vite: SSR transformed named imported functions don't work with debuggers "step-into"
Describe the bug
Originally reported at https://github.com/vitest-dev/vitest/issues/5569. I think I’ve seen 2-3 similar reports in the past too.
When a file is SSR transformed in Vite, debuggers like Chrome Dev tools and VSCode are unable to “step into” functions that are imported via named imports. In practice this means stopping debugger in line 3 and pressing “step into” button of debuggers in the example below.
1 | import { add } from "./math.ts";
2 |
3 | add(1, 2);
Debugger should stop at line 1 or 2, but instead it stops at 7.
1 | export function add(a: number, b: number): number {
2 | if (a === 1) {
3 | return 1;
4 | }
5 |
6 | return a + b;
7 | }
8 |
Source map of the math.ts in reproduction: https://evanw.github.io/source-map-visualization/#NDI2AGZ1bmN0….
Video demonstrating step-into issue with Vite SSR
I think this has something to do with the way Vite SSR transforms import into import() and saves the results into single variable. In the reproduction setup I’ve created “Vite simulation”, where SSR transformed file is loaded directly into Node.
In this example “step into” works correctly on lines 11 and 15. If Vite was able to destructure the named imports of __vite_ssr_exports__ into separate variables, I think debuggers would work just fine. But this would break ESM live bindings.
1 | globalThis.__vite_ssr_exports__ = {};
2 | debugger;
3 |
4 | await import("./transpiled.js");
5 | const { add, multiply } = globalThis.__vite_ssr_exports__;
6 |
7 | // Incorrect position when "step into"
8 | globalThis.__vite_ssr_exports__.add(1, 2);
9 |
10 | // Correct position when "stepinto"
11 | add(1, 2);
12 |
13 | // Same here
14 | globalThis.__vite_ssr_exports__.multiply(1, 2);
15 | multiply(1, 2);
16 |
Video showing how lines 11 and 15 work correctly
Reproduction
https://github.com/AriPerkkio/vite-ssr-transform-sourcemaps-issue
Steps to reproduce
See README https://github.com/AriPerkkio/vite-ssr-transform-sourcemaps-issue/blob/main/README.md.
System Info
System:
OS: macOS 15.0.1
CPU: (8) arm64 Apple M2
Memory: 471.58 MB / 16.00 GB
Shell: 3.2.57 - /bin/bash
Binaries:
Node: 20.17.0 - ~/.nvm/versions/node/v20.17.0/bin/node
Yarn: 1.22.22 - ~/.nvm/versions/node/v20.17.0/bin/yarn
npm: 10.8.2 - ~/.nvm/versions/node/v20.17.0/bin/npm
pnpm: 8.15.9 - ~/.nvm/versions/node/v20.17.0/bin/pnpm
Browsers:
Chrome: 129.0.6668.91
Safari: 18.0.1
Used Package Manager
pnpm
Logs
No response
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.
- Make sure this is a Vite issue and not a framework-specific issue. For example, if it’s a Vue SFC related bug, it should likely be reported to vuejs/core instead.
- 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.
About this issue
- Original URL
- State: open
- Created 2 years ago
- Reactions: 1
- Comments: 31 (20 by maintainers)
There’s also one big inconvenience with debugging related to the SSR transformations – inability to evaluate code that includes imported modules.
For example,
If
reshas unexpected value, I normally would want evaluatehello()andbye()to see what they return to make sense of theresvalue. However, they are not defined, as they live in the__vite_ssr_import_0__.helloand__vite_ssr_import_0__.bye.I guess no. That feature is not exposed in Node API (I’m not sure which feature you are referring to, but probably) and other APIs exposed from Node does not have a stable ABI. That means we’ll have to compile the lib for each Node version, otherwise the users has to compile it. Even if we compile it for each version, AFAIK there’s no way to do optional dependencies based on Node version, so all versions will have to be downloaded (unless we do it on our own with
postinstallscript).I think I found solution that works identical to ESM in Node.js (not in stackblitz for some reason - looks like they are using a vite-node like wrapper to import 😄 )
https://stackblitz.com/edit/node-7j9afi?file=test.js
Running this code in Node.js shows the same values for the native ESM module and the proxy+wrapper. One caveat - to make this work, we need to know all exports before we execute the code (we can just return them alongside dependencies)
The new algorithm would be something like this:
=assignment to assign values instead ofObject.definePropertybecause the module will be sealed before the values are assigned at runtime__vite_ssr_exports__.${method}when the export is reassigned__vite_ssr_identity__use(0, method)()__vite_ssr_exports__objects with all exported properties assigned toundefined(we kept the record of all exports) and sort them__vite_ssr_namespace__that throws an error if someone attempts to assign a value__vite_ssr_import__now returns__vite_ssr_namespace__instead of__vite_ssr_exports__for public usageBenchmark-wise it seems to be the slowest one 😄 (Even slower than getters, even though we don’t define a
getmethod on a proxy). Property access seems to be on par with reassignment (I recommend running benchmarks locally and not on stackblitz because it messes withimports):In real-world scenarios using the module runner, it makes it around ~4 mil operations slower:
Just a note: internally, module namespace exports are defined as getters/setters: https://github.com/nodejs/node/blob/main/deps/v8/src/builtins/accessors.cc#L265