As an alternative to the computed gotos, you can use regular functions with the `[[musttail]]` attribute in Clang or GCC to achieve basically the same thing - the call in the tail position is replaced with a `jmp` instruction to the next function rather than to the label, and stack usage remains constant because the current frame is reutililzed for the called function. `musttail` requires that the calling function and callee have the same signature, and a prototype.
You'd replace the JUMP_TARGET macro:
#define JUMP_TARGET goto *jump_table[(int32_t)l->input.p[l->pos]]
#ifdef __clang__
#define musttail [[clang::musttail]]
#elif __GNUC__
#define musttail [[gnu::musttail]]
#else
#define musttail
#endif
#define JUMP_TARGET return musttail jump_table[(int32_t)l->input.p[l->pos]](l, a, out)
See diff (untested): https://www.diffchecker.com/V4yH3EyF/
This approach has the advantage that it will work everywhere and not only on compilers that support the computed gotos - it just won't optimize it on compilers that don't support `musttail`. (Though it has been proposed to standardize it in a future version of C).
It might also work better with code navigation tools that show functions, but not labels, and enables modularity as we can split rules over multiple translation units.
Performance wise should basically be the same - though it's been argued that it may do better in some cases because the compiler's register allocator doesn't do a great job in large functions with computed gotos - whereas in musttail approach each function is a smaller unit and optimized separately.
Can't wait for mandatory TCO coming to Rust. But it's not there yet. https://github.com/phi-go/rfcs/blob/guaranteed-tco/text/0000...
Not sure I like the `become` keyword. Seems bizarre - someone encountering this word in code for the first time would have no idea what it's doing.
Why don't they just use `tailcall`? That would make it's obvious what it's doing because we've been using the term for nearly half a century, and the entire literature on the subject uses the term "tail call".
Even better would be to just automatically insert a tail call - like every other language that has supported tail calls for decades - provided the callee has the same signature as the caller. If it's undesirable because we want a stack trace, then instead have some keyword or attribute to suppress the tail call - such as `no_tail`, `nontail` or `donttail`.
Requiring tail calls to be marked will basically mean the optimization will be underutilized. Other than having a stack trace for debugging, there's basically no reason not to have the optimization on by default.
Rust does allow tail call optimization. But that's LLVM's decision to optimize tail calls on a case-by-case basis. An explicit syntax to denote tail calls would be the difference between tail call optimization and guaranteed tall call elimination, which is important because if you're writing a tail-recursive function then it's pretty trivial to blow the stack at any moderate recursion depth unless you can guarantee the elimination.
As for why it's not trivial for Rust to do this by default, consider the question of what should happen in the case of local destructors, which in an ordinary function would be called after `return myfunc()` returns, but in a tail-recursive function would need to be called beforehand. The proposals for `become` tend to handle this by making it a compiler error to have any locals with destructors in scope at the point of the tail-call, further motivating the explicit syntax.
As far as the name of the keyword: anyone who knows what "tail call" means will figure out "become" pretty quickly. If they don't get it from context clues, someone will just have to tell them, "oh, it's a tail call", and the confusion will dissolve, because "become" is really not a bad word for what happens in a tail call. (This is obviously less important than the implementation issues kibwen handled.)
I'm generally pretty conservative about keywords. But it changes the semantics of the return, so it makes sense to change the word used in that position.
On compilers that don't support musttail, won't this make the stack explode?