Rails - storing a date when the day is optional - sql

Question

I'm collecting a date to store in the database but have to account for the fact that the user may not know the exact date. I want to offer the option to only enter the month and year. What would be the best way to store these values on the database?
If I use a Date object without the day and then store it as a date in the database then it defaults to the 1st of the month, which would be inaccurate. I had the idea of potentially storing the day, month and year separately as integers and then having a method on the model that returned a Date object and whether or not the day was accurate (ie had been inputted by the user and not just defaulted by the system).
It seems a little messy though, is there a better way to do this?
Thanks.

Answer 1

There are multiple solutions available. Choose one which serves your use-cases the best:
1. Individual date-part fields
You've already experimented with this idea. It seems something like this:
create table t(
year int,
month int,
day int
)
A month of 2017-03 could be represented with (2017, 3, NULL).
Note that all fields are NULL-able. You can make year NOT NULL if you want to some information at least.
With this model, you must use client logic to construct some kind of date-like object for further use.
The big disadvantage of this model is that it is hard to index. You could index f.ex. make_date(year, coalesce(month, 1), coalesce(day, 1)) but using it in queries is rather inconvenient. Also, to disallow some value compositions, which make no sense (f.ex. given a year and a day, but not a month), you should add a (really long) CHECK constraint too, f.ex.
CHECK (CASE
WHEN year IS NULL THEN month IS NULL AND day IS NULL
ELSE CASE WHEN month IS NULL THEN day IS NULL END
END)
2. Sample date and precision
create table t(
sample_date date,
sample_precision date_precision -- enum of f.ex. 'year', 'month', 'day'
)
A month of 2017-03 could be represented with ('2017-03-28', 'month').
This doesn't require long CHECK constraints, but it is fairly hard to select by dates if sample_date is truly just a sample (f.ex. when the whole month of 2017-03 should be represented in a row, a sample date could even be 2017-03-28). When you use the first date as sample_date (from the values it can take, based on sample_precision) things will get slightly easier. But then the following CHECK constraint would be needed for integrity:
CHECK (date_trunc(sample_precision::text, sample_date)::date = sample_date)
(More on improving this further, later.)
3. Possible range
You can store a possible range of dates. With either a possible_start and possible_end or with PostgreSQL's daterange type.
create table t(
possible_start date,
possible_end date,
-- or
possible_range daterange
)
A month of 2017-03 could be represented with ('2017-03-01', '2017-03-31').
In this model when possible_start = possible_end then the date value is exact. You could query two different things now:
Which rows happening around given date(s) for sure (contains)
Which rows possibly happening around given date(s) (intersects)
Both of these types of queries can use indexes with daterange.
The beauty of this is that you are not limited to month ranges. You can use literally any length of ranges. Its only drawback is that the range must be contiguous.
2. + 3. ?
There is a variant, which has all of 3.'s advantages, but looks like 2. with the interval type:
create table t(
possible_start date,
possible_length interval day
)
A month of 2017-03 could be represented with ('2017-03-01', '1 month').
(The day qualifier restricts the minimum precision of the interval to be a day. It is not required for timestamp or timestamptz based solutions.)
The last possible date could be represented with (possible_start + possible_length - interval '1 day')::date. Or, the whole range as (for a daterange index): daterange(possible_start, (possible_start + possible_length)::date) (ranges are implicitly exclusive on their end).
http://rextester.com/AWIO2403

Answer 2

you can store a date in one column and precision in other, to compare values, you can use date_trunc(precision_column,DATE), eg:
t=# create table so36(d date,p text);
CREATE TABLE
t=# insert into so36 select now(),'day';
INSERT 0 1
t=# insert into so36 select '2017-03-01','month';
INSERT 0 1
t=# select *,date_trunc(p,d),date_trunc(p,now()) from so36;
d | p | date_trunc | date_trunc
------------+-------+------------------------+------------------------
2017-03-28 | day | 2017-03-28 00:00:00+00 | 2017-03-28 00:00:00+00
2017-03-01 | month | 2017-03-01 00:00:00+00 | 2017-03-01 00:00:00+00
(2 rows)