metadata::filter_platform fails on !amd64

[mwhudson]

Problem
metadata::filter_platform fails on e.g. powerpc64le-linux-gnu

Steps

1. install rustup
2. rustup default stable
3. git clone https://github.com/rust-lang/cargo
4. cd cargo
5. cargo test metadata::filter_platform

Possible Solution(s)

The failure is rather long (you can see it in https://launchpadlibrarian.net/526141668/buildlog_ubuntu-hirsute-amd64.cargo_0.51.0-0ubuntu1~ppa1_BUILDING.txt.gz) but the upshot is that /packages/2/dependencies/2 and /packages/2/dependencies/3 are the other way around from how the test expects (test expects alt-dep first, output has host-dep first).

Now, I'm no expert but I don't see anything in the code that guarantees the ordering of dependencies in the output. So I'm guessing this only passes on amd64 by chance. But I could be wrong!

Notes

Output of cargo version:

I observed this with c694096 and 0.51.0 built with "rustc 1.50.0 (cb75ad5db 2021-02-10)" on Ubuntu 20.04. I also saw the same test building cargo 0.51.0 on arm64, s390x and armhf with my rustc 1.50.0 packages.

[ehuss]

I don't see any errors in the linked build log, but I was able to verify with an aarch64 system.

Yea, it looks like the [target] tables are stored in sorted order. The test as written is assuming the host triple sorts alphabetically after wasm32, which is true for all x86* targets.

It'll be a bit tricky to fix, since there isn't any existing support code for ignoring the order of JSON arrays (I think it was removed in #8489). I'm guessing adding an option to ignore array order is an possibility. Another is to change that test to extract the JSON object, and selectively compare the important parts.

[mwhudson]

Oh indeed I sent a link to the wrong log. Glad you managed to repro it anyway. I guess another idea would to to sort the array in the "expected" object before comparison?