Re: [HACKERS] removing abstime, reltime, tinterval.c, spi/timetravel

Andres Freund <andres@anarazel.de> writes:
> On 2018-10-12 19:47:40 -0400, Tom Lane wrote:
>> So I went looking for a different example to plug in there, and soon
>> found that there weren't any.  If you change all the physically_coercible
>> calls in that script to binary_coercible, its output doesn't change.
>> I'm thinking that we ought to do that, and just get rid of
>> physically_coercible(), so that we have a tighter, more semantically
>> meaningful set of checks here.  We can always undo that if we ever
>> have occasion to type-cheat like that again, but offhand I'm not sure
>> why we would do so.

> Hm, I wonder if it's not a good idea to leave the test there, or rewrite
> it slightly, so we have a a more precise warning about cheats like that?

After thinking about this for awhile, I decided that
physically_coercible() is poorly named, because it suggests that it
might for instance allow any 4-byte type to be cast to any other one.
Actually the additional cases it allows are just ones where an explicit
binary-coercion cast would be needed.  So I still think we should
tighten up the tests while we can, but I left that function in place
with a different name and a better comment.

            regards, tom lane