Re: Using getline into a variable from a coprocess sometimes overwrites RT

[Aharon Robbins]

OK, I added a footnote. It'll be in CVS shortly.

Thanks,

Arnold

> Date: Sun, 4 Oct 2009 14:59:49 -0700 (PDT)
> From: Nick Hobson <address@hidden>
> Subject: Re: Using getline into a variable from a coprocess sometimes 
> overwrites RT
> To: address@hidden
>
> Ah, I was assuming that getline into a variable would split lines on 
> newline, but it makes sense that gawk should use the actual value 
> of RS!  (Of course, if necessary you can set RS = "\n" before the 
> getline and reset it afterwards.)
>
> Maybe section 3.8 (e.g., 3.8.10) of the user guide could be changed 
> to mention that when RS is a regexp, getline might change RT?  At 
> the moment 3.8.8 says: "In this version of getline, none of the 
> built-in variables are changed and the record is not split into 
> fields. The only variable changed is var."
>
> Thanks,
> Nick
>
>
> Greetings. Re this:
>
> > Date: Fri, 2 Oct 2009 17:21:22 -0700 (PDT)
> > From: Nick Hobson <address@hidden>
> > Subject: Using getline into a variable from a coprocess sometimes 
> > overwrites 
> > RT
> > To: address@hidden
> >
> > Hi,
> >
> > I think I've found a gawk bug.  In some cases, using getline into a
> > variable from a coprocess clobbers RT.  It doesn't touch $0 or NF.
> > I have demonstrated the problem in gawk 3.1.7 under Arch Linux.
> >
> > The test program is:
> >
> > BEGIN {RS = "u"}
> >
> > RT == "u" {
> >     printf callpgm("xargs file -b ", "imgs/auk.png")
> > }
> >
> > function callpgm(ext, src,  res) {
> >     print src |& ext
> >     close(ext, "to")
> >     print "1", $0, NF, RT
> >     ext |& getline res
> >     print "2", $0, NF, RT
> >     close(ext)
> >     return res
> > }
> >
> > The input data is:
> >
> > bug
> >
> > When run as echo 'bug' | bug.awk, the expected output is:
> >
> > 1 b 1 u
> > 2 b 1 u
> > PNG image, 500 x 600, 8-bit colormap, non-interlaced
> >
> > The actual output is:
> >
> > 1 b 1 u
> > 2 b 1 
> > PNG image, 500 x 600, 8-bit colormap, non-interlaced
> >
> > The getline sets RT to the empty string.  (The actual value RT is
> > overwritten with varies.  If you change the above BEGIN block to RS =
> > "[aeiou]", RT gets overwritten with "i".)
>
> This is not a bug.  RT is set to match the text of whatever input
> characters matched RS.  In the case of the getline, there were no
> "u" characters in the output from the coprocess, so RT is set to the
> null string.  When you set RS = "[aeiou]" then "i" is the first of
> those characters found in the output from `file', so that's what RT
> is set to.
>
> Since your getline used a variable, the fields and NF are not changed.
>
> Consider this simpler example:
>
>         $ echo foo | gawk 'BEGIN { RS = "x" }                
>         > { printf "RT = <%s>\n", RT ; print NF, $0 }'
>         RT = <>
>         1 foo
>
>         $
>
> Here, no text matched "x", so RT is empty, and the print $0 includes
> the original newline that was in the input, as well as the value of
> ORS that print automatically adds.
>
> This stuff can be subtle. :-)  In this example, at least, gawk is
> working correctly.
>
> Thanks for taking the time to report a possible bug, I appreciate it,
> even when it's not really a bug. :-)
>
> Arnold