How do i move a pieces of data between two words if the data has a % percent symbol at the begining?

Question

I have a log file looking somtheing like :

SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-RTP-Trust
                               ^
% Invalid input detected at '^' marker.

SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-Control-Trust
                               ^
% Invalid input detected at '^' marker.

SWEs-elmPCI-A-01(config)#
 policy-map AutoQoS-Police-CiscoPhone
SWEs-elmPCI-A-01(config-pmap)#

Now i have to look for all the lines begining with % symbol between the lines containing hash # and put it into another error log file.

if you consider the above then this should go into error log file.

   SWEs-elmPCI-A-01(config)#
     class AutoQoS-VoIP-RTP-Trust
                                   ^
    % Invalid input detected at '^' marker.

    SWEs-elmPCI-A-01(config)#
     class AutoQoS-VoIP-Control-Trust
                                   ^
    % Invalid input detected at '^' marker.

    SWEs-elmPCI-A-01(config)#

Answer 1

Not sure if this is what you want, but:

#!/bin/bash

awk '
BEGIN {
    err=0
}
/^SWE.*#$/ {
    if (err) {
        printf "%s\n%s", txt, $0
        txt=""
        err=0
    } else {
        txt=$0
    }
    next
}
/^% Invalid/ {
    err=1
    txt=txt "\n" $0
    next
}
{
    txt=txt "\n" $0
}
END {
    print ""
}
' "$1"

Result:

SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-RTP-Trust
                               ^
% Invalid input detected at '^' marker.

SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-Control-Trust
                               ^
% Invalid input detected at '^' marker.

SWEs-elmPCI-A-01(config)#

---

As an alternative, how I tought it was with extra lines to group:

awk '
BEGIN {
    i = 0
    sep="------------------------------------------------"
}
/^% Invalid/ {
    printf "%s %3d\n%s\n%s\n%s\n",
    sep, ++i, txt, $0, sep
    txt=""
    next
}
/^SWE.*#$/ {

    txt=$0
    next
}
txt != "" {
    txt=txt "\n" $0
}
' "$1"

Result:

------------------------------------------------   1
SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-RTP-Trust
                               ^
% Invalid input detected at '^' marker.
------------------------------------------------
------------------------------------------------   2
SWEs-elmPCI-A-01(config)#
 class AutoQoS-VoIP-Control-Trust
                               ^
% Invalid input detected at '^' marker.
------------------------------------------------

Answer 2

Maybe something like:

sed 'H;/#$/!d;s/^/\
/;x;/\n%/!d;s/.//;x;s/.*//;x'

Since you want both the prompt before the error and after the error, that one includes both except for prompts that would otherwise appear both as the prompt after one error and the one before the next error to match your sample output.

It's a lot simpler if we include both prompts regardless of whether errors follow each other or not:

sed 'H;/#$/!d;x;/\n%/!d'

If you wanted only the prompt before the error, that would also be a lot simpler:

sed '/#$/!{H;d;}
     x;/\n%/!d'

---

Probably the best way to explain those is translate to a more verbose language like perl. In sed, the hold space is like a static variable that is initialised with an empty line, and the pattern space a variable which sed assigns in turn to every line of input. Something like:

$hold_space = "\n";
LINE: while ($pattern_space = <>) {

  <sed commands go here>

  print $pattern_space; # unless -n option is passed
}

The H command translates to:

$hold_space .= $pattern_space;

The d command translates to:

next LINE;

That is stop executing the commands, discard the pattern space and start fresh with the next line of input.

x swaps pattern and hold space:

($pattern_space, $hold_space) = ($hold_space, $pattern_space);

So that last sed command translates to:

$hold_space = "\n";
LINE: while ($pattern_space = <>) {

  if (! ($pattern_space =~ /#$/)) { # anything other than a prompt
    $hold_space .= $pattern_space;
    next LINE;
  }

  ($pattern_space, $hold_space) = ($hold_space, $pattern_space);

  if (! ($pattern_space =~ /\n%/)) {
    next LINE;
  }

  print $pattern_space;
}

Which can be re-written in a more legible fashion as:

LINE: while ($pattern_space = <>) {

  if ($pattern_space =~ /#$/)) {
    if ($hold_space =~ /\n%/)) {
      print $hold_space;
    }
    $hold_space = $pattern_space;
  } else {
    $hold_space .= $pattern_space;
  }
}

Rename $pattern_space to $line and $hold_space to $current_block and it becomes even more legible.