On Tue, Mar 31, 2020 at 6:36 AM Eric Blake <eblake@redhat.com> wrote:
> On 3/31/20 2:00 AM, Oğuz wrote:
> > Thanks for the reply. This might not be a bug though; I sent a similar mail
> > (https://www.mail-archive.com/austin-group-l@opengroup.org/msg05881.html)
> > to Austin Group mailing list asking what's the expected behavior in this
> > case, and I was told (
> > https://www.mail-archive.com/austin-group-l@opengroup.org/msg05891.html)
> > both behaviors -yielding n or empty line- are correct and standard should
> > *probably* be amended to explicitly state that this is unspecified. And
> > apparently (
> > https://www.mail-archive.com/austin-group-l@opengroup.org/msg05893.html)
> > some other UNIXes adopted the same practice as GNU sed (or vice versa, I
> > don't know which one is older).
>
> The POSIX folks will probably declare that use of a \X sequence (for
> arbitrary X; 'n', 't', '1', and probably others all fit this category)
> inside a regex delimited by X is unspecified behavior.  But that still
> doesn't stop us from fixing GNU set to at least be consistent - we
> should either blindly declare that \X represents the special meaning of
> X when such a meaning is present regardless of X also being the regex
> delimiter (our current \n behavior - no way to represent the delimiter
> as a literal match), or that use of X as a delimiter renders the special
> meaning of \X useless for that regex (our \t behavior - no way to
> represent the special behavior as part of the match).  My personal
> preference is making things consistent to our \t behavior.
>
> >> In the code, the "match_slash" function [1] is used to find
> >> the delimiters of the "s" command (typically "slashes").
> >> Special handling happens if a slash is found [2],
> >> And in lines 557-8 there's this conditional:
> >>
> >>                else if (ch == 'n' && regex)
> >>                  ch = '\n';
> >>
> >> Which forces any "\n" to be a new-line, regardless if the
> >> delimiter itself was an "n".
> >>
>
> >> Interestingly, removing these two lines does not cause
> >> any test failures, so this might be easy to fix without causing
> >> any regressions.
> >>
> >>
> >> For now I'm leaving this item open until we decide how to deal with it.
>
> I'm thus in favor of removing that special-case of 'n'.

Thank you all. Sorry it's taken so long.
I expect to push the following tomorrow.