AssetReaders and AssetLoaders - lack of communication outside of meta files?

[laundmo]

I was working on:

• Enable setting headers and query parameters. Also add 'fake extensions' mechanism. johanhelsing/bevy_web_asset#35

Which improves a rather simple AssetReader for reading from web-requests.

During this, I was wondering why AssetReaders cannot communicate any context/metadata to AssetLoaders, or even decide which AssetLoader from a list of matching ones (multiple registered for the file extension, for example) to use.

For example, I had to implement a "fake file extension" mechanism which allows adding ..png to the end or a URL. This extension is then stripped before sending the request for cases where an API, for example, doesn't accept file extensions. It would be very nice here for the reader to read the Content-Type header, which could then be used to hint to the AssetLoader for images which format the image is (or even decide outright which AssetLoader to use).

As far as I can tell, though, it wouldn't be possible to return an ad-hoc .meta file only after the asset has been read.

I can also imagine other cases where this could be very useful, as anything which isn't exactly a filesystem will likely have some metadata available which isn't in .meta format.

Is something like web-requests or database queries just too far outside the target use case for bevy_asset?