# Default for Code Golf: Input/Output methods

It looks like we have a consensus that we want certain defaults for the format which answers are expected in for . On that poll, the question arose twice, which input/output formats should be allowed for programs and functions.

So here is another poll. This one works different though. All the input/output methods are independent of each other, so there will be one answer per method. Upvote all you think are reasonable for the default. Downvote those which you think shouldn't be allowed unless the OP explicitly permits them.

To keep this remotely manageable, I have not posted individual answers for all possible inputs for functions. So there are only four: functions can take input/output via their arguments and return values. Or functions can use any method full programs can. I don't think there is any point in (say) allowing programs to take input from STDIN (only) and to allow functions to take input from ARGV (only) or something like that. If you disagree, please leave a comment.

Note: Some votes have been reverted because they were detected as serial voting. If you vote on multiple answers, please leave some time between votes.

A method is allowed if it has 5 net votes and at least twice as many upvotes as downvotes.

### Update

The current results of the polls are now part of the tag wiki. Please notify me, if results change significantly and the wiki should be amended.

• Related (Can numeric input/output be in the form of byte values) Dec 13, 2016 at 16:10
• "The current results of the polls are now part of the tag wiki" which tag wiki? meta wiki has nothing for code-golf tag. ppcg wiki has codegolf.stackexchange.com/tags/code-golf/info but it's missing a lot of answers here. Jul 18, 2018 at 18:17

# Image output may consist of a matrix or 2d array for raster images

This is relevant e.g. for challenges. (e.g. for languages that otherwise have no way of writing an image file or opening a window for the graphical display)

# An array may be represented as a buffer terminated with a sentinel value

This can be applied to both inputs and outputs for languages such as C and assembly. This is the default for C strings (a buffer of char, terminated with 0), but can be applied to any other data type.

As a more concrete example:

• An input array of positive integers can be taken as a single pointer of uint32_t or any other convenient integer type, which is assumed to terminate with a single zero.
• For an output array of positive integers, a C function can take an output pointer as the start of the buffer, write several numbers on that buffer, and then write a zero at the end.

It should be noted that the sentinel value must be the one that is guaranteed to not appear in any valid input or output, since their presence breaks the premise of the I/O method. For example, if the input may contain zeros, you cannot use zero as the sentinel value for the input array.

## You may assume that the input is valid

When parsing an input, you may assume that the input is already valid. You are not required to validate it unless the challenge is specifically about testing whether the input is valid or not.

# Programs may input/output bytes in a bit-serial or 8-bit parallel manner with a clock and enable signal

This is mainly applicable to HDLs such as Verilog, as well as other embedded devices (such as microcontrollers). Currently, display is capable of writing to standard out in a simulator, but of course, not in a physical FPGA. • Why require an enable signal? If you take a clock as input, and can output a bit with every clock pulse, the enable signal is just wasting code. (I've used this in a submission already, although the question specified I/O methods explicitly rather than defaulting to this post.) – user62131 Jan 26, 2017 at 20:40 • @ais523 Sorry, I didn't think of that when I created this. It makes sense that the clock could be stopped, at least for input, but not sure how a program could communicate that it wants to delay output for a few clock cycles while calculating. Jan 26, 2017 at 20:46 • If the problem's simple enough, the program can guarantee that it'll never have to delay output. (Note that you could define a fairly slow clock speed.) Of course, for some algorithms (especially ones that are slower than constant or linear-bounded time) the enable signal might be required. – user62131 Jan 26, 2017 at 20:49 • Why need a clock signal? It is also possible to just use UART with a given Clock Frequency. And why 8 Bit? If you want what stop you from using 4, 12, 16 or 27 Bits? Jun 21, 2017 at 11:12 • @12431234123412341234123 There is often no good way to perform sequential or stateful computation in synthesizable Verilog without a clock signal. 8-bit is in alignment with the fact that 8-bit ASCII is common; you can submit a new answer that allows non-8-bit parallel inputs. Additionally, to output/input data a communication clock is required, whether for each word/bit or to actually run a UART. I would strongly doubt that anyone would want to spend valuable bytes on writing a UART, except perhaps in code-bowling challenges. Jun 21, 2017 at 21:32 # Numerical input and output can be given as a string of digits and vice-versa As long as the input/output format is consistent, you may take a numerical output such as f(123) as a string f("123") and vice-versa. This will also open up other string methods, such as taking a number as a list of digits as per this consensus. Note that if different behaviours are expected for numbers with leading zeroes, then you will have to take input as a string instead, since strings with leading zeroes will be mapped to the same number. You may not add leading zeroes if you are using string input instead of a number. # A program or function taking input by reference may output by modifying the input value in place A complement to this For example, in rust, one might write fn a(v:&mut Vec<i32>){v.push(7);} instead of fn b(mut v: Vec<i32>)->Vec<i32>{v.push(7);return v;} or fn c(v:&mut Vec<i32>)->&mut Vec<i32>{v.push(7);return v;} # Spreadsheet languages can take input from a specific cell / group of cells Most Excel / Google Sheets answers seem to do this anyway, I don't know why it wasn't here. For example, this answer takes input through the A1 cell of a spreadsheet and performs operations on that to arrive at the final output. # Assembly language / machine code must take input and output as defined by an ABI for the platform that predates the challenge For example, most commonly used ABIs for x86 state that 32-bit integer outputs are returned in the %eax register. As such, that's a valid way to output a 32-bit integer. However, outputting in, e.g., %esp would not be valid, unless a pre-existing ABI can be located where that's considered a valid method of output. (Some background for people unaware of assembly language technology: an ABI, or Application Binary Interface, is a specification for how pieces of code within a particular program, on a particular platform, communicate with each other via the state of the processor. Most ABIs specify specific memory addresses and/or registers to be used for the communication.) This is outright intended as a competing answer to the other assembly-language-related answers, because I feel they're too permissive and could lead to loopholes in which bytes are moved from the program to a selection of a register/memory address. (See this answer and the corresponding question for more detail.) • Assembly is free for platform etc it can have arg as one register and return its result as a flag... If it has to communicate or to be used to C language programs than yes has to follow API... – user58988 Feb 4, 2019 at 22:27 # Programs running on FPGAs may take input or give output through transceivers. • I'm not familiar with FPGAs, but I would upvote if there's no way to encode data into which transceiver is being used or how. Dec 3, 2018 at 23:28 # Mappings may be used in place of functions Whenever a function is expected, a mapping such as a dictionary/hashmap or array may be used instead. Arguments for: • in many languages, there is no syntax difference (K, MATLAB, Ruby - a Proc can be used with the same [] syntax as a Hash) • disallowing it borders on an unobservable requirement, particularly in the above languages, and having different rules for different languages is generally discouraged • it allows removing boilerplate in challenges which boil down to "construct a mapping and then look up into it", like (Python) lambda x:{k:v for k,v in ...}[x] Arguments against: • it feels a bit cheaty to me • I haven't downvoted, because I know mathematically a function can be viewed as just a (possibly infinite) mapping. But I also haven't upvoted because, as you say, it feels a little cheaty. I think if the return value isn't callable/executable in your chosen language, we shouldn't assume an interpreter to run it. May 30, 2021 at 1:23 • I have no reasons to disagree for challenges where the output is completely deterministic (i.e. a pure function), assuming that you can extract the output for any given input in finite time. The answer is invalid if the premise is not satisfied (which is equivalent to a function that falls into an infinite loop for that input). One thing to note is that we're already allowing this for sequence challenges in the form of infinite lazy lists. Jun 1, 2021 at 7:58 • Even if this is to be allowed as an output method, it should certainly not be allowed as an input method (for problems where you take a function as input), since a mapping includes extra information that a function does not, namely an enumeration of all items in the domain. Jun 9, 2021 at 2:54 # Functions may generate different output via different exceptions if they could be caught using different code For example, in C++, you could use throw 1; and throw true; to output two distinct values. They could be caught by catch(int) and catch(bool) in a try block respectively. (In most other languages, you could only throw values in an exception class.) # Image output may consist of a pixel-shader A pixel-shader is a function that takes the (x,y) coordinates of a pixel as an input, and outputs the color of this pixel (e.g. a scalar for grayscale images, a tuple for color images, a boolean value for binary images etc.) This is relevant e.g. for challenges. (also for GLSL / Shadertoy answers.) • Isn't this explicitly disallowed for graphical-output at the moment? I suspect it had a good reason too. Same for your other suggestion here. – Mast Nov 25, 2019 at 9:56 • @Mast It seems to me that the tag wiki only addresses ASCII-art, but these two types do not seem to be captured, that is why I also thought it would be a good idea to clarify that explicitly. Nov 25, 2019 at 10:15 • I find it weird that this lets solutions not specify the size of the image. Or does the function gave to produce some different value indicating out-of-bounds? – xnor Nov 25, 2019 at 10:25 • @xnor The size of the image should be specified, but I think the key is that if you want a bigger resolution of the same image you could just sample (x,y) more finely as an alternative to defining a bigger resolution in the start. Nov 25, 2019 at 10:59 • @flawr I'm not clear how you'd sample more finely. Are (x,y) floats? – xnor Nov 25, 2019 at 11:02 • If we normalize the coordinates e.g. to [0,1]^2 then (x,y) could be interpreted as floats, right. If you're using integers then the program itself also takes the limits (e.g. in shadertoy this is implicit) Nov 25, 2019 at 12:39 # Spreadsheet languages may output through cell values Most Excel / Google Sheets answers seem to do this anyway, I don't know why it wasn't here. For example, this answer outputs a created piece of text to the cell it's pasted into. # Functions may output via throwing the output value as an exception, if it could be caught, and read by the code catching the exception # In stack-based languages, inputs may be pushed onto the stack at the start of the program, and the output may be the top element on the stack at the end of the program This has already been proposed for functions, but it should also be allowed for full programs. # Programs may take input from a generator or iterator We already have a rule that programs may output via a generator. Just for completeness, I think we should specify it's a valid input method as well. Example in rust: |t:Box<dyn Iterator>|t.flatten() Other example, taking input via the returned generator in python: def k(): yield "Hello, "+str(yield "What is your name") Which can be used like t = k() print(next(k)) print(k.send("Charlie")) Another example: def k(f): return [int(next(f)),*f] • This would also work as (e.g. in Python) a function that takes a generator as an argument, and to take input, use next(g) (instead of yield). Aug 27 at 19:18 # black-box functions to be taken as input can be assumed to be predefined under a given name only if they cannot be passed as an argument. This holds for languages that do have classical functions, but cannot pass around functions like other values. This holds for instance for: Java up to version 7 This proposition is the opposite of that one which does not have this restriction. • -1 because I don't think this is well-defined (e.g. are functions in Underload passed as an argument? What about Python's functions, which aren't black box because you can inspect their bytecode?) Nov 26, 2017 at 23:39 • Also, Don't allow / ask for different things in languages with different capabilities (related issue). Nov 26, 2017 at 23:45 • @EsolangingFruit That doesn't seem to be the case here, it's not the challenge asking for different stuff for different languages, it's more like allowing more languages to compete by using more lenient consensuses for them, since otherwise they couldn't compete at all. Dec 10, 2017 at 12:14 • @EriktheOutgolfer Fair, but my view is that, to avoid ambiguity and rule-lawyering, we should allow all languages to do this. Dec 10, 2017 at 20:57 ## A full program may be input or output any time a function could If we can do this: x=>y=>x+y Then we can do this: print("print(int(input())+"+input()+")") • That Python code is...a curried program. what. Jun 20, 2017 at 2:18 • +1 from me. If functions can return other functions, full programs should be able to print other full programs Jun 16, 2019 at 18:25 # If a program is supposed to take a whole number as input, it may accept the number as a unary string. # Programs may take input based on their executable name i.e. argv[0] in C or0 in shell-like languages.

• This is a default loophole. If taking input through the executable name, those bytes count towards the score. Jan 6, 2021 at 1:57
• @Dingus no, as in the user must rename the executable to a string, and that string is one of the inputs. It is not stored program data Jan 6, 2021 at 7:51
• I don't really see how this isn't stored program data (where the data is one of the inputs). Leaving that aside, how is this different from hardcoding? Jan 6, 2021 at 10:50
• In some sense I guess this isn't much different from taking input from a file, which is allowed. But I imagine that file is meant to be a fixed target, not a moving one. Jan 6, 2021 at 11:02
• I don't really see how this is stored data. Instead of running ./program arg1 arg2, you run mv program arg0; ./arg0 arg1 arg2. Jan 6, 2021 at 11:10
• Or you can run it without mv via something like exec -a arg0 ./program arg1 arg2 in bash or execlp("program", arg0, arg1, arg2, 0) in C. From an OS perspective, \$0 is just a parameter that we usually leave set to the default value. May 30, 2021 at 1:26
• I'll say, I don't know of a situation where this could save bytes, but I don't see a reason to ban it. It's not the same as the loophole of assuming the program will have a fixed value controlled by the programmer. May 30, 2021 at 1:28

# Functions may take input from the output of a callback function

The callback function should be provided using a valid input method.

Complementing

# Languages that output on program completion can assume termination even when built to hang for the purpose of providing output.

That is, if a program is built to not a terminate, and requires output while running (say, for an Internal Truth Machine, or Scream Very Loudly, etc.), the program can assume to crash or be manually stopped to verify the output scales appropriately.

The argument for this is that several languages (such as Processing[Java], for example) will not print output unless a certain state is reached (the end of a Draw statement in Processing, per se). This means that programs stuck in While Loops would never provide any output. Languages like GolfScript, though it has a Print function, also doesn't actually Print the output until the program is complete - it holds the print in a buffer until the program eventually stops (for any reason, even interrupting). Likewise, GS prints the stack when it closes for any reason (depending on your interpreter; TIO fails at this) - so filling the stack with infinite As then waiting for a crash would also be within this parameter.

Several languages would otherwise fail infinitely-printing hanging challenges due to the act they wouldn't ever be capable of printing given the assumed infinite time and memory, which is the main argument for allowing this sort of output.

• As written, this would allow a GolfScript program to “solve” the halting problem: push false onto the stack, then run a universal Turing machine, then pop false and push true. If this is to be allowed, it should only be allowed for mechanisms that don’t allow the program to change its mind later. Jun 9, 2021 at 2:50
• I think we should not allow this. Verifying that the program is correct is essentially undecidable and if a program doesn't give output when it enters infinite loop, it will never give an output. Dec 13, 2021 at 15:09
• I, too, think we should not allow this. 1) It requires external input, namely some decision as to whether or not to terminate the program yet. 2) If the challenge is to output the count of objects that match a certain criterion, this would allow a program which searched an infinite search space, which would find any matching objects, but which would never know whether or not future searching would find any more. Jul 8 at 5:22

# A natural number may be represented as a collection of n empty collections, and a string as a collection of such numbers

In languages that do not have numbers or strings but do have lists, stacks, or other collections, a natural number (including 0) may be represented by a collection of empty collections, where the length of the outer collection represents the natural number. So [] would represent 0, [[], [], []] would represent 3, and so on. Strings may be represented as collections of these natural numbers, as Unicode codepoints are unsigned.

• Even if the language does not have a notion of a number or string, the interpreter can be made to do the necessary I/O via such conversion. Flurry already does that (a number is not a built-in, but stack contents are translated from/to numbers when the program starts/ends). Jun 2, 2021 at 3:22
• @Bubbler That's totally reasonable, but one of the main ideas of the language I was thinking about is that the interpreter has no conception of anything that isn't a stack.
– user
Jun 2, 2021 at 12:55

# Property setters may take input from the set value

• This could use some elaboration. Why is this not a special case of a function taking input as function arguments? And why would anyone answer a PPCG question with a property setter? Mar 23, 2017 at 8:55

## Functions are allowed to overwrite global data structures as its output

The premise of this is that your language does not support built-in return values.

For example, Assembly language doesn't have a built-in way of returning values, therefore a global accumulator is permitted as a return value.

In Scratch, there are no functions (only procedures), therefore global variables are allowed as the output of the function.

In stack-based languages, functions take arbitary items from the stack, and modify the stack. The return value is therefore stored onto the stack.

• In Scratch, variables can be outputted via say. And in a most stack-based languages, there are commands to output. So only if your language has NO output method. Mar 30, 2021 at 19:54

# Classes, objects or other entities may have separate methods to read the input and return the result

This includes the constructor and destructor.

• Can the destructor return the result? That seems to be along the same lines as outputting to /dev/null. May 23, 2017 at 3:15
• I don't believe a C++ destructor is allowed to return anything. Jan 11, 2018 at 8:21
• Is it okay if the destruct function output something?
– l4m2
Mar 8, 2018 at 12:09

# Functions may output by calling a callback function or event

The callback function should be provided using a valid input method.

Example in JavaScript that outputs Hello world!:

f=>f("Hello world!")

This may be shorter than returning a value in some languages:

import random as r
def f(x):
q=r.random()
return(q+q/2)/2

As opposed to:

import random as r
def f(x,g):
q=r.random()
g((q+q/2)/2)
• I don't see why this should be useful, considering you can just return the result. Feb 11, 2017 at 2:27
• Never mind, I see why this might be useful. Upvoted. Feb 13, 2017 at 0:16
• @Challenger5 Please explain why. Feb 28, 2017 at 12:34
• @haykam In a language like Python, something like import os [\n] def f(k):return os.system(k) is longer than import os [\n] def f(k,g):g(os.system(k)) Feb 28, 2017 at 17:51
• @Challenger5, the point of specifying default I/O methods is to allow answers to use idiomatic I/O, not to allow golfing a byte off print. Mar 1, 2017 at 15:04
• @PeterTaylor Might it be more idiomatic in some languages to use continuation-passing style rather than returning? Apr 26, 2017 at 16:49

# Cellular automata and 2d languages may output by interfacing with a user-built structure

Many of these languages do not have any other way to produce (non-visual) output. By "interfacing", I mean additional structures being created by the user to interact with the existing code in a way that the program provides only the necessary information (such as some sort of serial representation of the outputted values) in a way that could feasibly be used by the user's program to control another turing-complete automaton (so not in a tiny space in the middle of the program, or in an area that will be overwritten).

# Cellular automata and 2d languages may take input by interfacing with a user-built structure

Many of these languages do not have any other way to take input. By "interfacing", I mean additional structures being created by the user to interact with the existing code in a way that provides only the necessary information (such as some sort of serial representation of the inputted values) in a way that could feasibly be used by the user's program and controlled by another turing-complete automaton (so not in a tiny space in the middle of the program, or in an area that will be overwritten).

• This would seem similar to the "Languages without input like /// may input through insertion into their source code" Jul 25, 2021 at 14:42