I get these build warnings with current clang SVN trunk (SVN revision 343613, technically version 8.0 but probably introduced pretty recently): ==== warning: gfx/gl/GLContext.cpp:325:20 [-Wc++2a-compat] aggregate initialization of type 'mozilla::gl::GLContextSymbols' with user-declared constructors is incompatible with C++2a warning: gfx/gl/GLContext.cpp:2054:16 [-Wc++2a-compat] aggregate initialization of type 'mozilla::gl::GLContextSymbols' with user-declared constructors is incompatible with C++2a warning: gfx/gl/GLContext.h:3504:33 [-Wc++2a-compat] aggregate initialization of type 'mozilla::gl::GLContextSymbols' with user-declared constructors is incompatible with C++2a ==== This is about this declaration: > GLContextSymbols mSymbols = {}; https://dxr.mozilla.org/mozilla-central/rev/56b988a937689d5599400afa59b72c390b40abf2/gfx/gl/GLContext.h#3504 ...and this assignment: > // If initialization fails, zero the symbols to avoid hard-to-understand bugs. > mSymbols = {}; https://dxr.mozilla.org/mozilla-central/rev/56b988a937689d5599400afa59b72c390b40abf2/gfx/gl/GLContext.cpp#325 ...which it complains about because we do have a user-declared constructor -- THOUGH, we only declare it insomuch as to delete it: > struct GLContextSymbols final > { > GLContextSymbols() = delete; // Initialize with {}. https://dxr.mozilla.org/mozilla-central/rev/56b988a937689d5599400afa59b72c390b40abf2/gfx/gl/GLContextSymbols.h#30 This deleted-constructor scenario almost seems like it should be an exception to the warning, but apparently it was considered in the proposal -- see section 1.1 here: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1008r1.pdf Quoting section 1.1 of that paper, which gives an example like ours (deleted constructor, aggregate instantiation with {}): "Clearly, the intent of the deleted constructor is to prevent the user from initializing the class. However, contrary to intuition, this does not work: the user can still initialize [instances of the class] via aggregate initialization because this completely bypasses the constructors." In our case, the "clearly" thing here is not actually accurate. But maybe we're abusing this C++ feature? (A bit more info at https://clang.llvm.org/cxx_status.html -- see row for "Prohibit aggregates with user-declared constructors" there, which links to the PDF noted above.) This issue seems to have been introduced by bug 1411626 (which deleted the constructor) and, to a lesser extent, bug 1482974 (which added some of these "= {}" lines.) Setting both of those as dependencies.

Yeah, I deleted the constructors to force aggregate initialization, as they mention, to prevent uninitialized values. It looks like they're taking this pseudo-annotation away. For the curious, it seems like the compelling issue here is that there's no easy way to forbid aggregate construction on a class before this proposal. (You would have to include a dummy non-default-constructable member object though, I think? And even then they could agg-construct /that/)

I reported it in bug 1494603 -- It was caused by this upstream change: https://github.com/llvm-mirror/clang/commit/de8e63fd00e24e58a5ca05007cf7954916a6e6b9 --

I played around some, and I no longer see a way to encourage zero-initializing, besides static analysis to lint for lack of initialization.

Pushed by jgilbert@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/af0263577e4d Aggregates in c++2a can't have deleted ctors. r=dholbert