Re: Replication vs. float timestamps is a disaster

Stephen Frost <sfrost(at)snowman(dot)net> writes:
> * Tom Lane (tgl(at)sss(dot)pgh(dot)pa(dot)us) wrote:
>> While I'm generally not one to vote for dropping backwards-compatibility
>> features, I have to say that I find #4 the most attractive of these
>> options. It would result in getting rid of boatloads of under-tested
>> code, whereas #2 would really just add more, and #3 at best maintains
>> the status quo complexity-wise.

> +1.

BTW, I did a bit of digging in the archives, and found a couple of
previous discussions around this:

https://www.postgresql.org/message-id/flat/1178476313.18303.164.camel%40goldbach
https://www.postgresql.org/message-id/flat/1287334597.8516.372.camel%40jdavis

as well as numerous discussions of specific bugs associated with float
timestamps, eg this isn't even the first time we've hit a float-timestamp
replication bug:

https://www.postgresql.org/message-id/flat/CAHGQGwFbqTDBrw95jek6_RvYG2%3DE-6o0HOpbeEcP6oWHJTLkUw%40mail.gmail.com

In one or another of those threads, I opined that we'd have to keep float
timestamps available for as long as we were supporting pg_upgrade from
8.3. However, I think that that argument hasn't really stood the test of
time, because of the lack of packagers shipping builds with float
timestamps turned on. There may be some number of installations that
still have float timestamps active, but it's got to be a tiny number.
And the list of reasons not to like float timestamps is very long.

regards, tom lane