Use the strict access API

[robUx4]

So far the code was built without the define. But since we can change the API for 2.0 we should switch to a stricter API that makes things less public.

This requires libmatroska to merge Matroska-Org/libmatroska#95

We might be able to add some constexpr here and there despite using methods.

[mbunkus]

This PR completely breaks compilation of MKVToolNix. I'm still evaluating if I can convert all of that to only using the strict API.

In the meantime, please take a look at this new warning that clang emits now:

In file included from src/common/debugging.cpp:19:
lib/libebml/ebml/EbmlDummy.h:59:9: warning: binding dereferenced null pointer to reference has undefined behavior [-Wnull-dereference]
        EBML_CONCRETE_DUMMY_CLASS(EbmlDummy)
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/mosu/prog/video/mkvtoolnix/lib/libebml/ebml/EbmlElement.h:194:69: note: expanded from macro 'EBML_CONCRETE_DUMMY_CLASS'

Thanks.

[mbunkus]

Another issue that needs a fix:

EbmlElement defines a function called Context(). EbmlMaster defines a private member variable Context. This makes it impossible or at least very, very difficult to call the Context() function on an EbmlMaster and derived classes. MKVToolNix does indeed need this.

[mbunkus]

And while we're at it, please add a GetId() const member function to EbmlElement that does the same as operator EbmlId(). Using a member function instead of all those bloody operators is much cleaner style & muss less confusion for people not as familiar with the code.

[mbunkus]

I've successfully converted MKVToolNix to using only the strict API stuff (a lot via the provided macros to still compile with old library versions which don't provide all the required public functions directly, but sometimes via indirections such as Generic()).

What I'd like to see before merging are the things I've already pointed out:

1. Fix the undefined behavior the compiler warns about
2. Resolve the ambiguity between the Context() function in EbmlElement and the Context member variable in EbmlMaster (e.g. by renaming the function to GetContext() which would be more in line with the other Get…() functions)
3. Provide proper functions for all operator… conversion operator functions

[mbunkus]

Rebased onto current master

[mbunkus]

I'm working on changes on top of this branch that address the points I made.

[robUx4]

OK, not sure when I'll have time to work on this, so feel free to make any progress.

There are some macro helpers that are useful to switch from strict to non-strict, so you probably use that for the time being so your code compile with both variants.

[mbunkus]

I've just pushed a handful of commits that implement everything I wanted. Please review when you find the time.

I'd also like to push a cosmetics commit fixing indentation (we had this at 2 spaces at one point, but now it's becoming a really bad mixture of 2, 4 and other amounts of spaces again), but that'll make the diff way less obvious. Therefore I'm holding off on it until we get this PR merged.

[robUx4]

The changes look fine and it does build with EBML_STRICT_API

[mbunkus]

Trying to build MKVToolNix with this branch & the latest master from libMatroska fails during the linking stage:

ld.lld: error: undefined symbol: libebml::EbmlDummy::ClassId()
>>> referenced by ebml.h:124 (src/common/ebml.h:124)
>>>               r_matroska.o:(kax_reader_c::read_headers_internal()) in archive src/input/libmtxinput.a
clang-16: error: linker command failed with exit code 1 (use -v to see invocation)

This is due to the function not having an implementation anymore. Easily fixed by replacing the declaration with:

    static const EbmlId & ClassId() { return DummyRawId; }; 

Otherwise I'm still fine with the changes. Thanks!

[robUx4]

Thanks for testing (after this long there was invitably things that might break). It should work with the latest push.

[mbunkus]

Works, thanks!