filesystem: ErrorCode::IsDirectory when opening directory with WRITE

[wingo]

On Windows, it was possible to return a directory descriptor even if WRITE was in the permissions. Fixes wasmtime for
WebAssembly/wasi-testsuite#176.

[wingo]

In all honesty I haven't tested on Windows, it's just that this patch should fix the issue. To repro, check out the prod/testsuite-base branch of https://github.com/WebAssembly/wasi-testsuite and run:

wasmtime -Wcomponent-model-async -Sp3,http --dir 'tests\rust\testsuite\wasm32-wasip3\fs-tests.dir::fs-tests.dir' 'tests\rust\testsuite\wasm32-wasip3\filesystem-flags-and-type.wasm'

[wingo]

Rebased to fix the rustfmt error

[wingo]

My goodness, there are many CI checks in this project :) Patch fixes unused-variable warning/error when compiling for Windows.

[rvolosatovs]

My goodness, there are many CI checks in this project :) Patch fixes unused-variable warning/error when compiling for Windows.

You can add prtest:full to commit message description to trigger a full CI run for cases like these

[alexcrichton]

Test-wise this is something that'll be caught when updating to the latest wasi-testsuite, right? If so, I think it's ok to skip a test here. Mind leaving a comment though on this match arm explaining that this is sync'ing windows semantics to unix semantics?

[wingo]

I had misremembered the nature of this error, my apologies. The problem is not the lack of READ, it's the presence of WRITE; see the open specification. Thank you for your patience.

[wingo]

Lordie me, this PR has been an education for me, thanks all for patience. Ready now (again).

[alexcrichton]

Anything in particular we could make easier on our end? Or more of a "lots of stuff to discover" kind of education?

[wingo]

Anything in particular we could make easier on our end? Or more of a "lots of stuff to discover" kind of education?

Oh, I think things are pretty good on the wasmtime side, just that there were a number of considerations that I didn't fully grasp (the history of wasi versions and their implementations, the layers of error codes (std::io vs ErrorCode etc), the specificities of how CI works here...).

So things are good. Marginal improvements can be made tho:

• I didn't know about prtest:full, many thanks to @rvolosatovs; perhaps worth mentioning this when a merge fails, so as to avoid multiple merge failures
• Would be nice to be able to manually re-run tests that failed, or choose specific jobs, as in chromium's review system (image attached)
• It could make sense that if a PR includes a line that removes or adds cfg(), that we select the full build matrix
• One has to click through a few things to get to build errors (the list of checks, scroll to bad job, click, scroll down in the logs enough so that they are all loaded, search for a term); perhaps some summarization is possible

But like I say things are good, and thank you and Roman for your help & patience <3