Question

So my dear SOers, Let me be direct to the point:
specification: filter a text file using pairs of patterns.
Example: if we have a file:

line 1 blabla
line 2 more blabla
line 3 **PAT1a** blabla
line 4 blabla
line 5 **PAT1b** blabla
line 6 blabla
line 7 **PAT2a** blabla
line 8 blabla
line 9 **PAT2b** blabla
line 10 **PAT3a** blabla
line 11 blabla
line 12 **PAT3b** blabla
more and more blabla

should give:

line 3 **PAT1a** blabla
line 4 blabla
line 5 **PAT1b** blabla
line 7 **PAT2a** blabla
line 8 blabla
line 9 **PAT2b** blabla
line 10 **PAT3a** blabla
line 11 blabla
line 12 **PAT3b** blabla

I know how to filer only one part of it using ‘sed’:
sed -n -e ‘/PAT1a/,/PAT1b/{p}’
But how to filter all the snippets, do i need to write those pairs of patterns in a configuration file, read a pair from it, use the sed cmd above, go to next pair…?

Note: Suppose PAT1, PAT2 and PAT3, etc share no common prefix(like ‘PAT’ in this case)

One thing more: how to make a newline in quota text in this post without leaving a whole blank line?

Answer 1

I assumed the pattern pairs are given as a separate file. Then, when they appear in order in the input, you could use this awk script:

awk 'NR == FNR { a[NR] = $1; b[NR] = $2; next }
     !s && $0 ~ a[i+1] { s = 1 }
     s
     s && $0 ~ b[i+1] { s = 0; i++ }' patterns.txt input.txt

And a more complicated version when the patterns can appear out of order:

awk 'NR == FNR { a[++n] = $1; b[n] = $2; next }
     { for (i = 1; !s && i <= n; i++) if ($0 ~ a[i]) s = i; }
     s
     s && $0 ~ b[s] { s = 0 }' patterns.txt input.txt