value = match capture {
            // b. If captureI is undefined, let capturedValue be undefined.
            None => JsValue::undefined(),
            // c. Else if fullUnicode is true, then
            // d. Else,
            Some(range) => {
                if let Some(s) = input.get(range) {
                    s.into()
                } else {
                    "".into()
                }
            }
        };

        // e. Perform ! CreateDataPropertyOrThrow(A, ! ToString(?(i)), capturedValue).
        a.create_data_property_or_throw(i, captured_value, context)
            .expect("this CreateDataPropertyOrThrow call must not fail");
    }

    // 28. Return A.
    Ok(Some(a))
}

Self::abstract_builtin_exec()方法中存在global以及last_index這樣看來最終執行的方法就是在這裡了，仔細檢視該方法中的程式碼（程式碼寫的很詳細而且每一步都有註釋）

在第 12 步中：

1. lastIndex 超過文字長度且當 global 存在時將 lastIndex 置為 0

2. 獲取匹配到的值（match_value）

   1. 如果未匹配到則置為advance_string_index()方法的返回值
   2. advance_string_index()不在當前問題的考慮範圍 https://tc39.es/ecma262/#sec-...

第 13 步獲取匹配到的值的 endIndex

第 15 步將 lastIndex 置為 endIndex

至此也就整明白了g標誌的含義，在正則的原型鏈中存在一個lastIndex，如果匹配為真時lastIndex不會重置為 0 ，下一次開始時繼承了上次位置，

結論

在問題程式碼中分析

const reg = /[a-z]/g; // 宣告後，lastIndex 為 0
reg.test('a'); // => true；第一次匹配後，lastIndex 為 1
reg.test('a'); // => false；第二次匹配由於 lastIndex 為 1，且字元只有一個，得到 false，將 lastIndex 置為 0
reg.test('a'); // => true；下面依次迴圈前兩次的邏輯
reg.test('a'); // => false；
reg.test('a'); // => true；