Read a date in xkcd notation

CJam, 35 bytes

ll{1$e>}*](l+eeWf%$-8>Wf=\f=2/(o'-*

Try it here. It expects the input lines to be padded with spaces.

Explanation

ll reads two lines of input, and {1$e>}* performs a "scan" on the second one: it it takes all prefixes of its input and computes the maximum of each prefix. For the input line "0 1 2 7 8", this pushes "0001112227778". Our stack now looks like this:

"first line" '0 '0 '0 '1 '1 '1 ...

We need to re-capture the values into a list ourselves using ]; this captures our first line as well, so we pop it back out using (, to get

"0001112227778" "first line"

as expected.

eelee+ enumerates this line, then does the same for a third input line, and concatenates the results, leaving something like this on the top of the stack:

[[0 '5] [1 ' ] [2 ' ] [3 ' ] [4 ' ] [5 ' ] [6 ' ] [7 ' ] [8 '6] [9 '7] [10 '8] [11 ' ] [12 ' ]
 [0 '2] [1 ' ] [2 ' ] [3 ' ] [4 '4] [5 ' ] [6 ' ] [7 ' ] [8 '1] [9 ' ] [10 ' ] [11 ' ] [12 '3]]

Now our stack is ["0001112227778" X] where X is the enumerated list above.

We flip each pair in X (Wf%), sort the pairs lexicographically ($), and leave the last 8 pairs -8>. This gets us something like:

[['1 8] ['2 0] ['3 12] ['4 4] ['5 0] ['6 8] ['7 9] ['8 10]]

This works, because the sort places all the pairs with key ' (space) before all the digits in ascending order.

These are the "x-positions" of the characters 12345678 on the first and third lines: we only need to retrieve the characters from our (modified) second line that are vertically aligned with them.

To do this, we take each position (Wf=), index into the string we made earlier (\f=). We have "20610222" on the stack now: to add the dashes, first we split into segments of length two (2/), print the first segment without a newline ((o) and join the remaining segments with dashes ('-*).

EDIT: cool scan trick, Martin! Saved four bytes.

EDIT 2: saved two more bytes by replacing eelee+ with l+ee; this works, because the lines all have the same lengths, and list indexing in CJam is automatically modulo the list length, so the indices n+0, n+1, n+2... nicely map to 0, 1, 2...

EDIT 3: Martin saved another byte in the final step of the process. Nice!

Pyth, 48 43

j\-cj\-ctuuXN.xsTZK?qJhtHdKJ+hHeHGC.z*9d4 7

Test Suite

Requires padding with spaces into a rectangle.

I don't think this is the best approach, but basically it writes the middle value to the index in a string pointed at by the top or bottom value. Well I guess I had enough time to golf most of the obvious stuff I saw. :P

JavaScript (ES7), 115

Anonymous function. Using template strings, there is a newline that is signifiocant and included in byte count.

Requirement: the middle input line cannot to be shorter than the first or the last. This requirement is satisfied when the input is padded with spaces to form a rectangle.

x=>([a,z,b]=o=x.split`
`,d=i=0,[for(c of z)o[a[i]-1]=o[b[i++]-1]=d=+c||d],o.splice(4,2,'-',o[4],o[5],'-'),o.join``)

ES6 version 117 using .map instead of array comprehension

x=>([a,z,b]=o=x.split`
`,d=0,[...z].map((c,i)=>o[a[i]-1]=o[b[i]-1]=d=+c||d],o.splice(4,2,'-',o[4],o[5],'-'),o.join``)

Less golfed

x=>(
  o=[],
  [a,z,b] = x.split`\n`,
  d=i=0,
  [ for(c of z) (
      d = +c||d, // each new digit found in z goes in d (but not the spaces and not the '0' (d starts at 0 anyway)
      o[a[i]-1] = o[b[i]-1] = d, // if the index char is space, that gives index -1 that is ignored when joining later
      ++i
  )],
  o.splice(4,2,'-',o[4],o[5],'-'), // add the dashes in the right places
  o.join``
)

Test snippet

f=x=>(
  [a,z,b]=o=x.split`\n`,
  d=i=0,[for(c of z)o[a[i]-1]=o[b[i++]-1]=d=+c||d],
  o.splice(4,2,'-',o[4],o[5],'-'),o.join``
)


console.log=x=>O.textContent+=x+'\n';

[['2  3  1  4\n0  1  2  3  7\n5     67    8','2013-02-27']
,['2  3  1     4\n0  1  2  4  5\n   5  67 8','2015-12-24']
,['      1234\n1     2   \n5678','2222-11-11']
,['   1     3  24\n0  1  2  7  8 \n57    6     8','1878-02-08']
,['2   4   1   3\n0   1   2   6  \n5       678','2061-02-22']
,['      1 4 2 3\n0 1 2 3 4 5 6 8\n6 5 7         8','3564-10-28']
,['1234\n1   \n5678','1111-11-11']
,['1 2 3 4\n0 1 2 3\n8 5 6 7','0123-12-30']]
.forEach(t=>(k=t[1],r=f(t[0]),console.log(t[0]+'\n'+r+'\n'+(r==k?'OK\n':'Fail\n'))))

<pre id=O></pre>