Question

I have a following simple script for parsing out dates from irc logs (created by irssi)

#!/bin/bash                                                                                  
query=$1

grep -n $query logfile > matches.log 
grep -n "Day changed" logfile >> matches.log

cat matches.log | sort -n

It produces output like:

--- Day changed Tue Jul 03 2012
--- Day changed Wed Jul 04 2012
--- Day changed Thu Jul 05 2012
16:54 <@Hamatti> who let the dogs out
--- Day changed Fri Jul 06 2012
--- Day changed Sat Jul 07 2012
--- Day changed Sun Jul 08 2012
12:11 <@Hamatti> dogs are fun

But since I’m only interested in finding out dates for actual matches, I’d like to filter out all those

--- Day changed XXX XXX dd dddd

lines where they don’t follow by timestamp on the next line. So the example should output

--- Day changed Thu Jul 05 2012
16:54 <@Hamatti> who let the dogs out
--- Day changed Sun Jul 08 2012
12:11 <@Hamatti> dogs are fun

to get rid of all the disinformation that’s not useful.

edit.
After the answer by T. Zelieke I realised that I could make this more of a one-liner so I use the following now to save logfile from being iterated twice.

query=$1
egrep "$query|Day changed" logfile |grep -B1 "^[^-]" |sed '/^--$/d'

Answer 1

grep -B1 "^[^-]" data |sed '/^--$/d'

This uses grep to filter lines that do NOT start with a dash ("^[^-]"). -B1 asks to print the immediate line before a match.
Unfortunately grep separates then each match (pair of two lines) by an -- line. Therefore I pipe the output through sed to get rid of those superflouos lines.