In this letter, Dijkstra talks about readability and maintainability in a time where those topics were rarely talked about (1968). This letter was one of the main causes why modern programmers don’t have to trouble themselves with goto statements. Older languages like Java and C# still have a (discouraged) goto statement, because they (mindlessly) copied it from C, which (mindlessly) copied it from Assembly, but more modern languages like Swift and Kotlin don’t even have a goto statement anymore.
TIL that C# and Java have a goto statement.
Forget about goto, the latest version of C# introduced what’s essentially comefrom.
To be fair,
awaitis a bit more likecomefrom, and it’s been around for a few releases now.async/await was introduced in version 4.5, released 2012. More than a few releases at this point!
How is
awaitlikecomefrom, any more than threading is likecomefrom? The variable context is preserved and you have no control over what is executed before theawaitreturns.
What…? That is a terrible idea.
It’s scary as fuck, yeah, but, to be fair, it’s only intended to be used by code generators, and it’s quite awkward to use outside of them.
Java doesn’t. Well, it’s a reserved keyword but it’s not implemented.
Yeah but we got labels with continue and break, so we can pseudo goto.
Following that logic
if,elseandwhileare also “pseudo goto” statements.There’s nothing wrong with conditional jumps - we couldn’t program without them. The problem with goto specifically is that you can goto “anywhere”.
If pretty much gets compiled to a goto statement. Well more a jumpif but same principle
In C# at least,
gotocan take you between case labels in a switch statement (rather than using fallthrough), which I don’t view as being nearly as bad. For example, you can dogoto case 1orgoto defaultto jump to another case.The only other use of
gotoI find remotely tolerable is when paired with a labelled loop statement (like putting a label right before aforloop), but honestly Rust handles that far better with labelled loops (and labelled block expressions).I’ve programmed C# for nearly 15 years, and have used
gototwice . Once to simplify an early break from a nested loop, essentially a nestedcontinue. The second was to refactor a giant switch statement in a parser, essentially removing convolutedwhileloops, and just did agotothe start.It’s one of those things that almost should never be used, but the times it’s been needed, it removed a lot of silliness.
For C it makes sense. The point of C is that it can work as a low level language. Basically, everything doable with assembly SHOULD be doable with C, and that’s why we don’t need another low level language that’s basically C with goto.
Even though almost all of C users should never use goto.
C is one of the few languages where using
gotomakes sense as a poor man’s local error/cleanup handler.Yeah. Without a proper error handling mechanism, goto is actually useful for once.
Still don’t get why Go simultaneously picked this and introduced
defer
switch statements are three gotos in a trenchcoat.
Anything is a goto in disguise when you think about it
In most CPU instruction sets, the only conditional instruction is branch (aka goto).
And after each instruction, the CPU go to the next instruction
Duff’s device takes this to a whole new level.
This is very nice and clean
Egads! My eyes.
I’d rather it was just written in assembly. It’s the
do {opening a block under thecase 0, but then proceeding to have furthercasestatements inside that block. You now have case statements in two different scopes that are part of the same switch.
Languages don’t have goto because they mindlessly copied it.
True it wasn’t mindless - just idiotic
It’s not idiotic. You can do a lot of performance optimizations with GOTO so providing it as a “use it if you know what you’re doing” option is fine. And some things are easier to read with GOTO.
FWIW the Linux source code is full of GOTO statements. Nearly 200,000 of them in fact.
There’s a solid reason for goto in C.
Bringing goto into Java was (and is) idiotic.
If you’re trying to squeeze every ounce of performance out of your code then you’ll need those optimizations.
But any higher level language than C the entire point is to write easily maintainable and useful code that any idiot can go, read and update. A goto is antithetical to readability.
It can be, sure. But when used in a limited manner where it makes sense it can be the more readable option. I’ve used it in a try/catch to retry the operation after changing a variable. One label (“reconnect”), one goto, totally easy to understand on a surface level.
For such an influential letter, I don’t find his arguement all that compelling. I agree that not using
go towill often lead to better structured (and more maintainable) programs, but I don’t find his metric of “indexable process progress” to satisfyingly explain why that is.Perhaps it’s because at that time people would be running the programs in their heads before submitting them for processing, so they tended to use more of a computer scientist mindset - whereas now we’re more likely to use test cases to convince ourselves that code is correct.
I think it’s convoluted way to describe it. Very technically-practical. I agree it’s probably because of historical context.
The argument I read out of it is that using
gotobreaks you being able to read and follow the code logic/run-logic. Which I agree with.Functions are similar jumps, but with the inclusion of a call stack, you can traverse and follow them.
I think we could add a goto stack / include goto jumps in the call stack though? It’s not named though, so the stack is an index you only understand when you look at the code lines and match goto targets.
I disagree with unit tests replacing readability. Being able to read and follow code is central to maintainability, to readability and debug-ability. Those are still central to development and maintenance even if you make use of unit tests.
I wasn’t saying that unit tests replaces readability, I was saying that back in the 60s they’d reason and debug using their brains (and maybe pen and paper), with more use of things like formal proofs for correctness. Now that we write more complicated programs in more powerful environments, it’s rare to do this (we’d use breakpoints, unit tests, fuzzing, etc).
Perhaps it’s because at that time people would be running the programs in their heads before submitting them for processing, so they tended to use more of a computer scientist mindset - whereas now we’re more likely to use test cases to convince ourselves that code is correct.
This is 1968. You didn’t have an IDE or debugger. Your editor was likely pretty terrible too (if you had one). You may have still been using punch cards. It’s possible the only output you got from your program was printed on green-bar paper.
“Back in the day” it wasn’t uncommon to sit with a printout of your code and manually walk though it keeping state with a pencil. Being able to keep track of things in your head was very important.
GOTO existed in part for performance purposes. When your CPU clock is measured in megahertz, your RAM is measured in kilobytes and your compilers don’t do function in-lining it’s quicker and cheaper to just move the program pointer than it is to push a bunch of variables on a stack and jump to another location, then pop things off the stack when you’re done (especially if you’re calling your function inside a loop). Even when I was programming back in the '80s there was a sense that “function calls can be expensive”. It wasn’t uncommon then to manually un-roll loops or in-line code. Compilers are much more sophisticated today and CPUs are so much faster that 99% of the time you don’t need to think about now.
Oddly enough the arguments against GOTO are less important today as we have much better development tools available to us. Though I certainly won’t recommend it over better flow-control structures.
When your CPU clock is measured in megahertz, your RAM is measured in kilobytes
Ah yes, the good ol’ days when developers programmed for efficiency.
Mostly because they had to. Writing efficient code and easy-to-read code are often at odds with each other. I like being able to create lots of functions that can be called from a loop without needing to worry too much about function call overhead. I can prioritize readability for some time before I ever need to worry about performance.
I get that but it seems as though no one cares at all about efficiency these days.
People have been complaining about this exact thing forever. Even back when “people cared about efficiency”.
Does the catchphrase “blazing fast” ring any bells? Some people care.
(Arguably that’s just the pendulum swinging the other way; Ruby, Python, and Java ruled the software world for a while, and I think that’s a large part of why the Go and Rust communities make such a big deal about speed.)
Their main argumentation (from page 1) summarized:
You know the state and progress of a program from the line you are on. A
gotobreaks that.You can index the progress of a program through static line indexes and a dynamic loop index and function call stack. A
gotobreaks that. Including a “statements/lines since beginning of execution” is infeasible for understanding.gotodoes have some uses, such as single exit point, but should be used sparingly.Single exit point is an assembly rule that was mindlessly ported to C/C++. It doesn’t make sense in a structured language.
Maybe for your use-cases, for mine i find it quite useful.
I’ll be very interested to hear more, do you have a blog?
A common case is a function that uses resources like opening files, sockets, whatever. If there’s an error you want to ensure a clean exit, so at different parts of said function you
gotothe end where you check if said resources are being used and release them before exiting.Cleaner than constantly checking at every error.
Go has
gototoo. They surely did not “mindlessly copy” it.The standard library makes use of it. So they most definitely see a warranted use-case for it.
OP argument against using it in high level languages may still hold though. Go may have introduced it as a systems language which allows control over alternative implementations.
The article does say that there are good cases to use goto, but they are rare and most programmers won’t ever encounter such situations. I believe the jist is that it can do nore harm than good.
Java does not support goto.
And yet it’s still easy to write spaghetti code in Java. Just abuse inheritance. Where is this function implemented? No one knows but the compiler!
You mean
SpaghettiFactory()And don’t forget to use miracle sort for any sorting needs you have. Why do the work yourself when you can just hope it gets solved for you?!
That doesn’t make it spaghetti code though. In well-written OOP code you shouldn’t care where a function is implemented. The problem is a much too high level of abstraction. If your high level code is so abstract that it is only running tasks and handling messages there’s no way to write it in a way that prevents mistakes because you couldn’t possibly know what the actual implementations do.
Well it has labeled breaks but that’s the closest it gets to it.
PDF magic… It has grainy text. But the selectable text and displayed text have a 1-character offset.
I assume they display the original scan so it definitely does not contain errors, while still providing the image-parsed text for searchability, indexability, and select-+copyability?
Unfortunately, the grainy text is hard[er] to read.
