For my part the current state is ok. @alexfh and @aaron.ballman but should do their review before committing.
I would be interested in a exampleoutput for any real project.

Informationen my part think that both switch and else should stay here for now. Running it over llvm gave really a lot of warnings, and that a rather good codebase.

In D37808#869810, @Eugene.Zelenko wrote:

In D37808#869612, @JonasToth wrote:

In D37808#869602, @Eugene.Zelenko wrote:

I think will be good idea to extend -Wswitch diagnostics.

Ok. But it will introduce new warnings to llvm codebase itself. I prepare some example output i found right now.

If number of them will not be huge, it'll be worth to fix before extended -Wswitch will be committed.

I see, forgot that.
The doc were the interesting part for me, I trust that you test the code correctlty :)

I built the patch locally, here is my output:

I did read through it now. In general, the code is sound with my understanding of the complexity metric and there is a almost one to one wording to the document, which is nice.
Since we chatted in IRC directly, i would give a short summary to avoid forgetting what we talked about :)

Looking up to the Visitor, i will do this next.

In D37808#869823, @lebedev.ri wrote:

What about GNU extension case 1 ... 3: ?

In D37808#869602, @Eugene.Zelenko wrote:

I think will be good idea to extend -Wswitch diagnostics.

fixed first stuff, sorry for noise.

• fix the first issues i found

I added a clarifying comment for some outcommented code still in the patch. I would like to have a bit of discussion on it.
Maybe some parts of the check could be warning-area, i just implemented it here to have it on one place.

Recommited with fix in rL313067

• make matcher for ctor initialization more readable

This Patch broke the buildbot for vs-2015.
I will revert, when i figured out how to do this in svn :/

review in D36354

commited in rL313043

• fix, that deleted destructors are caught as problem as well
• clarify one comment, duplicated avoid was misleading

• fix typos

• address review comments, especially improve diagnostic for return types, that should be owner but arent

• [Doc] add list of aliased checks in release notes

There is an exception to the general rule (EXP36-C-EX2), stating that the result of malloc and friends is allowed to be casted to stricter alignments, since the pointer is known to be of correct alignment.
Could you add a testcase for this case, i think there would currenlty be a false positive.

• fix minor doc errors

• Adressed some review comments.
• Added testcases, where the owner<> annotation might be hidden by a typedef or using, like int using heap_int = gsl::owner<int*>;

• Merge branch 'master'
• mention current problems with typedefs of owner<> and adjust test cases

ping

Thank you for fixing this.
Why is std=c++11 necessary? I think this check should apply to older standards as well.

language in the docuementation improved

• fix typos and bad language

• adjusted expected diagnostics
• adjust diagnostics and remove private inheritance cases

• adjusted the diagnostic for member variables

• added inheritance cases, adjusted matcher is required

• added additional testcases, like @aaron.ballman requested
• fixed diagnostics

alright. i thought it would do something different, but the enforcement to handle all notes is a good thing. forget what i wrote :)

@aaron.ballman is it ok for you as well? otherwise i would commit it.

Note can be handled right now as well. E.g.

// CHECK-MESSAGES: [[@LINE-2]]:3: note: type deduction did not result in an owner

sry. misconfigured herald :(

addressed some issues, not all yet

• fix spelling and grammar errors
• adjust unit test to new diagnostics

struggling with arc...

fix patch, to diff against master again

• removing trailing comments

exception checking seems to end in the front end for such cases, so nothing to do here anymore. reimplement if there is need later.

• fix indendation in testcase

• get up to date with master
• added testcase for enums

• handle typedefs correctly now

as noted in one comment, there are false positives with the current matcher, since it does not look through typedefs/aliases. is there a way to do this easily?

• adjusting diagnostics

remove unnecessary outcommented lines

i think the diagnostics are better now, i added the expected and received type everywhere i could.

• adjust the diagnostics to include the types

• remove duplicated release note

• test case fixed expected warning
• address review comments from aaron

the new code to fix the issues is in D37060, so i will close this revision.

I added my thoughts for the check. Many variables in your code could be const, i didn't mention all cases.

added more generic cases, and improved the matcher, thanks to @lebedev.ri

plain macros seem to be handled correctly. Only troublemaker are templates.

i did work on the generic code with macros and templates.
i could not reproduce all issues, but in general template exception cause problems with diagnostics and with catching these cases.

• master remerged into this branch
• improve testcases for generic code

Did I run it over clang or athletic like that? I would be interested in an overall output.
I added my thoughts. This check seems very valuable.