I wrote:
> Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
>> Seems reasonable. If we rip it out completely we'll have to find all the
>> places it breaks and fix them. And we'll almost certainly get new
>> breakage. If it's hiding a real bug we'll have to do that, but I'd be
>> reluctant unless there's actual proof.
> Hard to tell. What I propose to do right now is change the \r filters
> as shown above, and see if the test I added in 004_logrotate.pl starts
> to show failures on Windows. If it does, and no other place does,
> I'm willing to be satisfied with that. If we see *other* failures then
> that'd prove that the problem is real, no?
So I did that, and the first report is from bowerbird and it's still
green. Unless I'm completely misinterpreting what's happening (always
a possibility), that means we're still managing to remove "data"
occurrences of \r.
The most likely theory about that, I think, is that IPC::Run::run already
translated any \r\n occurrences in the psql command's output to plain \n.
Then, the \r generated by pg_current_logfile() would butt up against the
line-ending \n, allowing the "fix" in sub psql to remove valid data.
One thing I noticed while making 91bdf499b is that some of these
substitutions are conditioned on "if $TestLib::windows_os" while others
are conditioned on "if $Config{osname} eq 'msys'". What is the reason
for this difference? Is it possible that we only really need to do it
in the latter case?
regards, tom lane