Worried about performance about phpstan-dba HOT 2 CLOSED

Hey Ondrej,

thx for the feedback.

as it stands right now we are asking the database, because this gives us the ability to have a result-set schema, not just a database schema. with this metadata we are able to also infer types for aggregate functions and we also get the nullable-types on joins for free without the need to parse the actual database query.

Sometimes the database doesn't even live on the same machine where PHPStan is running.

for this case phpstan-dba supports record & replay. when replay-ing the queries we also see a 2-2.5x improved performance.

note, recording has a open issue though

I think it'd be much more interesting to download the database structure before the analysis

as mentioned above, this would not bring us on feature parity with what we have atm - I guess.

maybe we could use something like https://github.com/vimeo/php-mysql-engine to simulate a real database. but this in turn might bring us in a similar direction we already have with record&replay without additional benefits?
since there is no such engine I am aware of for other database types (e.g. pgsql) this would render phpstan-dba even more coupled to mysql which is not the best direction I think.

these are just my thoughts though, and I am open for experiements to improve phpstan-dba performance if we can.

from phpstan-dba.

Yeah, "record and replay" seems reasonable, thanks :)

from phpstan-dba.