Re: awk inplace overrides symlinks

[Ed Morton]

Not saying gawk needs to behave the same way of course but FWIW sed -i 
and perl -i both also replace the link with a regular file:

   $ rm link; echo 'foo' > file; ln -s file link; cat link; ls -l file link
   foo
   -rw-r--r-- 1 name name 4 Feb  9 08:28 file
   lrwxrwxrwx 1 name name 4 Feb  9 08:28 link -> file

   $ sed -i 's/foo/bar/' link

   $ cat link; ls -l file link
   bar
   -rw-r--r-- 1 name name 4 Feb  9 08:28 file
   -rw-r--r-- 1 name name 4 Feb  9 08:28 link

   $ rm link; echo 'foo' > file; ln -s file link; cat link; ls -l file link
   foo
   -rw-r--r-- 1 name name 4 Feb  9 08:28 file
   lrwxrwxrwx 1 name name 4 Feb  9 08:28 link -> file

   $ perl -pi -e 's/foo/bar/' link

   $ cat link; ls -l file link
   bar
   -rw-r--r-- 1 name name 4 Feb  9 08:28 file
   -rw-r--r-- 1 name name 4 Feb  9 08:28 link

so maybe there's something to be said for consistency across the various 
tools?

Regards,

    Ed.

On 2/9/2024 7:47 AM, Andrew J. Schorr wrote:
> On Fri, Feb 09, 2024 at 05:21:05AM +0000, Avid Seeker wrote:
> > On Friday, February 9th, 2024 at 02:16, Andrew J. 
> > Schorr<aschorr@telemetry-investments.com>  wrote:
> >
> > > Hi,
> > >
> > > On Fri, Feb 09, 2024 at 12:28:54AM +0000, Avid Seeker wrote:
> > >
> > > > `touch file ln -s file link awk -i inplace '{print "newline"}' link`
> > > >
> > > > Results in replacing `link` with a regular file.
> > >
> > > Thanks for the bug report. The code currently calls stat() to check whether
> > > it's working on a regular file, and craps out if it isn't. So arguably we
> > > should be using lstat() and failing in this case instead of stomping on the
> > > symlink. Or do you expect the code to chase the link and replace the symlink
> > > destination file? I think the latter is probably too much to ask...
> > >
> > > Regards,
> > > Andy
> > It's more expected for the program edits the original file. Consider how
> > editors like emacs don't complain when editing a symlink. In the end, the
> > symlink chasing is done by `readlink` function.
> It could be a recursive symlink, so we'd actually need to use
> realpath, not readlink. But it's not quite clear to me that this
> behavior is within the scope of what the inplace extension is
> supposed to do. One could debate it...
>
> Regards,
> Andy