CI: Fix windows tests disk space failure

[cnpryer]

Fixes: #2783

Problem

The windows tests were failing due to insufficient disk space because of cargo management on the D: drive. Unlike other OS options, the windows GHA option uses the D: drive which has less capacity than the C: drive.

Solution

Use Swatinem/rust-cache@v1 or similar logic to manage caching deps

NOTE: I modified the on to trigger this GHA in order to demonstrate the passing test. Before merging I can pull this out.

Other

While looking into this I noticed:

• windows-2019 is often associated with random issues (changed to windows-latest in this PR)
• The first pass I submitted used a hacky workaround based on info from the issue I linked here

I'm curious how well this will hold up over time since nightly is used here. In case there's a need to fallback to the workaround, something along these lines (+ some cleaning logic) could probably be used:

jobs:
  test-rust:
    runs-on: windows-latest
    steps:
      - name: Install latest Rust nightly
        uses: actions-rs/toolchain@v1
        with:
          toolchain: nightly-2022-04-01
          override: true
      - name: Checkout branch
        run: |
          git clone ${{ github.event.pull_request.head.repo.full_name }} C:/polars --depth 1
      - uses: Swatinem/rust-cache@v1
        with:
          working-directory: C:/polars
      - name: Run tests
        working-directory: C:/polars
        run: |
          rustup override set nightly-2022-04-01
          git checkout ${{ github.event.pull_request.head.sha }}
          set RUSTFLAGS="-C debuginfo=0"
          cd polars && make test && make integration-tests

Note the cloning and checkout need to be further validated.

Other Resources:

• Cache usage meta tracking issue rust-lang/cargo#7150

[ritchie46]

Thanks a lot! So as I understand it, a hack was copying to C:/ and after updating to latest it wasn't needed anymore?

[cnpryer]

The workaround fixed it but probably only because the c drive had more capacity. Upgrading to windows latest wasn't the fix, I threw that in there because I kept seeing 2019 be related to other issues (this hopefully prevents problems in future).

Swatinem/rust-cache@v1 looks like it's the more robust fix.

The Swatinem/rust-cache@v1 action cleans up cargo management by adding "smart caching". So we no longer hit capacity on The d drive (we'll see how long this lasts).

A bunch of projects use it for the same problem that we saw here. The only catch I think is that with nightly the caches will be dropped with each new version that's used. So I'll keep my eye on this. Not sure if it cleans dropped caches.