Issue with Array.isArray typeguard after upgrading from 4.1.2 to 4.1.3

[mwgamble]

TypeScript Version: 4.2.0-dev.20201211

Search Terms:

re: Array.isArray

Code

interface TestCase<T extends string | number> {
    readonly val1: T | ReadonlyArray<T>;
    readonly val2: ReadonlyArray<T>;
}

interface MultiCaseFixture<T> {
    cases: T[];
}

function subDataFunc(): TestCase<string | number>[] {
    return [
        { val1: "a", val2: ["a", "b", "c"] },
        { val1: 2, val2: [1, 2, 3] },
        { val1: ["a", "z"], val2: ["x", "y", "z"] },
        { val1: [5, 10], val2: [10, 100, 1000] },
    ];
}

function dataFunc<T>(subFunc: () => T[]): MultiCaseFixture<T> {
    return { cases: subFunc() };
}

function testFunc() {
    const fixture = dataFunc<TestCase<string | number>>(subDataFunc);
    fixture.cases.forEach(({ val1, val2 }) => {
        if (Array.isArray(val1)) {
            const reversedVal1 = val1.slice().reverse();
            console.log(reversedVal1);
        } else {
            console.log(val1);
        }
        console.log(val2);
    });
}

testFunc();

Expected behavior:

This code should compile properly without issues. It compiles fine on 4.1.2 and also compiled fine on previous releases going all the way back to at least 3.8.

Actual behavior:

I get this output from tsc with the command-line yarn run tsc test.ts. There is no tsconfig.json file in the folder at the time.

test.ts:27:47 - error TS2339: Property 'reverse' does not exist on type 'string | any[]'.
  Property 'reverse' does not exist on type 'string'.

27             const reversedVal1 = val1.slice().reverse();
                                                 ~~~~~~~


Found 1 error.

Playground Link: https://www.typescriptlang.org/play?ts=4.2.0-dev.20201211#code/JYOwLgpgTgZghgYwgAgCoQM5gMJwxAHlWQgA9IQATDZLKUAc2QB9kQBXAWwCNoA+ZAG8AUMjHIoEOJQD2IADYBPZADc48gIwAuNC2QAlKbIWKAglChxFRPgG5R4ydLlLV6gEw7Dzk+cvXUO2EAX2FhUEhYRBQAWXZ5MGBcfAAxYHJ2SRshBzEEPEwdVABtAF17UOEYdhAERLladm4AETgwOBSahAAKAEoizBwCgjpGPQ4efjKc8QkIMEyQZGLc2aE3TR0AIjgtgBoNz2Wd-eQt7lOthC3S5GC91dnBDe1kdwO1eSPijQP35AAzLd7o9xM9Pq9iicDlsAF43D4eHRQ0iXRSXeHAh5rMEvZEAVgOGgADKVEV9kSSicTidSaVjVuUQmFqrV6ktKG0Ol0bN0ME1OrUdH1kABeAQlUr9ZBxBJJAppDJZQIzRzzRbrfL4DA6fncQU9Xp3Cosrrs5CQLAGkUiWYIORYZAwdILSRi5Cc9oGoiDZKEUYgJisCa8KB8Ph8pqtL1dXr2WbOpUQAB0WswyZgMigAFFEAALbrdcHqX6HO5G8WqnHAGDIbp+KzJ4AYBuKboQ3pG204tb2kCOyQqaD4SgANRL7ohyYw8mASD6ycHw4gfXjPbtDpk8hT8hkDG6S6gI-HmjjoLEwRI8nwVfXyD7GC3O737ZLZ-XoXXD6fyd3+8+7jvuIwRnpUlpgNacZAA

Related Issues:

According to #40463 and #41808, an issue with Array.isArray has already been fixed and is supposedly in 4.1.3.

[typescript-bot]

👋 Hi, I'm the Repro bot. I can help narrow down and track compiler bugs across releases! This comment reflects the current state of the repro in this issue running against the nightly TypeScript. If something changes, I will post a new comment.

---

Issue body code block by @mwgamble

👍 Compiled

Historical Information

Issue body code block by @mwgamble

Version	Reproduction Outputs
4.1.2	❌ Failed: - Property 'reverse' does not exist on type 'string | any[]'. Property 'reverse' does not exist on type 'string'.
3.7.5, 3.8.2, 3.9.2, 4.0.2, Nightly	👍 Compiled

[mwgamble]

Not to sound rude, but Repro Bot is incorrect. Unless there's something really weird about my setup, this issue is not occurring on 4.1.2. It started on 4.1.3.

This is the playground link for 4.1.2, where there are most certainly no errors:

https://www.typescriptlang.org/play?ts=4.1.2#code/JYOwLgpgTgZghgYwgAgCoQM5gMJwxAHlWQgA9IQATDZLKUAc2QB9kQBXAWwCNoA+ZAG8AUMjHIoEOJQD2IADYBPZADc48gIwAuNC2QAlKbIWKAglChxFRPgG5R4ydLlLV6gEw7Dzk+cvXUO2EAX2FhUEhYRBQAWXZ5MGBcfAAxYHJ2SRshBzEEPEwdVABtAF17UOEYdhAERLladm4AETgwOBSahAAKAEoizBwCgjpGPQ4efjKc8QkIMEyQZGLc2aE3TR0AIjgtgBoNz2Wd-eQt7lOthC3S5GC91dnBDe1kdwO1eSPijQP35AAzLd7o9xM9Pq9iicDlsAF43D4eHRQ0iXRSXeHAh5rMEvZEAVgOGgADKVEV9kSSicTidSaVjVuUQmFqrV6ktKG0Ol0bN0ME1OrUdH1kABeAQlUr9ZBxBJJAppDJZQIzRzzRbrfL4DA6fncQU9Xp3Cosrrs5CQLAGkUiWYIORYZAwdILSRi5Cc9oGoiDZKEUYgJisCa8KB8Ph8pqtL1dXr2WbOpUQAB0WswyZgMigAFFEAALbrdcHqX6HO5G8WqnHAGDIbp+KzJ4AYBuKboQ3pG204tb2kCOyQqaD4SgANRL7ohyYw8mASD6ycHw4gfXjPbtDpk8hT8hkDG6S6gI-HmjjoLEwRI8nwVfXyD7GC3O737ZLZ-XoXXD6fyd3+8+7jvuIwRnpUlpgNacZAA

This is the playground link for 4.1.3, where there is an error:

https://www.typescriptlang.org/play?ts=4.1.3#code/JYOwLgpgTgZghgYwgAgCoQM5gMJwxAHlWQgA9IQATDZLKUAc2QB9kQBXAWwCNoA+ZAG8AUMjHIoEOJQD2IADYBPZADc48gIwAuNC2QAlKbIWKAglChxFRPgG5R4ydLlLV6gEw7Dzk+cvXUO2EAX2FhUEhYRBQAWXZ5MGBcfAAxYHJ2SRshBzEEPEwdVABtAF17UOEYdhAERLladm4AETgwOBSahAAKAEoizBwCgjpGPQ4efjKc8QkIMEyQZGLc2aE3TR0AIjgtgBoNz2Wd-eQt7lOthC3S5GC91dnBDe1kdwO1eSPijQP35AAzLd7o9xM9Pq9iicDlsAF43D4eHRQ0iXRSXeHAh5rMEvZEAVgOGgADKVEV9kSSicTidSaVjVuUQmFqrV6ktKG0Ol0bN0ME1OrUdH1kABeAQlUr9ZBxBJJAppDJZQIzRzzRbrfL4DA6fncQU9Xp3Cosrrs5CQLAGkUiWYIORYZAwdILSRi5Cc9oGoiDZKEUYgJisCa8KB8Ph8pqtL1dXr2WbOpUQAB0WswyZgMigAFFEAALbrdcHqX6HO5G8WqnHAGDIbp+KzJ4AYBuKboQ3pG204tb2kCOyQqaD4SgANRL7ohyYw8mASD6ycHw4gfXjPbtDpk8hT8hkDG6S6gI-HmjjoLEwRI8nwVfXyD7GC3O737ZLZ-XoXXD6fyd3+8+7jvuIwRnpUlpgNacZAA

[orta]

Interesting, wonder what's up with the bot then - it should be the same underlaying code as the playground. Perhaps the repro bot has a default setting set which causes it to act differently.

Confirmed that the error occurs on master in a test case too. It can be fixed by reverting this:

-                // If the candidate type is a subtype of the target type, narrow to the candidate type,
-                // if the target type is a subtype of the candidate type, narrow to the target type,
-                // otherwise, narrow to an intersection of the two types.
-                return isTypeSubtypeOf(candidate, type) ? candidate : isTypeSubtypeOf(type, candidate) ? type : getIntersectionType([type, candidate]);
+                 // If the candidate type is a subtype of the target type, narrow to the candidate type.^M
+                 // Otherwise, if the target type is assignable to the candidate type, keep the target type.^M
+                 // Otherwise, if the candidate type is assignable to the target type, narrow to the candidate^M
+                 // type. Otherwise, the types are completely unrelated, so narrow to an intersection of the^M
+                 // two types.^M
+                 return isTypeSubtypeOf(candidate, type) ? candidate :
+                     isTypeAssignableTo(type, candidate) ? type :
+                     isTypeAssignableTo(candidate, type) ? candidate :
+                     getIntersectionType([type, candidate]);

Will remove the error - @weswigham, do you have a good argument for keeping it around?

[mwgamble]

Is there an update on this? I can see that some related PRs appear to have been merged.

[orta]

The changes should have all been reverted I believe, good reminder to close this out

[lorenzodallavecchia]

@orta is the fix in 4.3 already?
Testing in v4.3.0-beta still shows some problems.

If you look at the playground @mwgamble at line 27 the val1 variable is refined as any[] and so it appears to work. However, in typescript@4.1.2 it was correctly inferred as (string|number)[].

Also, see this playground of mine.

• on typescript@4.1.2
  This is the last version where everything worked.
• on typescript@4.3.0-beta
  The inferred type for elem at line 3 is any and is not even reported by noImplicitAny!
  Also note how the infeered type of the false branch of the guard is not refined, causing the error at line 5.