A cute algorithm for card-image templates

There's a trick I think I saw originally in REXX, and which I think originally comes from the IBM mainframe world.

Suppose you have a record with some fixed format and you want to reformat it. For example, you have this:

199712100036325SITTLER   KRAGEN

And you want to reformat it to this:

KRAGEN    SITTLER    $00363.25  10/12/1997

The thing that would make this easy would be if you could write a couple of "picture" lines showing the desired input and output, and have software apply the transformation automatically:

199712100036325SITTLER   KRAGEN    
19YyMmDd2345678OPQRSTUVWXopqrstuvwx
opqrstuvwxOPQRSTUVWX $23456.78  Dd/Mm/19Yy
KRAGEN    SITTLER    $00363.25  10/12/1997

So far that's nothing terribly special. You use the correspondence of the characters in the before-and-after picture to show where to move the input characters around to in the output.

The special part is that it turns out you can implement this with a simple character substitution, the same kind of thing you would use to transform uppercase to lowercase or vice versa, or remove accents from ISO-8859-1 text for accent-insensitive comparison, or translate between EBCDIC and ASCII. Here's what it looks like in Python.

>>> import string
>>> the_input = '199712100036325SITTLER   KRAGEN    '
>>> beforepic = '19YyMmDd2345678OPQRSTUVWXopqrstuvwx'
>>> afterpic  = 'opqrstuvwxOPQRSTUVWX $23456.78  Dd/Mm/19Yy'
>>> cipher = string.maketrans(beforepic, the_input)
>>> string.translate(afterpic, cipher)
'KRAGEN    SITTLER    $00363.25  10/12/1997'

So first we compute a character substitution that would convert beforepic into the_input. Then we apply that substitution to afterpic, and we get the desired output.

It's not a very versatile trick --- all the characters in beforepic have to be distinct, so it can't work in this form for anything over 256 bytes, it only handles fixed-width fields, and you can see I had a hard time coming up with reasonable-looking characters to use in the templates even in this small example. But the clever thing about it is that, given the existing ability to translate a string of characters according to such a table of correspondences, and the ability to construct such a table from a before and after string, it only takes a couple of lines of code.