fix: cache paths on Windows are broken

[bartlomieju]

Cache paths are not properly generated on Windows.

It's broken since v0.13.0.

Ref:
denoland/std#552
denoland/std#556

Example:

file:///C:/deno/js/unit_test_runner.ts
// for above URL this cache path should be generated 
file\c\deno\js\unit_test_runner.ts
// this is produced (probably)
C:/deno/js/unit_test_runner.ts

[bartlomieju]

@piscisaureus I think I managed to make it work. We're gonna need to do a release with this change to fix CI in standard lib

[piscisaureus]

Add an unreachable() here. Let's not silently discard part components.

[bartlomieju]

@piscisaureus added

[piscisaureus]

This fixes a very acute problem, so I'm all for landing this.
However, we have similar problems with other path components, which can also contain illegal characters (e.g. ? and &).

Tangentially related, there's also a correctness issue w.r.t. case sensitivity. In URLs the path and query component are case sensitive, while the protocol and host are not. When deno_dir lives on a case insensitive file system it won't be able to handle that properly.

[bartlomieju]

However, we have similar problems with other path components, which can also contain illegal characters (e.g. ? and &).

Haven't thought about it, but I believe Url won't return them from path_segments method. That inherently means that http://example.com/a.js?foo would be cached as http/example.com/a.js.

Tangentially related, there's also a correctness issue w.r.t. case sensitivity. In URLs the path and query component are case sensitive, while the protocol and host are not. When deno_dir lives on a case insensitive file system it won't be able to handle that properly.

Agreed, any ideas how to tackle that? Cast all protocols and hostnames to lowercase?

[piscisaureus]

Haven't thought about it, but I believe Url won't return them from path_segments method. That inherently means that http://example.com/a.js?foo would be cached as http/example.com/a.js.

That doesn't seem right either. Only the part after # can be safely discarded.

[bartlomieju]

That doesn't seem right either. Only the part after # can be safely discarded.

Agreed, shall I open a new issue with two described edge cases and we can land this patch?