Most of us probably do the same thing when we read a HN comment about something specific: "This rando seems to know what they're talking about. I'll assume it as fact until I encounter otherwise."
Not doing this might actually cause bigger problems... Getting first-hand experience or even reputable knowledge about something is extremely expensive compared to gut-checking random info you come across. So the "cheap knowledge" may be worth it on balance.
I wish the source citing was more explicit. It would be great if the AI summary said something like, “almost no info about xyz can be found online but one GitHub comment says abc” (link)
Instead it often frames the answer as authoritative
In case you find it interesting: I deployed an early version of a "lesson administering" bot deployed on a college campus that guides students through tutored activities of content curated by a professor in the "study mode" style -- that is, forcing them to think for themselves. We saw an immediate student performance gain on exams of about 1 stdev in the course. So with the right material and right prompting, things are looking promising.
Addressing your core question: Drizzle is a great ORM with a tastefully designed API—it's clearly a product of love. But it’s still an ORM, and it’s confined by certain design boundaries that come from being a library. For example, what if you want to use TypeScript, Go, and Python on your backend? Do you run three ORMs, each with different APIs? With Gel, you have one data model and a unified querying layer—the true source of truth.
We have a blog post about that and more [1].
---
By the way, if you visit Drizzle’s website, you’ll see that Gel is one of their biggest sponsors. We worked closely with Drizzle to ship a first-class integration with Gel. You can use Gel’s schema and migrations, and Drizzle will just work. You can even use the Drizzle query builder and EdgeQL side by side if you want.
> Do you run three ORMs, each with different APIs?
Yes and you absolutely have to. That's not a disadvantage, it's just how it is. Because SQL is the absolute bare minimum. The lowest common standard. And not a great one. Null handling and typesystem for example are way inferior to those of good programming languages. So why would I leave those productivity gains on the table?
Using EdgeQL simple means I have another programming language at hand.
> EdgeDB has a robust type system that's most [sic] comprehensive that most ORMs
Well yeah. And it is inferior to the programming language I use. Hence, this comes at a disadvantage to me.
> We've also built a query builder for TypeScript
Aha, so then... why not build a query builder for every language, just like with ORMs?
Sorry, not convinced. We would be better off by improving on SQL itself.
> And not a great one. Null handling and typesystem for example are way inferior to those of good programming languages
There is no mainstream programming language that I know off that offers what are table stakes for an RDBMS.
For example, even trivial SQL things like a constraint saying 'at least one of these two fields must be empty, but both can't be empty' is missing in the "advanced" type systems in mainstream languages.
First of all, your example constraint is not possible to be statically expressed in SQL anyways. In the typesystem, that is.
Sure, you can create a constraint for `'at least one of these two fields must be empty, but both can't be empty'` but that will only be applied at runtime. It's not like any dmbs (to my knowledge) will reject the query at parse-time. Or else, please show me an example of how you would do that. (and don't use constraints or indexes, because those only work at runtime)
Second, I deliberately said "good programming languages" and you changed that to "mainstream programming language". You know what? If all your mainstream programming languages (however you count or define those) don't support that stuff, maybe it's time to move on and choose a better language.
In Scala at least, it's trivially possible to define such a constraint with types. That ensures that you will never by accident violate that constraint. And that is guaranteed at compile time, so no one is gonna wake you up in the night because your query failed due to that condition you specified.
If you don't believe me, I'm happy to show you the code to encode that logic.
> First of all, your example constraint is not possible to be statically expressed in SQL anyways. In the typesystem, that is.
So? The DB will prevent you from violating the constraints, because it cannot tell in advance (i.e. before getting the query with the data in it) whether the data violates the constraint.
> It's not like any dmbs (to my knowledge) will reject the query at parse-time.
What parse-time?
echo "INSERT INTO tbl_test (c_one, c_two) VALUES (3, 4);
Where's the parse-time in that?
TBH, I am still failing to see your point - this looks like an artificial restriction (the query must be rejected before you present it to the DB).
Whether you call it runtime or compile-time or parse-time, the DB will not let you violate the type safety by accident.
The point is that the DB will enforce the constraints.
> Second, I deliberately said "good programming languages" and you changed that to "mainstream programming language".
Well, if your bar for "good programming languages" rules out all the mainstream languages, what's the point of even discussing your point? The point you are making then becomes irrelevant.
> If all your mainstream programming languages (however you count or define those) don't support that stuff, maybe it's time to move on and choose a better language.
For better or worse, the world has rejected those better languages and relegated them to niche uses. Shrieking shrilly about your favourite languages aren't gonna make them more popular.
You know what's more realistic? Teaching the users of the "poorer" languages that they can get all those benefits of type enforcement in their DB without needing to switch languages.
> In Scala at least, it's trivially possible to define such a constraint with types.
In more than a few languages it's possible to do that. I'm thinking more Prolog, and specific SAT solvers than Scala, though. There's benefits in doing so.
However, the minute you plug an RDBMS into your system, many benefits can be gained without switching languages at all. Like real constraints for XOR or composite uniqueness, referential integrity, NULL-prevention, default values, etc.
> What parse-time? (...) TBH, I am still failing to see your point
What I tried to say was: in some programming languages, this insert will not even compile. So you don't have to write a test, you don't have to spin up a test database or anything, it just doesn't compile. And I prefer that over having to wait until a query is actually sent before I get an error.
This is relevant for me because 1.) it makes me more productive since I get the error much quicker and I don't have to write a test and 2.) it prevents getting calls in the night because something broke.
I hope that makes it clear.
> However, the minute you plug an RDBMS into your system, many benefits can be gained without switching languages at all.
Yeah, but that doesn't invalidate my original point, does it?
> Yes and you absolutely have to. That's not a disadvantage, it's just how it is. Because SQL is the absolute bare minimum. The lowest common standard. And not a great one. Null handling and typesystem for example are way inferior to those of good programming languages. So why would I leave those productivity gains on the table?
I think... we're in a agreement? :)
EdgeQL doesn't have a NULL (it's a set-based language and NULL is an empty set, this tiny adjustment makes it easier to reason about missing data). And because it also has a more robust type system, allows for limitless composition and easy refactoring, it has far greater DX than SQL => you're more productive.
There's a footnote here: EdgeQL doesn't support some of the SQL capabilities just yet, namely window functions and recursive CTEs. But aside from that it is absolutely a beast.
> Well yeah. And it is inferior to the programming language I use. Hence, this comes at a disadvantage to me.
Maybe, but I'm curious how you arrived to that conclusion. I assume you mean that using the power of a high-level programming language you can force ORM to complete submission and that's just not true. Most of the time you'll either have grossly inefficient multi-roundtrip query code (hidden from you) or let ORM go and use SQL. Obviously that's an extreme scenario, but it's surprisingly common in complex code bases and logic.
> Aha, so then... why not build a query builder for every language, just like with ORMs?
We are. We started with improving the network protocol (it does less round-trips than Postgreses and is stateless) and crafting client libraries for MANY languages. All client libraries support automatic network & transaction error recovery, automatic connection pooling, and have generally great and polished API. Not to mention they are fast.
We do have a query builder for TypeScript. But we also have codegen for every language we support: place an .edgeql file in your project and you get fully typed code out of that. That said, we will eventually have query builders for every language we support. There's only so much you can do in 3 years since we announced 1.0.
> Sorry, not convinced. We would be better off by improving on SQL itself.
Significantly improving SQL without starting from scratch isn't possible. Adding sugar - surely is possible, but we are 100% that the productivity boost we deliver with EdgeQL is worth going all in (and our users agree with us).
In any case, with Gel 6 we have full SQL support (except DDL), so it's possible to use Gel along with ORMs if that's needed. We are not SQL haters at all.
> Most of the time you'll either have grossly inefficient multi-roundtrip query code (hidden from you)
Or you spend 5 minutes actually putting some effort in to configuring your ORM. Maybe spend 30 minutes learning if it's your first time. People will spend weeks handwriting SQL to save hours of ORM tuning.
In my experience it is the exact reverse. You spend weeks configuring or tuning your orm calls when you ccould just spend a few hours optimizing your sql. There is a reason most orms has a raw SQL escape hatch.
The sweet spot is typesafe sql libraries. You write your raw sql and library deduces the return type from the database and gives you typesafety (sqlx is really good for this, sqlc for go is similar). It gives you almost all the benefits of ORMs with almost no downsides.
For me, personally, I do a ton of fullstack work in JavaScript-land, but also have Python services for ML-heavy needs, and it’s nice to define one schema that at its root is still SQL while generating query clients for multiple languages.
