[BUG]: Extremely slow intellisense depending on schema size and amount

[ESchouten]

What version of drizzle-orm are you using?

0.27.0

What version of drizzle-kit are you using?

0.19.1

Describe the Bug

Intellisense gets unusably slow when providing the schema in the drizzle(sql, { schema }) instantiation.
The intellisense delay depends on the schema amount size, it approaches 8 seconds for us on a Macbook Pro M2 Max with 40 tables, with other, non Macbook M processor people are reporting 10-15 second delays.
It gets faster again when we only provide a subset of the schemas.

Expected behavior

Instant/fast intellisense.

I suppose the TS types are calculated on the go when intellisense activates? This might be the bottleneck. Other libraries generate types and store them on the filesystem. Is this an option for Drizzle?

Environment & setup

Macbook Pro M2 Max

[atanaskanchev]

+1 VSCode gets bogged down and often I have to restart the extension host to get it to load the types

[romulovalez]

+1 With 32 tables and more or less 30 columns per table it takes 14s to fire intellisense, when it used to take 2-3 seconds without drizzle, I measured it using npx tsc --extendedDiagnostics --incremental false --noEmit

Using the trick of casting the column names as string (ex.: integer('id' as string)) reduces the time to 5 seconds and only generates 40,000 new types (this would still be very high, wouldn't it?), without this trick the types count raised to +500,000 and it takes 14 seconds to load intellisense, as i said above.

[romulovalez]

Related issue: #437

[ESchouten]

Update from the Drizzle team: Tweet
"We'll make it a priority on Jul"

[oscarhermoso]

Nice, the column name as string workaround brings my tsc down from 4.9 seconds to 3.1 seconds (was 2.8s before adding Drizzle).

Regex find and replace (^.+?'\w+') with $1 as string in my schema directory patched 99% of the column names.

[MAST1999]

Nice, the column name as string workaround brings my tsc down from 4.9 seconds to 3.1 seconds (was 2.8s before adding Drizzle).

Regex find and replace (^.+?'\w+') with $1 as string in my schema directory patched 99% of the column names.

This actually does help!
But when I was reading it I was a bit confused so here's how I think it works in simple terms.

The issue is that drizzle orm is generating a lot of types by inferring things and most of that inferring depends on the column names. So if we add as string to the column names which otherwise would be a literal string it would reduce the number of types generated.

export const ticket = pgTable(
  'ticket',
  {
    id: uuid('id' as string).primaryKey().notNull(),
  });

I ended up using (^\s\s.+?'\w+') because I had 2 spacing in my tables (thanks to prettier) and this correctly selected all my columns.

[jeffminsungkim]

Any updates? This is very annoying tbh.

[ngregrichardson]

Seems like it'll be released sometime in August according to this thread on Discord

[jeffminsungkim]

Seems like it'll be released sometime in August according to this thread on Discord

Thanks for letting me know 👍🏼