Shell Programming and Scripting

I'm having a small problem with awk. I tried to use it for replacing ";" with <tab>. But I have to "touch" one variable to get the <tab> in OFS to be used.

Code:

$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print $0; }'
xx;yy;zz                # no OFS used
$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print $0; print $1,$2}'
xx;yy;zz
xx      yy              # only OFS used for "field for field"
$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print $0; print $1,$2; print $0}'
xx;yy;zz
xx      yy
xx;yy;zz                # still no tab for $0
$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print $0; print $1,$2; print $0; $1=$1; }'
xx;yy;zz                # tested to "touch" $1 at the end, no success
xx      yy
xx;yy;zz
$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print $0; print $1,$2; print $0; $1=$1; print $0}'
xx;yy;zz
xx      yy
xx;yy;zz
xx      yy      zz      # ta-dam! here is OFS used, AFTER the "touch"
# and here is a "funny" result, in the first line OFS is used, but not for $0:
$ echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { print "",$0; $1=$1; print "",$0 }'
        xx;yy;zz
        xx      yy      zz

I have googled but haven't find any information. Am I missing something? Or I have to do this:

Code:

echo "xx;yy;zz" | awk -F';' 'BEGIN { OFS="\t" } { $1=$1; print }'

Shell: bash, system: FreeBSD (is that correct info?)

Wouldn't it be better to use sed?

Code:

echo "xx;yy;zz" | sed 's/;/\t/g'

This goes round and round...
In this thread ... I was first thinking of an awk-solution, but because of the problem described here I suggested a sed-solution like yours - but you must use a "real" tab instead of "\t". The other thread is about that problem. Either you haven't tested your solution, or your system works different.

PS. The problem is not how to solve this, I want to know how awk works.

The original record, $0, is not modified by the awk field splitting process nor is it modified by setting the value of OFS. When you print $0, it is printed as is, regardless of the value of OFS.

OFS is used in two situations:
1. To separate the print statment's arguments.
2. As the field separator when $0 needs to be rebuilt, which happens when a field is assigned a value (this includes the use of the sub/gsub functions) or when you assign to a new field beyond the current value of NF.

The awk standard document: http://pubs.opengroup.org/onlinepubs...ities/awk.html

Regards,
Alister

Hi

The \t certainly works on Ubuntu using the bash shell

Whether \t is supported or not by sed depends only on the sed implementation itself. The shell is irrelevant. The operating system is irrelevant (except as an indication of which sed implementation is provided by default).

I believe that the only commonly-used sed that supports \t in either portion (regular expression or replacement text) of a substitution command is GNU sed.

According to the document standardizing sed (and the regular expression standard to which it refers), \t is an undefined sequence. When encountered, an implementation could treat it as a tab (as GNU sed does), it could treat it as a literal backslash followed by the letter tee (as just about every other sed does), it could abort with a syntax error, or it could choose to delete a file (though this is extremely unlikely ). The same is true of any sequence consisting of a backslash followed by an ordinary character, except for ...

and

Source of the 1st quote, which applies to all basic regular expressions: Regular Expressions
Source of the 2nd quote, which applies to sed regular expressions : sed

Long story short, if you use \t in a sed command, it will only work with GNU sed, which practically means Linux-only. Don't expect such scripts to run as intended on non-Linux platforms.

Regards,
Alister

Good to know