Ryan May

@kgoebber good point. We could certainly reimplement with two calls to `.sel()`. I think that will have minimal performance impact.

@sgdecker Yes, the units *should* trigger an error in that case.

Attn: @DennisHeimbigner

If you used a random temp directory for the intermediate test output that would let the scripts avoid stomping on each other.

Adding another case here, there's also a conflict between those two same test scripts and `tst_special_atts`: ``` 2022-11-21T23:34:24.6009450Z 58: Test command: /usr/bin/bash "-c" "export srcdir=$SRC_DIR/ncdump;export TOPSRCDIR=$SRC_DIR;$SRC_DIR/build-shared/ncdump/tst_nccopy4.sh tst_nccopy4.sh " 2022-11-21T23:34:24.6010479Z 58:...

And then another with `tst_group_data` in `ncdump`: ``` 2022-11-21T23:34:22.6601112Z 58: Test command: /usr/bin/bash "-c" "export srcdir=$SRC_DIR/ncdump;export TOPSRCDIR=$SRC_DIR;$SRC_DIR/build-shared/ncdump/tst_nccopy4.sh tst_nccopy4.sh " 2022-11-21T23:34:22.6602041Z 58: Working Directory: $SRC_DIR/build-shared/ncdump 2022-11-21T23:34:22.6603354Z 58: Test timeout computed to...

Ah, I see the reason we don't see such issues here isn't that we're not relying on it, it's that they're run with `--rerun-failed` (at least that's why I assume...

I'd also argue that relying on side effects of tests instead of having that as an explicit separate dependency (e.g. a fixture/setup) is a problem with the tests that should...

I agree that conda-build is probably the right place, but could probably extend to any included file if the content is identical.

The code parsing the return from the Wyoming web service is here: https://github.com/Unidata/siphon/blob/7c24f4003b738bd88d8bcf965b39992d1ee97c49/src/siphon/simplewebservice/wyoming.py#L70-L125