Append to line that is preceded AND followed by empty line - perl

Question

I need to append an asterisk to a line, but only if said line is preceded and followed by empty lines (FYI, said empty lines will NOT have any white space in them).
Suppose I have the following file:
foo
foo
foo
foo
foo
I want the output to look like this:
foo
foo
foo
foo*
foo
I tried modifying the following awk command (found here):
awk 'NR==1 {l=$0; next}
/^$/ {gsub(/test/,"xxx", l)}
{print l; l=$0}
END {print l}' file
to suit my uses, but got all tied up in knots.
Sed or Perl solutions are, of course, welcome also!
UPDATE:
It turned out that the question I asked was not quite correct. What I really needed was code that would append text to non-empty lines that do not start with whitespace AND are followed, two lines down, by non-empty lines that also do not start with whitespace.
For this revised problem, suppose I have the following file:
foo
third line foo
fifth line foo
this line starts with a space foo
this line starts with a space foo
ninth line foo
eleventh line foo
this line starts with a space foo
last line foo
I want the output to look like this:
foobar
third line foobar
fifth line foo
this line starts with a space foo
this line starts with a space foo
ninth line foobar
eleventh line foo
this line starts with a space foo
last line foo
For that, this sed one-liner does the trick:
sed '1N;N;/^[^[:space:]]/s/^\([^[:space:]].*\o\)\(\n\n[^[:space:]].*\)$/\1bar\2/;P;D' infile
Thanks to Benjamin W.'s clear and informative answer below, I was able to cobble this one-liner together!

Answer 1

A sed solution:
$ sed '1N;N;s/^\(\n.*\)\(\n\)$/\1*\2/;P;D' infile
foo
foo
foo
foo*
foo
N;P;D is the idiomatic way to look at two lines at the same time by appending the next one to the pattern space, then printing and deleting the first line.
1N;N;P;D extends that to always having three lines in the pattern space, which is what we want here.
The substitution matches if the first and last line are empty (^\n and \n$) and appends one * to the line between the empty lines.
Notice that this matches and appends a * also for the second line of three empty lines, which might not be what you want. To make sure this doesn't happen, the first capture group has to have at least one non-whitespace character:
sed '1N;N;s/^\(\n[^[:space:]].*\)\(\n\)$/\1*\2/;P;D' infile
Question from comment
Can we not append the * if the line two above begins with abc?
Example input file:
foo
foo
abc
foo
foo
foo
foo
There are three foo between empty lines, but the first one should not get the * appended because the line two above starts with abc. This can be done as follows:
$ sed '1{N;N};N;/^abc/!s/^\(.*\n\n[^[:space:]].*\)\(\n\)$/\1*\2/;P;D' infile
foo
foo
abc
foo
foo*
foo*
foo
This keeps four lines at a time in the pattern space and only makes the substitution if the pattern space does not start with abc:
1 { # On the first line
N # Append next line to pattern space
N # ... again, so there are three lines in pattern space
}
N # Append fourth line
/^abc/! # If the pattern space does not start with abc...
s/^\(.*\n\n[^[:space:]].*\)\(\n\)$/\1*\2/ # Append '*' to 3rd line in pattern space
P # Print first line of pattern space
D # Delete first line of pattern space, start next cycle
Two remarks:
BSD sed requires an extra semicolon: 1{N;N;} instead of 1{N;N}.
If the first and third line of the file are empty, the second line does not get an asterisk appended because we only start checking once there are four lines in the pattern space. This could be solved by adding an extra substitution into the 1{} block:
1{N;N;s/^\(\n[^[:space:]].*\)\(\n\)$/\1*\2/}
(remember the extra ; for BSD sed), but trying to cover all edge cases makes sed even less readable, especially in one-liners:
sed '1{N;N;s/^\(\n[^[:space:]].*\)\(\n\)$/\1*\2/};N;/^abc/!s/^\(.*\n\n[^[:space:]].*\)\(\n\)$/\1*\2/;P;D' infile

Answer 2

One way to think about these problems is as a state machine.
start: state = 0
0: /* looking for a blank line */
if (blank line) state = 1
1: /* leading blank line(s)
if (not blank line) {
nonblank = line
state = 2
}
2: /* saw non-blank line */
if (blank line) {
output noblank*
state = 0
} else {
state = 1
}
And we can translate this pretty directly to an awk program:
BEGIN {
state = 0; # start in state 0
}
state == 0 { # looking for a (leading) blank line
print;
if (length($0) == 0) { # found one
state = 1;
next;
}
}
state == 1 { # have a leading blank line
if (length($0) > 0) { # found a non-blank line
saved = $0; # save it
state = 2;
next;
} else {
print; # multiple leading blank lines (ok)
}
}
state == 2 { # saw the non-blank line
if (length($0) == 0) { # followed by a blank line
print saved "*"; # BINGO!
state = 1; # to the saw a blank-line state
} else { # nope, consecutive non-blank lines
print saved; # as-is
state = 0; # to the looking for a blank line state
}
print;
next;
}
END { # cleanup, might have something saved to show
if (state == 2) print saved;
}
This is not the shortest way, nor likely the fastest, but it's probably the most straightforward and easy to understand.
EDIT
Here is a comparison of Ed's way and mine (see the comments under his answer for context). I replicated the OP's input a million-fold and then timed the runnings:
# ls -l
total 22472
-rw-r--r--. 1 root root 111 Mar 13 18:16 ed.awk
-rw-r--r--. 1 root root 23000000 Mar 13 18:14 huge.in
-rw-r--r--. 1 root root 357 Mar 13 18:16 john.awk
# time awk -f john.awk < huge.in > /dev/null
2.934u 0.001s 0:02.95 99.3% 0+0k 112+0io 1pf+0w
# time awk -f ed.awk huge.in huge.in > /dev/null
14.217u 0.426s 0:14.65 99.8% 0+0k 272+0io 2pf+0w
His version took about 5 times as long, did twice as much I/O, and (not shown in this output) took 1400 times as much memory.
EDIT from Ed Morton:
For those of us unfamiliar with the output of whatever time command John used above, here's the 3rd-invocation results from the normal UNIX time program on cygwin/bash using GNU awk 4.1.3:
$ wc -l huge.in
1000000 huge.in
$ time awk -f john.awk huge.in > /dev/null
real 0m1.264s
user 0m1.232s
sys 0m0.030s
$ time awk -f ed.awk huge.in huge.in > /dev/null
real 0m1.638s
user 0m1.575s
sys 0m0.030s
so if you'd rather write 37 lines than 3 lines to save a third of a second on processing a million line file then John's answer is the right one for you.
EDIT#3
It's the standard "time" built-in from tcsh/csh. And even if you didn't recognize it, the output should be intuitively obvious. And yes, boys and girls, my solution can also be written as a short incomprehensible mess:
s == 0 { print; if (length($0) == 0) { s = 1; next; } }
s == 1 { if (length($0) > 0) { p = $0; s = 2; next; } else { print; } }
s == 2 { if (length($0) == 0) { print p "*"; s = 1; } else { print p; s = 0; } print; next; }
END { if (s == 2) print p; }

Answer 3

Here's a perl filter version, for the sake of illustration — hopefully it's clear to see how it works. It would be possible to write a version that has a lower input-output delay (2 lines instead of 3) but I don't think that's important.
my #lines;
while (<>) {
# Keep three lines in the buffer, print them as they fall out
push #lines, $_;
print shift #lines if #lines > 3;
# If a non-empty line occurs between two empty lines...
if (#lines == 3 && $lines[0] =~ /^$/ && $lines[2] =~ /^$/ && $lines[1] !~ /^$/) {
# place an asterisk at the end
$lines[1] =~ s/$/*/;
}
}
# Flush the buffer at EOF
print #lines;

Answer 4

A perl one-liner
perl -0777 -lne's/(?<=\n\n)(.*?)(\n\n)/$1\*$2/g; print' ol.txt
The -0777 "slurps" in the whole file, assigned to $_, on which the (global) substitution is run and which is then printed.
The lookbehind (?<=text) is needed for repeating patterns, [empty][line][empty][line][empty]. It is a "zero-width assertion" that only checks that the pattern is there without consuming it. That way the pattern stays available for next matches.
Such consecutive repeating patterns trip up the /(\n\n)(.*?)(\n\n)/$1$2\*$3/, posted initially, since the trailing \n\n are not considered for the start of the very next pattern, having been just matched.

Answer 5

Update: My solution also fails after two consecutive matches as described above and needs the same lookback: s/(?<=\n\n)(\w+)\n\n/\1\2*\n\n/mg;
The easiest way is to use multi-line match:
local $/; ## slurp mode
$file = <DATA>;
$file =~ s/\n\n(\w+)\n\n/\n\n\1*\n\n/mg;
printf $file;
__DATA__
foo
foo
foo
foo
foo

Answer 6

It's simplest and clearest to do this in 2 passes:
$ cat tst.awk
NR==FNR { nf[NR]=NF; nr=NR; next }
FNR>1 && FNR<nr && NF && !nf[FNR-1] && !nf[FNR+1] { $0 = $0 "*" }
{ print }
$ awk -f tst.awk file file
foo
foo
foo
foo*
foo
The above takes one pass to record the number of fields on each line (NF is zero for an empty line) and then the second pass just checks your requirements - the current line is not the first or last in the file, it is not empty and the lines before and after are empty.

Answer 7

alternative awk solution (single pass)
$ awk 'NR>2 && !pp && !NF {p=p"*"}
NR>1{print p}
{pp=length(p);p=$0}
END{print p}' foo
foo
foo
foo
foo*
foo
Explanation: defer printing to next line for decision making, so need to keep previous line in p and state of the second previous line in pp (length zero assumed to be empty). Do the bookkeeping assignments and at the end print the last line.