Given two question IDs, try to figure out how similar they are by looking at the answers.


You will be given two question IDs for codegolf.stackexchange.com; you may assume that there exist questions for both IDs that are not deleted, but are not necessarily open. You must run through all of the answers and determine the minimum Levenshtein distance between the code in the answers to the two questions (not including deleted answers). That is, you should compare every answer in question 1 to every answer in question 2, and determine the minimum Levenshtein distance. To find the code in an answer, assume the following procedure:

How to find the code snippet

A body of text is the answer's actual code if it is in backticks and is on its own line, or if it is indented with 4 spaces, with an empty line above it, unless there is no text above.

Examples of valid and not-valid code snippets (with . as a space) (separated by a ton of equal signs)

This is `not a valid code snippet because it is not on its own line`
This is:
`A valid code snippet`
This is
....not a valid code snippet because there's no spacing line above
This is

....A valid code snippet because there's a spacing line above
....Valid code snippet because there's no other text

If there are no valid code snippets in the answer, ignore the answer completely. Note that you should only take the first codeblock.

Final Specs

The two question IDs can be inputted in any reasonable format for 2 integers. The output should be the smallest Levenshtein distance between any two valid answers from either challenge. If there are no "valid" answers for one or both of the challenges, output -1.

Test Case

For challenge 115715 (Embedded Hexagons) and 116616 (Embedded Triangles) both by Comrade SparklePony, the two Charcoal answers (both by KritixiLithos) had a Levenshtein distance of 23, which was the smallest. Thus, your output for 115715, 116616 would be 23.


You may assume that the question has at most 100 answers because of an API pagesize restriction. You should not ignore backticks in code blocks, only if the code block itself is created using backticks and not on its own line.


I terminated the bounty period early because I made a request to a mod to get a one-week suspension and I didn't want the bounty to be automatically awarded to the highest scoring answer (which happens to be the longest). If a new submission comes in or a submission is golfed enough to become shorter than 532 bytes before the actual end of the bounty period (UTC 00:00 on Jun 1), I will give that a bounty to stay true to my promise, after the suspension expires. If I remember correctly, I need to double the bounty period next time so if you do get an answer in, you might get +200 :)

  • 1
    \$\begingroup\$ I'm confused by what counts as a valid code snippet. Why not just whatever's in <code> tags in the html? \$\endgroup\$ Apr 24, 2017 at 6:24
  • \$\begingroup\$ @HelkaHomba What about the newline restrictions? I could try to find another way to incorporate those. \$\endgroup\$
    – hyper-neutrino
    Apr 24, 2017 at 6:26
  • \$\begingroup\$ @HelkaHomba Essentially, if the answer contains backtick-delimited code within a line, it should be ignored. \$\endgroup\$
    – hyper-neutrino
    Apr 24, 2017 at 6:32
  • \$\begingroup\$ This is one of those answers, where it's easier to do the main part of the question. Downloading the page and extracting the code blocks is harder than doing the levenshtein distance. \$\endgroup\$
    – Bálint
    Apr 24, 2017 at 12:05
  • 1
    \$\begingroup\$ Cool. Just checking. \$\endgroup\$
    – Matt
    May 29, 2017 at 2:51

3 Answers 3


Java + Jsoup, 1027 bytes

The first two arguments are the question IDs.


import org.jsoup.*;import org.jsoup.nodes.*;class M{String a1[]=new String[100],a2[]=new String[100],c[];int i1=0,i2=0;public static void main(String a[])throws Exception{String r="https://codegolf.stackexchange.com/questions/";M m=new M();m.c=m.a1;m.r(Jsoup.connect(r+a[0]).get());m.c=m.a2;m.r(Jsoup.connect(r+a[1]).get());int s=m.ld(m.a1[1],m.a2[1]);for(int i=2;i<m.a1.length;i++)for(int j=2;j<m.a2.length;i++){if(m.a1[i]==null)break;int d=m.ld(m.a1[i],m.a2[j]);if(d<s)s=d;}System.out.print(s);}void r(Document d){a:for(Element e:d.select("td")){for(Element p:e.select("pre")){ a(p.select("code").get(0).html());continue a;}}}void a(String d){c[c==a1?i1++:i2++]=d;}int ld(String a,String b){a=a.toLowerCase();b=b.toLowerCase();int[]costs=new int[b.length()+1];for(int j=0;j<costs.length;j++)costs[j]=j;for(int i=1;i<=a.length();i++){costs[0]=i;int nw=i-1;for(int j=1;j<=b.length();j++){int cj=Math.min(1+Math.min(costs[j],costs[j-1]),a.charAt(i-1)==b.charAt(j-1)?nw:nw+1);nw=costs[j];costs[j]=cj;}}return costs[b.length()];}}


import org.jsoup.*;import org.jsoup.nodes.*;

class M {
    String a1[]=new String[100],a2[]=new String[100],c[];
    int i1=0,i2=0;
    public static void main(String a[])throws Exception{
    String r="https://codegolf.stackexchange.com/questions/";
    M m=new M();


    int s=m.ld(m.a1[1],m.a2[1]);
    for(int i=2;i<m.a1.length;i++)for(int j=2;j<m.a2.length;i++){if(m.a1[i]==null)break;int d=m.ld(m.a1[i],m.a2[j]);if(d<s)s=d;}

void r(Document d) {
    a:for(Element e:d.select("td")) {for(Element p:e.select("pre")) { 
        continue a;

void a(String d){c[c==a1?i1++:i2++]=d;}

int ld(String a, String b) {
    a = a.toLowerCase();
    b = b.toLowerCase();
    int [] costs = new int [b.length() + 1];
    for (int j = 0; j < costs.length; j++)costs[j] = j;
    for (int i = 1; i <= a.length(); i++) {
        costs[0] = i;
        int nw = i - 1;
        for (int j = 1; j <= b.length(); j++) {
            int cj = Math.min(1 + Math.min(costs[j], costs[j - 1]), a.charAt(i - 1) == b.charAt(j - 1) ? nw : nw + 1);
            nw = costs[j];
            costs[j] = cj;
    return costs[b.length()];


  • \$\begingroup\$ beat me to it!!!! Nice! \$\endgroup\$
    – tuskiomi
    May 29, 2017 at 21:37
  • 1
    \$\begingroup\$ Welcome to PPCG! It's not against the rules to use a third-party library, but we require that the use of the library is noted with the language (so a Java answer that uses a library called JavaHTML would be labelled "Java + JavaHTML"). \$\endgroup\$
    – user45941
    May 29, 2017 at 21:42
  • \$\begingroup\$ Ok, thanks! I'll keep that in mind for next time! \$\endgroup\$ May 29, 2017 at 21:45
  • \$\begingroup\$ There is nothing stopping you from using a library in this challenge if you wanted to. \$\endgroup\$
    – Matt
    May 30, 2017 at 1:00
  • \$\begingroup\$ I might have to now that somebody has topped my answer! \$\endgroup\$ May 30, 2017 at 1:45

PowerShell, 532 Bytes

$a={irm "api.stackexchange.com/2.2/questions/$args/answers?pagesize=100&site=codegolf&filter=!9YdnSMKKT"|% i*}
$1=&$a $1;$2=&$a $2
    $c=&$r $1[$_]
        &{$c,$d=$args;$e,$f=$c,$d|% le*;$m=[object[,]]::new($f+1,$e+1);0..$e|%{$m[0,$_]=$_};0..$f|%{$m[$_,0]=$_};1..$e|%{$i=$_;1..$f|%{$m[$_,$i]=(($m[($_-1),$i]+1),($m[$_,($i-1)]+1),($m[($_-1),($i-1)]+((1,0)[($c[($i-1)]-eq$d[($_-1)])]))|sort)[0]}};$m[$f,$e]} $c $d

I left newlines in there for some readability. The are still reflected in my byte count.

Pretty sure I have a handle on this. The tough part for me was actually getting the Levenshtein distance since PowerShell does not have a builtin for that as far as I know. Because of that I was able to answer the related challenge on Levenshtein distance. When my code refers to an anonymous function for LD you can refer to that answer for a more detailed explanation as to how that works.

Code with comments and progress indicator

The code can get real slow (because of the LD) so I built in some progress indicators for myself so I could follow the action as it unfolded and not assume it was stuck in a loop somewhere. The code for monitoring progress is not in the top block nor counted in my byte count.

# Assign the two integers into two variables. 

# Quick function to download up to 100 of the answer object to a given question using the SE API
$a={irm "api.stackexchange.com/2.2/questions/$args/answers?pagesize=100&site=codegolf&filter=!9YdnSMKKT"|% i*}

# Quick function that takes the body (as HTML) of an answer and parses out the likely codeblock from it. 

# Get the array of answers from the two questions linked.
$1=&$a $1;$2=&$a $2

# Hash table of parameters used for Write-Progress
# LD calcuations can be really slow on larger strings so I used this for testing so I knew 
# how much longer I needed to wait.
$parentProgressParameters = @{
    ID = 1 
    Activity = "Get LD of all questions" 
    Status = "Counting poppy seeds on the bagel"

$childProgressParameters = @{
    ID = 2
    ParentID = 1
    Status = "Progress"

# Cycle each code block from each answer against each answer in the other question.
    # Get the code block from this answer
    $c=&$r $1[$_]

    # Next line just for displaying progress. Not part of code. 
    Write-Progress @parentProgressParameters -PercentComplete (($_+1) / $1.count * 100) -CurrentOperation "Answer $($_+1) from question 1"

        # Get the code block from this answer   
        $d=&$r $2[$_]

        # Next two lines are for progress display. Not part of code. 
        $childProgressParameters.Activity = "Comparing answer $($_+1) of $($2.count)"
        Write-Progress @childProgressParameters -PercentComplete (($_+1) / $2.count * 100) -CurrentOperation "Answer $($_+1) from question 2"

        # Anonymous function to calculate Levenstien Distance
        # Get a better look at that function here: https://codegolf.stackexchange.com/a/123389/52023
        &{$c,$d=$args;$e,$f=$c,$d|% le*;$m=[object[,]]::new($f+1,$e+1);0..$e|%{$m[0,$_]=$_};0..$f|%{$m[$_,0]=$_};1..$e|%{$i=$_;1..$f|%{$m[$_,$i]=(($m[($_-1),$i]+1),($m[$_,($i-1)]+1),($m[($_-1),($i-1)]+((1,0)[($c[($i-1)]-eq$d[($_-1)])]))|sort)[0]}};$m[$f,$e]} $c $d
# Collect results and sort leaving the smallest number on top.

My logic for finding the code blocks is to take answer as HTML and look for a code tag set, optionally surrounded by a pre tag set that starts on its own line. In testing it found all the correct data on 6 different question sets.

I tried to work from the markdown code but it was too difficult to find the right code block.

Sample runs

Challenge-Similarity-Detector 97752 122740

Challenge-Similarity-Detector 115715 116616
  • \$\begingroup\$ I have spent the better part of 3 days looking into this. This challenge is in my top 5 for most fun attempting. TFTC(Thanks for the challange) \$\endgroup\$
    – Matt
    May 30, 2017 at 0:35
  • \$\begingroup\$ Nice job! Thanks, I'm glad you enjoyed it! :) \$\endgroup\$
    – hyper-neutrino
    May 30, 2017 at 0:59
  • \$\begingroup\$ Note: I awarded the bounty earlier than stated because I am requesting a suspension so I cannot award it any later. Good job! :) \$\endgroup\$
    – hyper-neutrino
    May 30, 2017 at 21:19
  • \$\begingroup\$ Requesting a suspension? \$\endgroup\$
    – Matt
    May 30, 2017 at 21:34
  • \$\begingroup\$ Yes, I asked Dennis to give me a 1-week suspension so I can focus on schoolwork. It's been done before (though I'm still here... I don't know when I'll disappear). \$\endgroup\$
    – hyper-neutrino
    May 30, 2017 at 21:40

Mathematica, 540 bytes

f=Flatten;l=Length;P=StringPosition;(H[r_]:=Block[{s,a,t,k},t={};d=1;k="https://codegolf.stackexchange.com/questions/"<>r;s=First/@P[Import[k,"Text"],"<pre><code>"];a=f[First/@P[Import[k,"Text"],"answerCount"]][[1]];While[d<l@s,If[s[[d]]>a,AppendTo[t,s[[d]]]];d++];Table[StringDelete[StringCases[StringTake[Import[k,"Text"],{t[[i]],t[[i]]+200}],"<pre><code>"~~__~~"</code></pre>"],{"<pre><code>","</code></pre>"}],{i, l@t}]];Min@DeleteCases[f@Table[EditDistance[ToString@Row@H[#1][[i]],ToString@Row@H[#2][[j]]],{i,l@H[#1]},{j,l@H[#1]}],0])&


["115715", "116616"]



uses built-in EditDistance which "gives the edit or Levenshtein distance between strings or vectors u and v."

As for the test case mathematica


returns 23

I guess I can golf it a bit more
Takes some minutes to run


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.