"The algorithm uses exactly the same terminology and is presented in bottom-up form. (If you prefer top-down design, please read the rest of this section backwards.)"
These classic-era structured programmers really loved their sentinels and really hated their gotos/breaks/returns. You can't tell me that it's easier to prove the correctness of this with Hoare-Dijkstra-style assertions than if you had an early return; assertions flow across control flow edges just fine, and when you have multiple control flow predecessors your precondition is the logical disjunction of those in-flowing assertions. Bob Floyd used Hoare-style triples for flow charts before Hoare.
@pervognsen Like honestly my day job is working in a pre-existing large codebase, and the number of times I have to restrain myself from fixing some convoluted mess with a simple "goto" is easily >=3 per week.
If the code is scary, it should look scary. And goto is scary by design. You should be scared! But not pathologically phobic.
@TomF Well, in this case you'd just need a break or an early return. When you look at a lot of the old articles on GOTO vs structured programming, including Knuth's famous article, they're working in a world where neither break or early return is a thing in structured languages, or at least not considered admissible. Almost all Knuth's examples (as far as I remember) in favor of GOTO are solvable with single-level continue, break or early return. Labeled break/continue probably handles the rest.
@TomF Most of the remaining examples that "need" GOTO rather than labeled break/continue come from irreducible control flow graphs, usually from non-trivial finite state machines where instead of representing the state as a variable and using a loop with a switch on the state variable, you implicitly encode the state via the program counter and use direct GOTOs between states. Those cases do occur but pretty infrequently. The common "GOTO the error handler" example is just labeled break.
@TomF As an aside, modern compilers perform an optimization called jump threading that turns the above example of a loop + state switch into the equivalent machine code of the GOTO version. So from the perspective of the final machine code (which was Knuth's concern in that article), there's no advantage even in that remaining class of examples.
@pervognsen I mean LLVM has no flow control constructs except for "goto" so yeah good job bending over backwards to avoid that, structured programming fetishists.
@lritter@TomF@pervognsen WASM also only has structured control flow, but to my surprise my hilarious CPU emulator switch-case-goto contraption performs quite well in WASM compared to native (trigger warning for "goto considered harmful" believers):
@StompyRobot This is not really true example of goto either. But you're right that this can only be expressed via goto in some languages if you don't want to use conditionals. Rust does not have goto but you can express the pattern you mentioned with blocks and labeled break. Similarly, Wasm does not have goto (and in fact cannot express irreducible CFGs directly) but you can express that clean-up pattern by branching to an outer block.
@TomF@StompyRobot Tom, I didn't say a bad thing about goto. You have to take your meds and stop harassing me in my own thread after I took you off the reply line several times, intentionally.
I guess you could use Pascal's nested functions to make a loop termination test that hand-implements the short-circuiting, which I think would be clearer than the sentinel shenanigans, but yeah, "break" still wins.
@sgf@TomF The short-circuiting is a funny example of something we all take for granted now. Even in Turbo Pascal, which was the Pascal that was foundational for my own early experiences, short-circuiting was an option you had to explicitly enable (with the equivalent of a scoped pragma/feature at the level of a source file or block of code) since it was not short-circuiting in classic Pascal. And I still remember that it was usually presented as an "optimization" rather than load-bearing.
@regehr It's a global constant: w for width, x is an array [0..w], keeping in mind that this is inclusive Pascal arrays so it actually has w+1 entries.
@regehr That part isn't too objectionable to me. And I like Go's take on this with named (multiple) return values. I'm guessing that came to Go via Robert Griesemer, who went to ETH Zurich and probably was directly influenced by Wirth and his languages. As for having the same name as the function in ALGOL derivatives, well, there's a certain elegance to it compared to having to make up a generic variable name like 'result'. Though I agree it looks weird to modern eyes.
@regehr How much do you hate Haskell's /=? If you're an ingrained C programmer that one is fun. At least it's defensible on the grounds of visual similarity to a strike-through equality sign in hand-written math. And it has precedent in a language (the first language!) since it's what FORTRAN uses. I've never figured out from what origin or analogy ALGOL's diamond-shaped <> is supposed to have originated. That one must be a sui generis product of late 1950s pipe smoke.
Add comment