[SOLVED] Diff regex matching

nikmit · 06-15-2012, 04:36 AM

I am diff-ing two files, containing ip addresses.
I want to avoid output about differences in those ip addresses, so I am using:

Code:

diff -I '1[.2.3.4|.2.3.5]' file1 file2   # works

and that works as much as it does ignore 1.2.3.4 and 1.2.3.5
It is not however how one would construct a regex, and what I would think is a better way of doing it fails:

Code:

diff -I '1\.2\.3\.[4|5]' file1 file2   # fails
diff -I '1.2.3.[4|5]' file1 file2   # fails

The square brackets seem to have functionality specific to diff, which is not explained in the man pages.
Matching for two expressions also has it's quirks as I needed to escape the pipe, but pipe does't need to be escaped withing square brackets:

Code:

diff -I 'checksum|hostname' file1 file2    # fails
diff -I 'checksum\|hostname' file1 file2   # matches either checksum or hostname
diff -I '[checksum|hostname]' file1 file2  # works as well, as above

The question - is this expected and normal behaviour or am I basing my script on hacks which might not be available after next update?
Is there any way to escape a dot other than by devising some solution using square brackets?

pan64 · 06-15-2012, 05:40 AM

have you tried:
diff -I '1.2.3.[45]' file1 file2
or
diff -I '1.2.3.[4\|5]' file1 file2
?

nikmit · 06-15-2012, 09:02 AM

Quote:

Originally Posted by pan64

have you tried:
diff -I '1.2.3.[45]' file1 file2
or
diff -I '1.2.3.[4\|5]' file1 file2
?

The problem is not in getting the 'or' bit to work, but that I don't know how to escape the dots in the ip.
That is why if I move all the dots into the 'or' brackets, like in

Code:

diff -I '1[.2.3.4|.2.3.5]' file1 file2   # works

everything works. Why it works is another thing I don't know

I expected that escaping like '1\.2\.3\.4' should work, but it doesn't. I didn't expect that '1[.2.3.4|.2.3.5]' would work, but it does.

David the H. · 06-15-2012, 11:52 AM

I'm not all that familiar with the intricacies of diff, but the info page says that "-I" accepts "grep-style regular expressions". I'm assuming that means basic regex, and not extended, but that's not perfectly clear.

In any case, for characters with special meanings, just give each one you want to be literal its own bracket expression. (You can also backslash-escape them, but I think it's generally cleaner and safer to use brackets.)

Code:

diff -I '1[.]2[.]3[.][45]' file1 file2

BTW, this really shouldn't work, or at least doesn't do what you expect it to:

Code:

'1[.2.3.4|.2.3.5]'

It's equivalent to this:

Code:

'1[2345.|]'

That is, a 1, followed by a single character that's either 2-5, period, or pipe. Bracket expressions don't accept pipes for "or" patterns, as they already are "or" patterns, for single characters only.

Most versions of regex do accept '(string1|string2)' for "or" patterns of longer strings. In gnu programs with basic regex, you have to backslash escape the parentheses and pipe to make them special, so this might work for you too:

Code:

'\(1[.]2[.]3[.]4\|1[.]2[.]3[.]5\)'