See [diff in Tcl] 

The code that was here was crap (according to the author) and has been removed. 

 [Arjen Markus] We have faced a slightly different problem: two files that should be compared with special care for (floating-point) numbers. The solution was simple in design:

   * Read the files line by line (all lines should be comparable, we did not need to deal with inserted or deleted lines)
   * Split the lines into words and compare the words as either strings or as numbers.
   * By using [[string is float]] we identified if the "word" is actually a number and if so, we compared them numerically (even allowing a certain tolerance if required).

This way you are immune to numbers formatted in different ways:
0.1, +.1, 1.0E-01, +1.00e-001 all spell the same number and you can encounter all of these forms (sometimes you have less than perfect control over the precise format).
----
[Arjen Markus] Question: would not this be a nice addition for the fileutil module in Tcllib?

[GPS] maybe it would...

[Arjen Markus] If so, it would benefit (in my opinion) from two custom procedures:

   * A procedure one can supply to compare the lines (for instance: ignore white-space or interpret numbers as numbers - my original problem)
   * A procedure to process the output (in a manner as [Tkdiff] does for instance)

----
[Arjen Markus] A few thoughts for improving the performance:

   * Store the lines as {lineno content}
   * Sort by content (lsort has this ability via "-index")
   * Use binary search to replace the inner loop.

This would bring back the number of iterations from O(N^2) to O(NlogN).
But perhaps it is not worth the trouble :-)

----
See also [diff in Tcl] and [Using Snit to glue diff, patch, and md5sum]