Shell Programming and Scripting

No sed, but Ruby(1.9+)

Code:

$ ruby -0777  -ne 'i=$_.rindex("AAA")-1; puts $_[0..i-1] + "\nEEEEE" + $_[i+"AAA".size+1..-1]' file
AAAaaa
BBBbbb
CCCccc
EEEEEeee
DDDddd

Code:

printf "G?AAA\n:s/AAA/EEEEE/g\n:wq\n" | vi -s infile >/dev/null

Hi, ctsgnb:

Regarding that usage of vi, you can't count on that working. Some (most?) vi clones will not read from stdin when it's not a terminal. The standard explicitly mentions this case and leaves it as undefined behavior. Also, it does not include a -s option for vi.

Even if it did, the sequence of commands is faulty. G makes the last line the current line and ?AAA then searches backward for "AAA". What if there's an "AAA" on that last line in the file and another "AAA" on a previous line? The search will not see that the current line is a match and will instead stop at the previous, earlier line, which is not what's desired. The backwards search needs to begin from line 1.

(The following is nothing but a stylistic note of personal preference. It's not really important but perhaps you may like it.)

Since printf reuses the format string until it has exhausted all of its arguments, you can rewrite it so that printf newline-terminates each string argument for you. It's easier to read when the embedded \n sequences are replaced with whitespace-delimited arguments. Additionally, should the ed/ex commands use %, \n, \t, etc..., there's no need to escape them (as there would be in the format string).

Code:

# Works fine but harder to read and vulnerable to metacharacters
printf "G?AAA\n:s/AAA/EEEEE/g\n:wq\n"

# Easier to read and no worries regarding format string metacharacters
printf '%s\n' 'G?AAA' :s/AAA/EEEEE/g :wq

Regards,
Alister

with gnu awk:

Code:

gawk '{P=P $0 "\n"} END{printf gensub("(.*)AAA","\\1EEEE","",P)}' file

This works for last occurance on last line too eg (red text replaced):

Code:

AAAaaa
BBBbbb
CCCccc
AAAbbbbAAAAAAcccc
DDDddd

If your not worried about this case, no need for gnu awk:

Code:

awk '/AAA/{printf P;P=""}{P=P $0 "\n"} END{sub("AAA","EEEE",P); printf P}' file

@alister

First of all, thanks a lot for your comments, i really appreciate their quality.

As I am not very familiar with 'ed' usage, if you could show me the solution you would suggest for this thread as well as if you would point me a good link about it, i would really enjoy it.

Thanks again for your time sharing knowledge

if you have tac installed, do this:

Code:

tac yourFile |sed '0,/AAA/{s/AAA/EEEE/}'|tac

Hi, ctsgnb:

My original post suggested ed, but ex is also a good option. If you already know vi, then you probably know quite a few ex commands (the colon commands in vi, but in ex you don't use the colon). Perhaps the following:

Code:

printf '%s\n' 1 '?AAA?s//EEEEE/' wq | ex -s file

ed is simpler than ex and behaves a little differently. In ed, entering nothing but a line number to make that line the current line also prints it (ex does not), so a redirect to /dev/null is probably a good idea to prevent unwanted text on stdout. Also, while ex always supports wq to write and quit, some eds do and some dont. The standard doesn't. So w and q should be specified as separate commands. Using ed:

Code:

printf '%s\n' 1 '?AAA?s//EEEEE/' w q | ed -s file > /dev/null

In ed/sed/ex, an empty regular expression is shorthand for the last regular expression used (whether in a previous substitution command or in an address).

There's also the heredoc option, of course, which is often the easiest of all to read (even though sometimes we all have an illogical inclination to squeeze everything into one line):

Code:

ed -s file >/dev/null <<'EOED'
1
?AAA?s//EEEEE/
w
q
EOED

As for a good source of info, I think the man pages and a terminal to play with is sufficient. ed is pretty simple. Off the top of my head, I'd say there are only about 15 commands to learn.

ed

Enjoy your weekend,
Alister