Michael137 wrote: > > LGTM > > Been wondering if we should follow the same principle as we did with > > `std::string` and simulate the layouts for all the STL types, keeping a > > record of them as they change. So we know we don't break the old layouts. > > Of course it has a separate maintenance cost > > Yeah, I think the biggest hurdle is the initial work in getting a stripped > down version of the class in place. The problem is that the container types > use a lot of template metaprogramming to implement various optimizations like > empty base class and whatnot. Copying that verbatim would make the test huge > (and in the case of libstdc++ its probably not even possible), and recreating > it from scratch is not trivial. I believe that after I wrote the std::string > test this way I tried to also do the same for some of the container types (it > may have actually been variant), and just gave up after I saw hold long it > would take. > > Nonetheless, I think there's definitely value in tests like this (for one, > going through the implementation in this way makes it obvious how many corner > cases need testing), and it's nice to see there's interest for these tests. > I'll keep that in mind when working on future changes.
Agreed, they can get pretty unwieldy. I'll see what the `std::(unordered_)map` simulator could look like since we're making changes there at the moment https://github.com/llvm/llvm-project/pull/97568 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
