Add support for sentinel and length refinement of tuples

[popham]

Fixes #4050, fixes #4377, fixes #4556, and fixes #4873.

[bradennapier]

+1 ! Definitely need this!! :-)

[anru]

it's absolutely killer feature for flow.

@vkurchatkin @mroch could we get merged this PR/functionality into the codebase ?

[petejodo]

I'd like to use this for better refining React children. To give a semi-contrived example:

type LayoutOneProps = {
    layoutOne: boolean,
    children: [
        React.Element<typeof Level>,
        React.Element<typeof Sidebar>,
        React.Element<typeof Content>
    ],
};

type LayoutTwoProps = {
    layoutTwo: boolean,
    // currently required to have `null` as the last value -------------------------v
    children: [React.Element<typeof LeftColumn>, React.Element<typeof RightColumn>, null],
};

const LayoutWrapper = (props: LayoutOneProps | LayoutTwoProps) => {
    if (props.layoutOne) {
        return (
            <div className="container">
                <div className="row">
                    {props.children[0]}
                </div>
                <div className="row">
                    {props.children[1]}
                    {props.children[2]}
                </div>
            </div>
        );
    } else if (props.layoutTwo) {
        return (
            <div className="container">
                <div className="row">
                    {props.children}
                </div>
            </div>
        );
    }
};

<LayoutWrapper layoutTwo>
    <LeftColumn />
    <RightColumn />
    {null /* currently required to do this */}
</LayoutWrapper>

[bradennapier]

Not sure why no one from the Flow team reviews or accepts any PR's... it kind of sucks... this is a year old and still flow fails at any kind of tuple refinement.

[goodmind]

Should probably use Dtoa.ecma_string_of_float

[goodmind]

Hey @mroch. Maybe you will find this useful too for tuples

[ilya-bobyr]

Rebased this change on top of master in #7974

[STRML]

It's a shame to see great PRs like this get left unreviewed and unmerged - this is a hugely useful feature.

[ilya-bobyr]

It's a shame to see great PRs like this get left unreviewed and unmerged - this is a hugely useful feature.

We got the "length" part in, in this PR: #7974

I have rebased the original change to work in compile codebase as of about a year ago.
But as @goodmind pointed out in the discussion for PR 7974, a technical discussion might need to happen before a final implementation would go through a review.

[STRML]

Thanks for pointing that out. Length refinement is rather limited but workable. Truthiness checks and less than / greater than does not trigger the refinement.

See this example.

[facebook-github-bot]

This pull request has been merged in c0401c3.