feat: Add Deno.sqlite bindings

[littledivy]

#14460

[littledivy]

This patch:

./target/release/deno run -A --unstable sqlite.js
cpu: Apple M1
runtime: deno 1.21.3 (aarch64-apple-darwin)

benchmark                        time (avg)             (min … max)       p75       p99      p995
------------------------------------------------------------------- -----------------------------
SELECT * From "Order"         17.83 ms/iter   (17.55 ms … 18.31 ms)  17.94 ms  18.31 ms  18.31 ms
SELECT * From "Product"       41.42 µs/iter  (40.79 µs … 160.83 µs)  41.33 µs  44.21 µs  45.83 µs
SELECT * From "OrderDetail"  335.29 ms/iter (310.42 ms … 399.25 ms) 323.45 ms 399.25 ms 399.25 ms

Node better-sqlite:

node ../better-sqlite.mjs
cpu: Apple M1
runtime: node v16.13.1 (arm64-darwin)

benchmark                        time (avg)             (min … max)       p75       p99      p995
------------------------------------------------------------------- -----------------------------
SELECT * From "Order"         25.57 ms/iter   (25.37 ms … 26.09 ms)  25.65 ms  26.09 ms  26.09 ms
SELECT * From "Product"       68.05 µs/iter  (66.42 µs … 371.75 µs)  67.88 µs  73.46 µs  78.46 µs
SELECT * From "OrderDetail"  546.51 ms/iter (515.57 ms … 602.88 ms) 571.54 ms 602.88 ms 602.88 ms

Benchmarks for FFI / WASM modules can be found here

[bartlomieju]

Discussed during the last CLI design meeting, there's no support for this feature from the team (it should be done as a userland module either in WASM or using FFI API), so we're closing this PR.

[fdionisi]

That's a real pity. I was looking forward to this. In conjunction with #13122, it would have been a killer feature (at least for my use case 🙃).

[sant123]

Discussed during the last CLI design meeting, there's no support for this feature from the team (it should be done as a userland module either in WASM or using FFI API), so we're closing this PR.

@bartlomieju Could this be re-considered? Looking at the commits I was hoping the release to be completed soon and for my surprise got this message.

[lucacasonato]

Could this be re-considered?

It's was the third time we had reconsidered this feature in 6 months 😅

What is your use-case that requires this to be built in, instead of using a user-land module?

[mikrosystheme]

This is very unfortunate. SQLite is a fundamental piece of technology that is useful in almost every software. @lucacasonato Instead of asking for use-cases (there are too many and it is pointless to list them all), what are the reasons to don't have it in core? Using a third party module adds friction (which one? will it be maintained in the long run? it is optimal or a subpar WASM version?) that should be avoided.

[fdionisi]

@lucacasonato, of course, user-land modules are a viable option, but I agree with @mikrosystheme observations.

Landing SQLite would allow fast prototyping in Deno with a database out of the box; if/when the prototype grows, you then can change the db with a more scalable solution. Same idea behind including TypeScript in the runtime (JS -> TS).
Then, with Jupyter integration, it would help answer the question "why not Python?", which is always hard to answer.

But then again, I suppose it is not sustainable to add everything shiny into Deno. 😅 It is just that seeing this PR growing got me excited about the possibilities that Deno could provide with just a single executable.

To answer your question, I have no solid use case that third-party modules cannot solve. But it would be a tangible improvement to DX, and more solid ground to push for Deno adoption.

[sant123]

Could this be re-considered?

It's was the third time we had reconsidered this feature in 6 months 😅

What is your use-case that requires this to be built in, instead of using a user-land module?

Well basically the out of the box feature is something that most people will like. I know there are third party modules that would solve the issue but they will not be as performant as a native built in integration. Also there is a library that uses SQLite bindings using FFI that I think is pretty cool but unfortunately cannot be used in Deno deploy. So having the SQLite built in Deno would be useful for both fast prototyping and secure ACID compliant operations.

[sant123]

@bartlomieju @lucacasonato could you consider it again please? 😅🙏🏼 I apologize for being such insistent.

[bartlomieju]

@sant123 I've been and still am supportive of this feature, however there's no broader support from the rest of the team. We might revisit this topic in a few weeks. I suggest to open a new issue or revive an existing one to get more feedback from the community.

[kitsonk]

Issue #11657 was re-opened