I don't know without debugging the new test you've added there, but often calls to POSIX apis on Fuchsia return -1 and set errno=ENOSYS, as they're just empty stubs. So I'd guess that somewhere along the way the test is hitting one of those paths.

An #if is probably fine, or disabling it on Fuchsia.

I think I'll disable it instead, just so it gets attention - maybe it's fine / doesn't matter, but best to be safe.  Curiously, there is another test in net/ that uploads a non-existent file, and seems to be enabled on Fuchsia, and fails with the right error code.

Can you give an example of a path which returns ERR_NOT_IMPLEMENTED ? And a path which returns ERR_FILE_NOT_FOUND?

It's possible that the NOT_IMPLEMENTED code is being returned from a different filesystem type than the FILE_NOT_FOUND filesystem, hence the asymmetry. Hard to say if it's WAI without knowing more, though.

For the one that returns ERR_NOT_IMPLEMENTED, I create a temp directory, and then use a path to a file in it that does not exist.

Hrm...  The ERR_FILE_NOT_FOUND_ONE looks to use c:\path\to\non\existant\file.randomness.12345"...  Even on Linux.  Wanting a cross-platform path was why I went with the other approach.

Wait...that's not right.  My test, because I was lazy, uses the path of a file used in other tests (Since I had it handy) that lives in content/test/data, and then appends a "foo" directory after the [valid, extant] file name.

Given that the case is weird, I can presumably try to do something less weird, and the test will likely pass - but I think that case should be returning a file not found error.

Removing HTTP label - seems to solely be a platform issue.  Using a "normal" non-existent file seems to resolve the issue.

Archiving P3s older than 1 year with no owner or component.