Switch is considered exhaustive or not based on the code after it

[kendfrey]

TypeScript Version: 2.5.3

Code

With --noImplicitReturns:

1:

function foo(bar: "a"): number
{
    switch (bar)
    {
        case "a":
            return 1;
    }
}

2:

function foo(bar: "a"): number
{
    switch (bar)
    {
        case "a":
            return 1;
    }

    let unreachable;
}

3:

function foo(bar: "a"): number
{
    switch (bar)
    {
        case "a":
            return 1;
    }

    function doNothing()
    {
    }
}

Expected behavior:

Either:

(1), (2), and (3):

error TS2366: Function lacks ending return statement and return type does not include 'undefined'.

Or:

(1) and (3) compile.
(2):

error TS7027: Unreachable code detected.

Actual behavior:

(1) compiles.
(2) and (3):

error TS2366: Function lacks ending return statement and return type does not include 'undefined'.

Additional notes:

I'm not sure which is the correct behaviour, but right now it's inconsistent.

Originally posted as a question here: https://stackoverflow.com/q/46518643/785745

[ahejlsberg]

Currently, when an exhaustive switch statement is the last statement of a function body, we consider the end point unreachable. We could potentially ignore trailing declarations that contain no executable code in this analysis.

[RyanCavanaugh]

Accepting PRs - when detecting whether the switch is "final", skip over any declarations that are hoisted (functions but not classes), declarations that are type-only (interface, type, const enum), or declarations that have no side effects (var)

[charlespierce]

@RyanCavanaugh To make sure I understand correctly: Should a var declaration only be considered side-effect free if it has no initializer? I.e. var a; should be skipped but var a = 1; should not be?

Also, is there any reason why we can't consider any top-level exhaustive switch to be "final"? Since an exhaustive switch will necessarily return along all code paths, any code after it should be unreachable and thus a function with a top-level exhaustive switch won't ever have an implicit return.

[DanielRosenwasser]

If it's truly exhaustive (i.e. it has a default and all branches throw or return) then that does seem reasonable.

[charlespierce]

@DanielRosenwasser The current logic that determines a Switch is exhaustive requires that the switch:

1. Has no default branch

2. Expression is a literal type, enum, or union of literal types.

3. All sub-types of the expression are covered by the case statements within.

4. All branches throw or return.

That seems at first glance to be more restrictive than your conditions. I'll take a look at implementing it so that functions with exhaustive switches never have implicit returns.

[charlespierce]

PR Opened with the above logic #20063

[mhegazy]

thanks @charlespierce !