Shell Programming and Scripting

OK. This set of requirements is MUCH different than I expected from your original problem statement: "I want to get all strings that starts with '+' and ends with '/'. Then I want the strings to be separated by ' + '."

Let me see if I understand what you want by stating rules that will outline the logic of an algorithm to process your input data:

If an input line does not contain a '/' character, skip to the next line. In this case no output will be produced for that line.

Otherwise the following steps shall be performed in sequence:Is this what you want? If not, please tell me what I have wrong.

Yes, precisely.

For this:

Code:

+$/ABBREV+ 
+$A$/NOUN
$A$/NOUN+At/NSUFF_FEM_PL+K/CASE_INDEF_ACC+
$A$/NOUN+At/NSUFF_FEM_PL+K/CASE_INDEF_GEN

Output shall be this:

Code:

$
$A$
$A$ + At + K 
$A$ + At + K

Thanks a lot!

OK, good. Here are a couple of ways to do it:

Code:

sed '/^[^\/]*$/d;s|/[^+]*+|+|g;s|/.*$||g;s/^+//;s/+$//;s/+/ + /g' input_file

and

Code:

awk '
index($0, "/") {                # for any line containing a "/"
        gsub("/([^+]*|$)", "")  # Remove all "/*" up to next "+" or EOL
        gsub(/(^[+]|[+]$)/, "") # Remove leading and trailing "+"
        gsub(/[+]/, " + ")      # Replace all "+" with " + "
        print                   # Print the modified line
}' input_file

If you are using a Solaris system, use /usr/xpg4/bin/awk or nawk instead of awk.

Another sed approach (not sure if it's standard sed!):

Code:

$ sed '/\//!d; s:/[^+]*+*: + :g; s:^+\|+ *$::g;' file
$ 
$A$ 
$A$ + At + K 
$A$ + At + K

It's almost standard sed. The last sed command here s:^+\|+ *$::g uses a non-standard extension supplied by some implementations of sed to make | behave as it does in an ERE. This behavior violates the standards. In a standards conforming sed this BRE would only match a line that consisted solely of a <plus_sign>, a <vertical_bar>, and a <plus_sign> followed by zero or more <space>s. On OS X (and, I expect, on several other sed implementations) this script, therefore, produces the output:

Code:

+$ + 
+$A$ + 
$A$ + At + K + 
$A$ + At + K +

instead of the desired output.

It also handles strings containing adjacent '+'s in input lines differently, but I'm not sure that that would make any difference for the input that would be expected here.

Your use of /\//!d is much easier to read than the /^[^\/]*$/d I used and serves exactly the same purpose (delete any line that does not contain a '/').

The use of +* at the end of the search pattern serves to find 0 (at end of line) or n + in order to simplify the code. Should undesired side effect show up, we needed to resort to a standard solution as you proposed. In fact, that might be easier to read than a complex ERE...

In fact now I got

Code:

$ +  
$A$ +  
$A$ + At + K
$A$ + At + i

Is there a way to get rid of the + as the last token? And get something like:

Code:

$
$A$
$A$ + At + K
$A$ + At + i