Zod V4 implementation

[SamJB123]

Ready for testing and refinement by community - but expect rough edges!

Current version includes several opinionated decisions with a goal of writing a schema once.

• Uses ZodV4 internals to map Zod's wide range of higher-level niche typings (IPV6, email etc) to their base primitives.
• The alignment allows for a bidirectional schema, two-way validation (writes and reads), and type safety in IDE, at build time, and at run time.
• Tuples were the trickiest zod "primitive" to deal with. The helper stores as objects at rest, and the helper restores them to being real tuples automatically for client-side actions and functions on the Convex backend (so array methods are still available).
• Transforms required the biggest value judgment. Landed on respecting Zod's intent that they are one way trips. There's still unused functions in ZodV4.ts from my attempts to create full two-way conversion for people who want to dabble, but given some transforms are lossy (e.g. cropping a longer body of text), the complexity-benefit payoff didn't make sense to me.
• After a few weeks of testing, I mentally treat transforms like the mirror of Convex queries, but for writes. Once transforms are written to the database, those outputs are still validated - but against the post-transform type.
• I have not attempted to touch overwrites in this PR. At best, they might work out of the box in the same way as normal transforms. But I can't say for sure.

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[ianmacartney]

in #707 i put up some fixes but I couldn't remember the right way to try to upstream them to yours. let me know if I should fork your fork or if I could have permissions to push to that branch to collaborate

• I moved the gui test into the root convex/ and src/ folder and out of the package
• If you could use the linting & prettier formatting for the repo it'll help
• running npm run dev in the root will run the tests, keep generated code up to date, etc.

It's really promising! Thanks for all the work

[ianmacartney]

This doesn't seem to be a nullable optional..

[SamJB123]

Yeah..this might be a hangover from before I revised the approach to optionals (overhauled it at one point during writing to recognise that objects can happily handle optionals without any special treatment)

[mikecann]

The tests are failing but I ran it through AI and was able to fix all outstanding issues. I can commit if you like?

[SamJB123]

@mikecann I'm happy if you're happy! Added you as collaborator too in case it makes things easier somehow.

[ianmacartney]

I'll close out #707 as I just was able to push up my changes here

[mikecann]

Okay I have added a commit that fixes the tests. Just to be clear the AI did all the hard work here based upon the expected values found in the failing tests.

So hopefully those expected values were correct otherwise these fixes will be incorrect.

I don't have enough context on Zod4 to say either way if this is correct or not but after scanning through the 3k lines of tests it all looks okay 🤷

I do like the look of the bidirectional schema that looks nice.

[hasanaktas]

Is there a package with zod v4 support on npm? The latest version of npm is 0.1.101-alpha.0

[ianmacartney]

@hasanaktas the latest alpha now has these changes

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/get-convex/convex-helpers@701

commit: 3605661