Still, Loren’s solution is the easiest to follow.

Although I can follow a lot of Seth’s solution, I am stumped by the regexp call & the guts of the for loop.

I understand the power of regular expressions & know that they are complex enough that several books have been published on just them. But, I’ve always had trouble understanding the syntax & I’ve never found a complete, comprehensive FRP for just the arguments for regexp.

Looking at Seth’s command, & the MatLab FRP for regexp, he is clearly using the 3rd syntax option. I understand the s is the cell array from above & under Remarks in the FRP, it does explain why s can be a cell array.

Further, I see that the expression is the ‘@[\w.]+’ string. But, I can’t find any documentation that allows the use of the square brackets in this syntax. In reading the FRP of regexp, it seems to me that ‘@’ all by itself should work.

Next, I see that ‘match’ is an allowable qualifier that forces the returned result to be the matching string. But, that isn’t what he wants here. He wants to return the index of the ‘@’, not the ‘@’ itself. And, frankly, since you know you are searching for the ‘@’ character, you could populate the cell array sl with the ‘@’ characters without needing to use the regexp function.

Finally, I see how the how ‘once’ is an allowable option.

This illustrates why MatLab either needs a more user friendly pattern matching function or it needs a much better written FRP for regexp.

I understand that given the history of regular expressions, it clearly is important for MatLab to include it. But, since the documentation out there in the world of Unix, C and C similar languages for regular expressions is generally inadequate for a user to master it, MatLab must not just base its documentation on these flawed sources & must write its own comprehensive set.

Cell arrays could in principle, be used, except they appear to use way too much memory. (Mine would be very sparse).

>> doCompare(10000)

ans =

seth: 1.5446
dave: 2.2605
steve: 0.4783
loren: 0.1521

What I found interesting is how the timing changes if you change the number of unique domains. Here are the results with 10,000 e-mail addresses, but only 10 unique domains:

>> doCompare(10000)

ans =

seth: 0.1973
dave: 2.2455
steve: 0.4725
loren: 0.1494

Of course, running time, although relatively easy to measure, is only one aspect to consider. One big difference between these algorithms is their memory usage.