While functions for getting and setting via json might be useful, I think they are peripheral to the core datatype problem here.
The interface to dynamic columns in Maria is SQL. The internal datatypes are SQLish. We can be completely sure of what was set, in SQL terms, by a COLUMN_CREATE(). But if we use a JSON getter then we might get something different because JSON datatypes don't match SQL datatypes. Sometimes it's no problem but not in general -- depends on the datatypes involved and the application's tolerance of differences.
What's missing is a getter that tells us the value and datatype of a dynamic column, a getter that tells us what was set.
What if we had a getter that returns a string like:
COLUMN_GET_SQL(`product`, 'price')
>> "CAST(123.456 AS DECIMAL(11,3))"
And Maria promises us that we could use that expression to write the value back and the result will be the exact same thing. A sort of idempotency promise (kinda).
This doesn't make the interface any less clunky (on the contrary) but I have resigned myself to a clunky interface. The only alternative is to forget all this and start something different. SQL is an insanely clunky old language and that's why we use some kind of abstraction software to make it nicer. If we're going to use SQL we may as well try to use what's there.
Something like this COLUMN_GET_SQL() would at least allow the application the possibility to get back what it put in.
Tom
@Roberto. A comment to "
i think it's a nice human readable format for arrays/objects, nothing less nothing more".
But why will you then need to *store* as JSON? Could not something like "SELECT .. INTO JSON ..." do the trick?
(with hindsight: "SELECT .. INTO XML ..." could have been implemented in MySQL when other XML functionalities were implemented but never was, it seems. You can mysqldump to XML but not "SELECT .. INTO XML ..." - using OUTFILE or not - refer http://dev.mysql.com/doc/refman/5.6/en/select-into.html)
-- Peter