Amend `make publish` to support removing dev-dependencies about rust_icu HOT 8 CLOSED

Alternative: is it possible to somehow arrange the crates so that the circular dependency is removed altogether?

from rust_icu.

We could theoretically move the tests that cause the cycle out into a new crate, but that seems like a risky practice.

from rust_icu.

To short-circuit the search, which crates are forming a cycle? Perhaps indeed we can move things around so that we don't need to patch things up to publish them.

from rust_icu.

It's ucal and udat, because

• udat has a method that takes a &ucal::UCalendar

• In ucal tests, we're parsing dates (udat::UDateFormat)

from rust_icu.

I removed my earlier message, since I misunderstood the options we had.

So now I see and understand how this came about. Given that rust_icu_udat is a higher-level crate than rust_icu_ucal, it is wrong for tests in rust_icu_ucal to require a loop back. So I kind of side with the crates.io objecting in the end; even though strictly speaking dev deps could be treated separately.

In retrospect, I see that my review recommendation to use date parsing for these few tests have caused this loop back. I wasn't aware that this would be an issue.

Would it be very horrible to lift these tests into rust_icu_udat, thus removing the dependency back-link? Yes, we'd be testing UCalendar elsewhere, but it's probably better than fiddling with the release process.

It seems like a better option compared to moving tests to a different crate, since test are still coupled with one of the two crates that are effectively under test. WDYT?

from rust_icu.

Could we just avoid the UDateFormat dependency and go back to using hard-coded timestamps with comments?

from rust_icu.

Could we just avoid the UDateFormat dependency and go back to using hard-coded timestamps with comments?

Yes, that'd work too. Feel free to make a pull request to that effect. I'll remove #101 .

from rust_icu.

One more thing: is there a way to configure cargo to alert us to such dep cycles?

from rust_icu.