Linux - Conan prev-prev is allowed to fail

[uilianries]

The CI job conanprevprev is unstable, it fails frequently because runs Conan's Development.

The CI job conanprevprev is unstable, it fails frequently because runs old old conan version, which is not required by some recipes.

It should not be considered as an important branch here, as all features are not matured.

[danimtb]

I am not sure about this change. Maybe we want to know that it fails in order to fix conan develop as soon as possible.

In the case we want to avoid develop, I'd rather remove it from the CI. WDYT @jgsogo?

[jgsogo]

I think it is important to keep this check... and realize that a commit to Conan repository broke the example. If we are breaking an example there are two options:

• We have introduced a regression in Conan
• The example was not using stable features from Conan
• (recipe in ConanCenter changed and we need to update the example)

All of them, IMHO, manifest a problem that should be fixed ASAP.

---

What is the scenario we want to bypass here?

[uilianries]

The latest failed build on master branch (2 days ago): https://travis-ci.org/github/conan-io/examples/jobs/756008647

It reports both Folly and Imgur can not be installed, because requires Conan 1.32. That's correct, but prev-prev:

Last Conan version found: 1.33.0
 - prev is 1.32
 - prevprev is 1.31

We could remove Folly or even disable, but when 1.34.0 be available, 1.32 will be prev-prev and acceptable for Folly.

[jgsogo]

In that case, maybe we should add something as we do with ConanInvalidConfiguration in C3i. We can capture that error and not fail the CI.

I mean, instead of thinking about something like this (which is needed in case of new features)

examples/.ci/run.py

Line 57 in e8c84db

version.parse("1.29.0"): [

we can add a try/except around this line to capture that specific error (conan_minimum_required) and not fail

examples/.ci/run.py

Line 195 in e8c84db

result = subprocess.call(build_script, env=env)

Does it make sense?

[jgsogo]

I like it! And the CI will like it too 😄

[uilianries]

I can't see any error on AppVeyor. Timeout maybe? Restarting ...

[uilianries]

subprocess.run is Python 3.5, but it doesn't work here, probably the patch version is incompatible. https://docs.python.org/3.5/library/subprocess.html#subprocess.run

[jgsogo]

There is a reason to use lower Python 3.6 (we can move away from 3.5, but let's keep it at 3.6): https://github.com/conan-io/tribe/blob/main/design/003-codebase-python.md

The plan is to move this CI (and the rest) to our Jenkins now that @czoido has migrated Conan-TestSuite. If it is broken in Appveyor/Travis, we can focus on moving to our CI and then take care of this issue.

[uilianries]

I agree, I'll change to support 3.5 again

[jgsogo]

@uilianries Python 3.6 is enough (if 3.5 fails for any reason)

[uilianries]

@jgsogo It's totally possible emulate check_output. Reading Python docs, it was introduce in 3.7, that's why older versions complained. Anyway, Python 3.5 should work again.

[uilianries]

After many tries, our current villain is brew upgrade pyenv command. Travis takes more than 50 minutes only to run before_install step, resulting in a timeout error. Travis offers caching feature to speed up the installation, but first we would need to pass all steps and cache, but unfortunately 50 min is not enough.

As OSX build are a pain in the neck for us, we could migrate Example to Jenkins. WDYT @jgsogo @danimtb ?
If yes, do we have a template or example for Jenkinsfile?