Improve union types deduplication

[reicheratwork]

Fixes issue #513
Rewrote is_same_type function to (more) correctly deduplicate union types, but only when boost::variant is used, as this does not have access by index
Added unittests to test more constructs in addition to those implemented by Jim Hague in PR #519

@banburybill , could you give this PR a look to see whether this fixes #513 for you?

[banburybill]

My specific failure case was a union with bounded and unbounded strings. This looks good to me, and the refactoring is great.

I guess it's vulnerable to someone using the templating to replace std::string with a genuine bounded string class. But I think that is probably a very niche usage, and fixing the out-of-the-box problem is far more important.

[reicheratwork]

I guess it's vulnerable to someone using the templating to replace std::string with a genuine bounded string class. But I think that is probably a very niche usage, and fixing the out-of-the-box problem is far more important.

That is also being checked for in the following code part:

const bool bstr_unique = (strcmp(gen->bounded_string_format, gen->string_format) != 0),
                  bseq_unique = (strcmp(gen->bounded_sequence_format, gen->sequence_format) != 0);

So in the case when a "normal" string is std::string and a bounded string is something else, the boolean bstr_unique is true and the bounded versions of strings will have their bound values added to the array identifying the type being compared. And therefore will "look like" different type, when compared to the unbounded versions.

The same is checked for when encountering (un)bounded sequences.

[banburybill]

Ah, I missed that bit. You're way ahead of me. Looks great.

[eboasson]

LGTM 😀 Thank you @reicheratwork !