[bug-gawk] more on OFS

[Nelson H. F. Beebe]

The reported problem with setting of OFS may be more subtle.  Here is
a test script run on my system with current versions of three awk
implementations, and their side-by-side output:

        % cat foo.sh
        #! /bin/sh -

        awktest()
        {
          echo a:b:c | $AWK -F":" '{$2="x"; OFS=FS; print}'
          echo a:b:c | $AWK -F":" '{$2="x"; print; OFS=FS; print}'
          echo a:b:c | $AWK -F":" '{$2="x"; print $1; OFS=FS; print}'
          echo a:b:c | $AWK -F":" '{$2="x"; print; $2=$2; OFS=FS; print}'
        }

        AWK=gawk ; awktest > foo.gawk
        AWK=mawk ; awktest > foo.mawk
        AWK=nawk ; awktest > foo.nawk

Running the script produces three output files that can be viewed like
this, with my comments in a final column:

        ================================================
        gawk    mawk    nawk            comments
        ================================================
        a:x:c   a x c   a:x:c           1 = 3, 2 differs
        a x c   a x c   a x c           1 = 2 = 3
        a x c   a x c   a x c           1 = 2 = 3
        a       a       a               1 = 2 = 3
        a:x:c   a x c   a:x:c           1 = 3, 2 differs
        a x c   a x c   a x c           1 = 2 = 3
        a:x:c   a x c   a:x:c           1 = 3, 2 differs
        ================================================

Thus, my latest gawk, and nawk (version 20110810) agree, but mawk
differs in 3 of the 7 lines.  That suggests a gray area in the
language that hampers portability, and is perhaps best avoided.

Repeating the tests on about twenty platforms gave identical
results.

I then ran the test with 69 historical versions of gawk, back to 3.0.0
from 13-May-1996, and got identical output from all of them.

I personally rarely use OFS, in favor of using printf() to control my
output, but I agree that OFS can be useful for writing simple filters.

In any event, it would seem desirable to add some extra-vigorous tests
to the gawk test suite to ensure that the handling of OFS remains
stable, and in agreement with its documentation.

-------------------------------------------------------------------------------
- Nelson H. F. Beebe                    Tel: +1 801 581 5254                  -
- University of Utah                    FAX: +1 801 581 4148                  -
- Department of Mathematics, 110 LCB    Internet e-mail: address@hidden  -
- 155 S 1400 E RM 233                       address@hidden  address@hidden -
- Salt Lake City, UT 84112-0090, USA    URL: http://www.math.utah.edu/~beebe/ -
-------------------------------------------------------------------------------