Make Interval Parsing More Permissive

[samuelcolvin]

Describe the bug

While working on apache/datafusion#12448 I noticed that the string 5 day hour is not interpreted correctly by parse_interval_month_day_nano_config, it should be considered the same as 5 day 0 hour, instead it return an error.

To Reproduce

run that function, or look at apache/datafusion#12448

Expected behavior

parse_interval_month_day_nano_config should match postgres behaviour in general

Additional context

Related to #6211 where I improved interval parsing a lot, although apparently not enough.

[alamb]

I think 5 day hour is a pretty uncommon usecase . Thank you for filing this

[samuelcolvin]

The most likely scenario where it could come up is select interval '5 day' hour, but I agree even that seems like either allowing a common mistake, or a side effect of their very permissive parser.

---

While we're talking about weird things postgres supports, as per apache/datafusion-sqlparser-rs#1345 (comment), the following are also supported:

select interval '1 month + 2 day'
select interval '1 minute + 2 s'  -- makes sense, all units are supported in arithmetic expressions
select interval '1 minute - 2 s'
select interval '1 minute + 2'  -- makes sense since raw numbers are interpreted as seconds
select interval '1 minute * 2s'  -- this is crazy! the "*" operator is interpretted as "+" so this evaluates as 0:01:02
select interval '1 minute * 2' -- same! 
select interval '1 minute minute' -- repeated units are ignored
select interval '1 month + 2 day' day -- same thing with the units just concatenated

[ByteBaker]

@alamb do we have a plan on integrating this inside Arrow?

[alamb]

@alamb do we have a plan on integrating this inside Arrow?

I am not sure what you mean by this question @ByteBaker

[ByteBaker]

I meant to ask if we plan to handle more cases in our parsing logic? To be closer to parsing abilities of Postgres? As @samuelcolvin pointed out above.

[alamb]

I meant to ask if we plan to handle more cases in our parsing logic? To be closer to parsing abilities of Postgres? As @samuelcolvin pointed out above.

I guess in general I think it would be driven by contributor needs -- if someone wants needs more exotic parsing support, we can definitely consider it, but if it gets too specific, it probably makes sense to keep the parsing in the application and not in arrow-rs

[ByteBaker]

In that case we need to decide if we wanna keep this one open or not.

[alamb]

Let's leave it open so it is easier to find if someone else hits the problem