Vite 6 compatibility issue `resolve.conditions`

[thebanjomatic]

Describe the bug

When using a browser-based evironment (happy-dom, jsdom, browser), vitest still passes conditions: ['node'] in the resolve config.

When using Vite 5, this doesn't really matter because it would infer that we were targetting a web environment and it would resolve in that context anyway. I'm not 100% sure on how this worked before, but tryNodeResolve was being passed a targetWeb boolean and the returned ID will match the browser / default import conditions instead of the node condition.

With Vite 6 a lot of this default environment magic was stripped out as part of the Environments api refactor, and now it's just using the resolve.conditions from the corresponding client/ssr configs. This results in a change in behavior where Vitest now uses the node condition for browser environments when using Vite 6.

Using the node condition instead of browser can result in test failures due to differences in behavior between browser and node implementations of a library, but I've also run into issues where esbuild as part of deps optimization generates invalid code where the module itself is "babel compatible" but tries to deconstruct the named exports from import_mod1.default instead of import_mod1.

These issues reproduce in the latest Vitest 2.1.8 using a resolution for Vite 6.0.3, as well as Vitest 3.0.0-beta.2.

Reproduction

I'm still working on creating a reproducer that I can share and should be able to do so by the end of the day.

In terms of the actual code changes, I think these two locations can be changed to use:

conditions: [testConfig.environment === 'node' ? 'node' : 'browser']

vitest/packages/vitest/src/node/plugins/index.ts

Lines 88 to 94 in 4e60333

	resolve: {
	// by default Vite resolves `module` field, which not always a native ESM module
	// setting this option can bypass that and fallback to cjs version
	mainFields: [],
	alias: testConfig.alias,
	conditions: ['node'],
	},

vitest/packages/vitest/src/node/plugins/workspace.ts

Lines 67 to 73 in 4e60333

	resolve: {
	// by default Vite resolves `module` field, which not always a native ESM module
	// setting this option can bypass that and fallback to cjs version
	mainFields: [],
	alias: testConfig.alias,
	conditions: ['node'],
	},

But it's also a little unclear to me if we should be clearing out the mainFields: [], for the browser-based environments also.

System Info

System:
    OS: Windows 11 10.0.22631
    CPU: (20) x64 12th Gen Intel(R) Core(TM) i7-12800H
    Memory: 11.40 GB / 31.64 GB
  Binaries:
    Node: 20.18.0 - ~\AppData\Local\fnm_multishells\18680_1733347780000\node.EXE
    Yarn: 4.5.3 - ~\AppData\Local\fnm_multishells\18680_1733347780000\yarn.CMD
    npm: 10.9.1 - D:\source\delete-me\test-vite6\node_modules\.bin\npm.CMD
    bun: 1.1.29 - ~\.bun\bin\bun.EXE
  Browsers:
    Edge: Chromium (127.0.2651.74)
    Internet Explorer: 11.0.22621.3527
  npmPackages:
    vite: ^6.0.0 => 6.0.3

Used Package Manager

yarn

Validations

• Follow our Code of Conduct
• Read the Contributing Guidelines.
• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Check that this is a concrete bug. For Q&A open a GitHub Discussion or join our Discord Chat Server.
• The provided reproduction is a minimal reproducible example of the bug.

[sheremet-va]

There is no compatibility issue here. happy-dom and jsdom tests are running in Node.js with browser globals. There is no widespread decision on how the browser condition should be treated - some libraries expect imports to follow bundlers logic (don't use extensions) which breaks in Vitest when externalized, some just expect globals to be present. Vitest added these conditions for a reason and we are keeping them. The same can be said about the mainFields treatment - since Node.js can't resolve those, it's better to remove them altogether to avoid module hazard if the module is externalized.

When using a browser-based evironment (happy-dom, jsdom, browser)

The browser mode doesn't modify resolve conditions.

[sheremet-va]

This results in a change in behavior where Vitest now uses the node condition for browser environments when using Vite 6.

I missed this part - we might need to inject default conditions now in the config, but we shouldn't remove the node condition.

[thebanjomatic]

Reproducer:
https://stackblitz.com/edit/vitest-dev-vitest-xzepei6n?file=package.json&view=editor

Vite 6.0.3:

Vite 5.4.9:

So, it does look like the node conditions were still used in the vite 5 case as that test case fails in both versions, but the tslib error when using optimized deps only reproduces in vite 6 since in vite 5, since the "vite:resolve" plugin was going down a different path (ssr: false implies targetWeb: true) and returning the browser condition instead of node.

This tslib issue is the failure that drove me down this path, and ultimately what is blocking my teams from adopting Vite 6 at this time. If that issue should be filed with the Vite repo instead, let me know.

[sheremet-va]

Reproducer: stackblitz.com/edit/vitest-dev-vitest-xzepei6n?file=package.json&view=editor

I just want to note that "browser" here is the "happy-dom" environment, not the actual browser, the browser mode in Vitest uses a separate Vite server.

[thebanjomatic]

Yes that's correct. I probably should have been more clear in the naming of the project. I think that's fine and I mainly want things to continue working the same as they do in Vite 5. Since there is no intention of changing away from conditions: ["node"], I have updated the reproducer to reflect that and to instead only focus on the failure case:

https://stackblitz.com/edit/vitest-dev-vitest-xzepei6n?file=package.json

Vite 5:

Vite 6:

[thebanjomatic]

From testing, it looks like adding back in the default conditions for vite >=6 fixes this issue. I'll submit a PR.