Fletch Rydell's Blog

While working on a compiler for Appel's “Tiger” language, I encountered an interesting edge case I'd never thought of: what does break do inside of a while loop's condition?

Inside the body of the loop, the break of course applies to the innermost containing loop expression, stopping iteration over the body. However, since this is an expression-oriented language, it's also perfectly valid to put a break in the condition of the while loop, for example:

while (
        if done then break;
        i:=i+1;
        i <= 10
    )
do (
    if g(i) then done := 1;
    print(g(i))
)

So, what should this do? Given that the while loop condition is executed once per iteration, I opted to make the break behave like one inside the body, jumping to the end of the loop. This has the advantage of allowing do/while-like constructs by putting the entire body in the condition. However, I can see an argument for treating such breaks as part of the surrounding body, breaking out of any outer loop (or erroring if no such loop exists).

Looking at actual languages, it's surprisingly hard to find any where this is even syntactically possible! C and many syntactically similar languages require the loop condition to be an expression, while break is a statement. Expression-oriented languages don't have this issue, but many of them (mostly functional languages) don't have loops at all (or like Scala, have loops but not break).

Out of the languages I tested, the only ones supporting this are Rust, Perl, and Kotlin. Rust throws error E0590, requiring the loop to break out of to be labeled explicitly. Perl takes my approach, breaking out of the loop containing the conditional. Kotlin actually does the opposite, interpretting while(break) {} the same as break when inside a loop. However, it also gives an Unreachable code warning (leading one to wonder exactly what code it considers unreachable given the empty loop body).