Locking down fragment/query on bare package names about modules

[giltayar]

I have to admit that this seems counter-intuitive to me. Maybe logical, but counter-intuitive:

If I do

import './foo?x=4'

Then I import the file ./foo.mjs, and the x=4 is part of the cache key, so is used as a "cache buster".

But if I do

import 'foo?x=4'

Then we're importing the package in the folder/file node_modules/foo?x=4? I would expect to load node_modules/foo with a cache-busting x=4.

While I understand the logic—bare imports are not URLs, and should not be treated as such—I believe most people wouldn't understand this and would be surprised by this behavior, especially if

import 'foo/bar?x=4'

is treated (somewhat) as a URL and the x=4 suddenly becomes a cache-buster again.

In the interest of the principle of least-surprise, I hope it doesn't land this way.

[jdalton]

Wouldn't this also impact how folks were planning to handle reloading modules
(e.g. the cache buster hash trick from the browser)?

[jdalton]

\cc @iarna for npm since it's package name related.

[guybedford]

In the current implementation the above falls into the ResolveModule path at https://github.com/nodejs/node/blob/master/src/module_wrap.cc#L594. Which then concatenates the specifier into the node_modules lookup. Once found, the path is taken to be the part after the package name length.

So the above example would check the directory node_modules/foo?a directly on the filesystem.

So we're actually already pretty locked down and in agreement with the package map approach here.

[bmeck]

@jdalton to an extent?

@guybedford that is a good catch and would have been considered a bug if I saw it before this issue. Since it was not my intention when writing that to have that specific behavior it seems we should still talk about it though.

[bmeck]

@giltayar you can post the issue I created in package name maps, but I did not see a way to argue about it if prohibiting the characters is seen as a bad idea. Any way it goes, we should match package name maps.

[bmeck]

Conclusion should stand and we aren't discussing this anymore.