Assuming the regressions are similar to the one intests/expected/cpp/31593-sf593.cpp: shouldindent_continue avoid incrementing the indentation when "nested statements are on the same line"?

The other modified "expected" .cpp files should not contain more indentation than necessary I think, could you confirm that?

Yes, the regression I observed are similar to the one intests/expected/cpp/31593-sf593.cpp. If you need, I can provide some examples.

Regarding the other modified tests, it could probably be a subjective thing. For exampletests/expected/cpp/60055-issue_3116.cpp, the body of the lambda function is intended twice compared to theB function. While technically there are three levels of indentation (A -> B -> lambda body) and so it would explain the extra indentation, on the other hand it does look a bit awkward. What's your take on it?

Yes, the regression I observed are similar to the one intests/expected/cpp/31593-sf593.cpp. If you need, I can provide some examples.

It would be helpful, thanks!

Regarding the other modified tests, it could probably be a subjective thing. For exampletests/expected/cpp/60055-issue_3116.cpp, the body of the lambda function is intended twice compared to theB function. While technically there are three levels of indentation (A -> B -> lambda body) and so it would explain the extra indentation, on the other hand it does look a bit awkward.

It does look awkward indeed, but it is consistent with how lambdas are indented normally. In this case, the) of B is on the same line with the} of its lambda parameter, and that is what makes it look weird, I think. Had the closing parentheses been on its own line, it would have looked nice and correct:

A(    B([] (const std::string &s) -> bool {            s = "hello";            return true;        }    ), 1);

Moreover, if you add extra parameters to B, with this PR you will get something like this, which I think it is correct:

A(    B(foo,        bar,        [] (const std::string &s) -> bool {            s = "hello";            return true;        }    ), 1);

Compare it with the current behavior; doesn't this look even weirder with those parameters squashed to B()'s level?

A(    B(foo,    bar,    [] (const std::string &s) -> bool {        s = "hello";        return true;    }    ), 1);

Like you said, it may be a subjective thing, and another option to adjust for personal preferences would be needed I guess.

I have attached an example showing some regressions, including the config used to test it. Note that I use negative values for indent_continue.

Re lambda indentation, I see the point in your comment. Also there is an interesting idea proposed in issue#3815, so the solution may be to merge this PR (once the other regressions are fixed) and then add an option as proposed by@PoeticPete.

pr_3816.zip