Build and test in a matrix

[wiktor-k]

No description provided.

[Jakuje]

To catch the #282, we would need to also run the tests on this target.

[wiktor-k]

To catch the #282, we would need to also run the tests on this target.

In this case, to be precise we don't actually need to run tests we just need to compile them. cargo check ... --all-targets checks tests and examples too (see cargo targets) and IMO we were missing that.

I think we don't need to strictly cargo build since we throw away the artifacts so check is sufficient. Running the tests may also be a good idea (but separate from the check) although I'm not sure how does it work (I don't think it emulates, say, aarch64 to run the test).

I've updated the matrix job and it's clearly visible it's not just that particular 32bit target that was failing.

To be honest I'd move the ci.sh to yaml since it's pretty straightforward and GitHub shows separate sections for all steps but I don't know how much is @hug-dev attached to that 😉

[wiktor-k]

I've cherry-picked your commits from the other PR... let's see how many tests will be green now :)

[Jakuje]

Sounds like all of them were fixed :)

I like the idea of running on the whole matrix. It makes it easier to catch all possible oddness (but really the most common one is the ulong size which demonstrates on 32b arches). The linux-x86_64 is missing though.

On the other hand, I would probably be interested in running the test on 32b arch too (can be separate PR based on what I did in #283). We can probably do it easily only on i386 though. We might get the arm64 runner as mentioned in #285 and we might be able to run there the arm32 binaries, but I never played around this and I an skeptical it will show other errors than the i386 ones.

[Jakuje]

I would probably change this to run only if the arch triplet matches the system we run on. Either simply by running this when TARGET is x86_64-unknown-linux-gnu . Or somehow dynamically if we will use different runners.

Otherwise we run the same tests 13 times in the same configuration here.

[Jakuje]

we are missing the "default" target here now. Something like x86_64-unknown-linux-gnu, which was probably excluded from the previous list as it was running when there was no --target provided (and which now builds when we call cargo test.