3 ms·

My Favorite PostgreSQL 18 Feature: Virtual Generated Columns

54 points by fforflo 1y ago

saltcured 1y ago
Aside from the SQL syntax, what is the difference between this and the longstanding virtual column feature, where you can define a function that takes the table's record type as input and return a scalar? PostgreSQL already let you use table.funcname as a virtual column, as I recall.
- mdaniel 1y ago
  It's possible the following argument is what you meant by "from the SQL syntax," and if so apologies for the noise At least one demonstrable difference I can see between your recollection and this feature is that the column name being part of the DDL pushes the logic down into the DDL, versus every client having to carry that magic expression around with them. I think it's likely the age-old application-logic-client-side versus stored-proc camps db=# select id, cast(length(name) AS text)||'0' as silly from my_table as compared to <<select id, silly from my_table>> for the consumer
  - saltcured 1y ago
    The virtual column feature I remember is halfway in between. It is not part of the DDL nor reported table schema, so the developer needs to know about it. But if a function called silly with input type my_table existed, you could still do your second query. It has syntactic sugar for making a bare function name implicitly invoke on the current row record. I can't remember if that sugar could also be used in an expression index. I never really used this syntactic feature, as I thought being more explicit with function calls would be better coding style.
- sharlos201068 1y ago
  That's still not a column and isn't included with things like select *
sgarland 1y ago
Welp, that’s it. That was the last thing MySQL had that Postgres didn’t (modulo clustering index, which OrioleDB does, soooo…). Can’t wait!
icedchai 1y ago
In the past, I've used views to add "virtual columns" where I didn't want to hard code them into the application.
garyclarke27 1y ago
Stored generated columns would be much more useful if they could reference other rows and tables, they are limited to only the current row. Automatic incremental view maintenance would also solve this problem, but is unfortunately still a glaring omission in Postgres capabilities.
_el1s7 1y ago
I don't really find these kind of features necessary. You can, and probably should, do these things in your application code. I mean, it's not really that hard to make a string lowercase, you don't need to define a "virtual colum" for that...
- combatentropy 1y ago
  You're right, it's not that hard. But why would you rather do it there?

54 points by fforflo 1y ago

Aside from the SQL syntax, what is the difference between this and the longstanding virtual column feature, where you can define a function that takes the table's record type as input and return a scalar? PostgreSQL already let you use table.funcname as a virtual column, as I recall.

mdaniel 1y ago
It's possible the following argument is what you meant by "from the SQL syntax," and if so apologies for the noise At least one demonstrable difference I can see between your recollection and this feature is that the column name being part of the DDL pushes the logic down into the DDL, versus every client having to carry that magic expression around with them. I think it's likely the age-old application-logic-client-side versus stored-proc camps db=# select id, cast(length(name) AS text)||'0' as silly from my_table as compared to <<select id, silly from my_table>> for the consumer
- saltcured 1y ago
  The virtual column feature I remember is halfway in between. It is not part of the DDL nor reported table schema, so the developer needs to know about it. But if a function called silly with input type my_table existed, you could still do your second query. It has syntactic sugar for making a bare function name implicitly invoke on the current row record. I can't remember if that sugar could also be used in an expression index. I never really used this syntactic feature, as I thought being more explicit with function calls would be better coding style.
sharlos201068 1y ago
That's still not a column and isn't included with things like select *

sgarland 1y ago

Welp, that’s it. That was the last thing MySQL had that Postgres didn’t (modulo clustering index, which OrioleDB does, soooo…). Can’t wait!

icedchai 1y ago

In the past, I've used views to add "virtual columns" where I didn't want to hard code them into the application.

garyclarke27 1y ago

Stored generated columns would be much more useful if they could reference other rows and tables, they are limited to only the current row. Automatic incremental view maintenance would also solve this problem, but is unfortunately still a glaring omission in Postgres capabilities.

_el1s7 1y ago

I don't really find these kind of features necessary. You can, and probably should, do these things in your application code. I mean, it's not really that hard to make a string lowercase, you don't need to define a "virtual colum" for that...

combatentropy 1y ago
You're right, it's not that hard. But why would you rather do it there?