Deprecate and unlist package version 3.0.0 and change release habits

[MarcoRossignoli]

Last week we released new major version 3.0.0

Unfortunately the version has got an ugly bug has described by our amazing @daveMueller #1037 (comment)

We did a great improvement on coverage testing, today we have a way to check if we're able to recognize pattern and verify line numbers and branch numbers.
I'm very happy of the quality of our coverage.
With this version we're very close to OpenCover as reported by @JSkimming #1037 (comment) one of the best coverage solution around.

Anyway the bug is escaped from our UTs and integrations because it manifests only on multi-document project and we're testing 1 document at time(we check coverage for specific scenario and we generate 1 asm for 1 class and we tests only that class so only 1 cs file).
Also offering nightly build to our friends didn't help #1037 (comment) and I'm ok with it, it's a nightly not an "official preview".

In the spirit of "learning from our(my) errors" I propose to add one more step before official release. I think that when the new fix/improvements are "enough"(also if we have 1 big change) we should release an official preview to nuget and maybe ask to our friends of coverlet community to give it a try.

We could also open an coverlet Twitter account to ask some collaboration for preview testing.

Coverlet is becoming the choice for xplat .NET Coverage, we have a lot of download https://nugettrends.com/packages?months=12&ids=coverlet.collector&ids=coverlet.msbuild&ids=coverlet.console and users ~61K(github numbers) and we're on MS docs https://docs.microsoft.com/en-us/dotnet/core/testing/unit-testing-code-coverage?tabs=windows with the privilege of VSTest strict integration --collect "XPlat Code Coverage", for all these reason I think we need to do a step further also on this part of "chain" to improve the quality and move @tonerdo's baby to the adolescence😄

Please let me know what you think @tonerdo @petli

[Malivil]

You can add me to that list of people willing to test an official preview =)

I can't update to nightly builds but if you tag me in a preview test I can dedicate some time to run it through my solution.

[MarcoRossignoli]

Sure thx @Malivil
I apologize wasn't my intention exclude/offend/forget nobody I wrote down first names came into my mind related to my past experience on coverlet with bugs/tests etc...that "etc" was not appropriate, I'm going to update immediately.

[MarcoRossignoli]

Another idea could be add an announcement label and let coverlet users to subscribe to it when we release preview.
There we can invite with a cc:

[Malivil]

Sure thx @Malivil
I apologize wasn't my intention exclude/offend/forget nobody I wrote down first names came into my mind related to my past experience on coverlet with bugs/tests etc...that "etc" was not appropriate, I'm going to update immediately.

No worries at all. I didn't take it that way, I just wanted to put my name forward to help out as well =)

[JSkimming]

In the spirit of "learning from our(my) errors" I propose to add one more step before official release. I think that when the new fix/improvements are "enough"(also if we have 1 big change) we should release an official preview to nuget and maybe ask to our friends of coverlet community to give it a try.

Just adding my 2 cents. I'm not convinced such caution is necessary, though happy to be persuaded otherwise.

This is a test library, so not going to break runtime code. Also, by definition, coverlet's usage is 100% covered by tests. If someone receives a degraded result, like #1037, it's quite likely to be something consumers are monitoring, and therefore have the option to reject the upgrade, which is exactly what I did.

I have the same view as with libraries like xunit, Moq, Fluent Assertions, etc. When Dependabot pushes upgrades, I quickly check the build's output, and if all is good, accept.

[MarcoRossignoli]

Thanks for your opinion @JSkimming!
Understood your point on "test library" meaning.

[petli]

After thinking a bit about this I think I agree with @JSkimming. In general bugs will be found faster if they get pulled into PRs by dependabot across the ecosystem, rather than just by a subset of preview users. If people automatically also merge those PRs without reviewing them is up to them.

This particular bug, by affecting almost every single user, means that we got a bunch of duplicate issues reported which added some noise, but rather than that a bug not being found at all.

[MarcoRossignoli]

Thanks @petli

[pkrukp]

Anyway the bug is escaped from our UTs and integrations because it manifests only on multi-document project and we're testing 1 document at time(we check coverage for specific scenario and we generate 1 asm for 1 class and we tests only that class so only 1 cs file).

I have not seen any mentions about improving automatic tests. "multi-document project" is the normal case, nobody has projects with only one document. Did you consider adding such test?

[MarcoRossignoli]

I have not seen any mentions about improving automatic tests. "multi-document project" is the normal case, nobody has projects with only one document. Did you consider adding such test?

We should try to add something on that side. Not super easy btw, at the moment we test all "patterns" not whole projects because every project is different so it's hard find a way to have a complete tests scenario. We'll improve for this specific case for sure.

Thanks for the comment.

[pkrukp]

Maybe following could work and be easy to implement:

Collect coverage from some third party project. The project should be non trivial, should already have a good number of UT and a good coverage. Collect coverage from the UT before each release (or even after each commit if possible). If coverage numbers change and it was not expected, you would have to manually review why it changed.

This assumes normally coverage numbers do not change (given same version/commit of the third party project) and so it should have low maintenance cost.

As a bonus, you can track/notice performance changes.

[MarcoRossignoli]

Thanks @pkruk2 it's a good idea...we could try with some project like dotnet/roslyn, dotnet/runtime(they're using coverlet)