Open fpapado opened 2 months ago
I'm having the same issue. We're integrating VGS for the first time, and started off with version 1.2.0.
Same thought, we could use 1.1.0, at least to unblock ourselves, but ideally this would be fixed so we can continue to update versions
Hi @fpapado, @filipebarcos, I see the issue thanks for the detailed issue explanation, I was impressed by this :) We will release a fix as soon as possible.
Hi guys, weβre replacing nanoID with a built-in JavaScript function.
New version 1.3.0
Expected Behavior
The expected behavior is that
@vgs/collect-js-react
can be used in React trees that are consumed by a CJS node process, such as Server-Side Rendering (SSR), without crashing. This sounds quite terse, the "current behavior" section below explains it better πCurrent Behavior
@vgs/collect-js-react
is published as a dual-module package: there is a CJS version marked by themain
field inpackage.json
(dist/index.js
) and an ESM version marked by themodule
field (dist/index.modern.js
). Bundlers and node know how to resolve each one.However, there is a quirk:
nanoid
is an ESM-only package, which is ok by itself. However, when collect-js-react'sdist/index.js
is loaded by node as CJS, the resolution eventually reaches therequire('nanoid')
statement. The current stable node versions do not allow synchronous requires of ESM modules, leading to theERR_REQUIRE_ESM
error at runtime.This scenario can happen in SSR setups, where a node process is typically used to render React trees to HTML. While it is possible to run ESM-only node servers, our current setup at work is still CJS, and I believe that is a common setup elsewhere. A module graph that imports collect-js-react will crash in such scenarios.
In many cases, TypeScript can actually highlight these issues while building the library. I noticed that the moduleResolution setting in this library's tsconfig.json is
node
, which is the old/legacy node behavior (nowadays known asnode10
). If you configure"moduleResolution": "node16"
, you should be seeing an error about the ESM-onlynanoid
being imported in a CJS context at type-checking time.Here is TypeScript's error, after tweaking the config and dependencies in this repository to use
typescript@^5
:Possible Solution
First idea:
nanoid@^3
One possible solution is to use
nanoid@^3
, which is still supported, and is also a dual package (nanoid themselves recommend this approach, citing that they still support it). This might be the simplest approach, as far as I can see.Second idea:
useId
, with some caveatsMy other thought was to use React's native
useId
to support the ids for the fields.nanoid
is only used to initialise and query fields, which I don't think needs a full UUID construct (but please tell me if I am wrong π)However, there are some constraints around using
useId
.The first constraint, is that collect-js-react supports react 16/17/18 (according to
package.json
peerDependencies
).useId
is only available in React 18 and upward. While we could feature-detect its presence in the React exports, there would still be an issue for users of other react versions. Conditionallyrequire()
ingnanoid
is also not great in this scenario, because the ERR_REQUIRE_ESM error would persist for those users.A possible solution is to use a shim for
useId
, such as a global counter, for React <18. A global counter can lead to hydration errors in SSR scenarios, but the id itself is never actually rendered in the DOM on the server (because@vgs/collect-js
runs only on the client, right?).There are also some other smaller issues with
useId
, such as having toCSS.escape()
it, because it ends up being used indocument.querySelector
.I am happy to help prototype this, if you want. However, using
nanoid
version 3 might solve this with less complexity.Steps to Reproduce (for bugs)
I made a minimal demo with a reproduction inside: https://github.com/fpapado/vgs-collect-js-react-esm-require-error. There are a few moving pieces, so this was simpler π
Context
We have been using
@vgs/collect-js
and wanted to use the first-party React bindings. Version1.1.0
was working well for us, but1.2.0
surfaced this issue.I believe this is a breaking change between version
1.1.0
and1.2.0
, since the public-facing API of the module (the way it is packaged, and its runtime compatibility) have changed.While we could stay on
1.1.0
, we worry that we will fall behind on future feature additions and security updates, so we would like to resolve this hurdle somehow. We like VGS and have spent a while thinking about library packaging, so we thought to offer some context and help πPlease let me know if there is any other information that I can provide, and thank you for all the work you do on VGS!
Your Environment