Support simd_json as an optional alternative to serde_json #934
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The approach is to abstract _all_ original serde_json references to a new lib crate in the workspace - json-impl. That crate as a feature swicth between serde_json and simd_json. In The initial phase, is to to get all the other crates to depend on json-impl with the serde feature, instead of serde_json directly, so in the end nothing should change. The next phase is to activate the simd feature and see what fails, then start mitigating the failures.
|
I have run NO tests yet, even for the serde_json abstraction. |
added 2 commits
October 19, 2024 17:58
The _problem_ with simd_json is that it requires _mutable_ slices to operate while serde_json does not. This FIRST PASS (certainly not the end solution) is to provide &mut adapter of the from_slice function for serde_json, a pseudo-safe adapter of from_str for simd_json, and then change all the places where stuff is flagged as needing to be mutable to be so - MOSTLY, however, this is in TESTS - so eventually we'll clean that up.
Now we see the REAL problems: * No "std::cmp::Eq" available on simd::OwnedValue (at least as it is now) * SOME use of &str passed to a deser visitor and then fed back into simd_json::from_str ... which needs a mut reference (and is technically unsafe)
|
(Where's the bang-head-against-wall emoji when you need it!) - so, up to now, the mut and unsafe stuff - well we can deal with it and it's nicely encapsulated to the Oh but simd_json isn't done with us yet, and this one requires some serious mods to simd_json - simd-lite/simd-json#398 - so we're kinda blocked until that's solved. (But, then again, one of my suggestions in that issue is a feature pulling simd_json Value representation closer to serde_json, so PERHAPS we can encapsulate ALL - or most - of the issue mitigations in such a feature) |
added 25 commits
October 22, 2024 13:57
... all we have a couple of missing functions and a mutability difference/error (in the base crates - haven't gotten onto the main client interface crates yet.
…ambda_http Of course, lambda_http IS the target for all this and the thing that will expose the mut and unsafe aspects of using simd to the world - that's for another day!
…references And add a very useful "from_bytes" combinator to aid in parsing from stuff like HTTP bodies (though from_reader is probably cheaper).
Plus the from_vec using an owned vec gets us around some awkward lifetime issues when trying to write "generic" code for both simd_json and serde_json
simd_json deserializer
THAT is NOT easy to create for simd_json.
We've removed Deserialize from of lambda_http::request::LambdaRequest ... that makes EVERYTHING compile in the simd_json context ... with the exception of th calls to lambda_runtime::run which require the expected request type to be serde::Deserialize. So now it's back to that issue - which, if I recall well, was the blocker back when I first started this!
At the moment the only way around the deserializer-must-be-simd issue is an unsafe "cast" ... and even the type equivalence check that I tried before that cast will itself not compile, so that is currently commented out.
added 16 commits
October 25, 2024 12:53
And this mod was proof that simd_json does, indeed, mess with the source slice! Docs on switching to simd_json will need to make that very clear.
simd_json http request deserialization isn't working from external caller
Turns out I was casting to the wrong thing
Using local patch instead
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
THIS IS A DRAFT PR TO MAKE KNOWN WHAT I'M PLAYING AT - PLEASE DO COMMENT AS YOU SEE FIT
The approach is to abstract all original serde_json references to a new lib crate in the workspace - json-impl.
That crate as a feature swicth between serde_json and simd_json. In
The initial phase, is to to get all the other crates to depend on json-impl with the serde feature, instead of serde_json directly, so in the end nothing should change.
The next phase is to activate the simd feature and see what fails, then start mitigating the failures.
📬 Issue #, if available:
✍️ Description of changes:
🔏 By submitting this pull request
cargo +nightly fmt.cargo clippy --fix.