Wouter van Oortmerssen
Wouter van Oortmerssen
@stewartmiles 1. I think ARMv5 is the last mainstream architecture that couldn't read unaligned with a single instruction, or am I missing something? Given that this would be a forward...
@adsharma We already have this? You can annotate a field as `key` (and the other fields are effectively the value) and then use it with binary search lookups if you...
@vglavnyy `std::bit_cast` takes a `const T &`, which you cannot legally construct an unaligned version of, so doesn't seem that useful? To be fully correct we'd have to go via...
@AustinSchuh but force defaults potentially wastes a lot of space, so I don't think it's a particularly good solution. Particularly calls like `Create` which serialize all fields are assuming default...
@mzaks yes, as opt-in with some kind of `allow_cycles` it could work, if there's enough demand for such a thing. We'd need signed offsets in general then though, or generate...
@adsharma multiple keys would also require multiple sort orders to be efficient, or indices separate from the vector containing the values.. the current solution is simple because it works almost...
@oberstet @mikkelfj @krojew I'd say a `struct` makes for a pretty good typedef already: `struct UUID { v:[uint:4]; }` We could consider native `typedef`, but I am not sure what...
@dbaileychess - Yup we can allow specifying the size of the union type, and maybe default it to 16 to protect those that are going to otherwise run into trouble...
@mikkelfj there would be no branching. The features I am proposing are all at codegen time, at runtime, the accessor code will only expect one kind of format, much like...
@mzaks interchanging signed and unsigned is not quite as easy in C++ as you make it sound.. you can't just make it a typedef an hope other code keeps working....