@ralsina@_rygo6 But that's a non-argument. When I say a "low level language" I mean "close to the ISA". Assembly is almost isomorphic to the ISA, and C is close to it.
You're changing the goalpost to some meaningless, unobservable artifact of "the real program".
@ptrschmdtnlsn Yes. Chisnall is incoherently shouting at silicon wafers. It was talked about when this paper was first published in 2018 and it's still true today.
@johnmyleswhite The AFC test cited in that post requires none of the bizarro requirements Steven is requiring
If you want to live in fear of this, that's your prerogative. The vast majority will not. You can read the relevant US codes yourself, or ask your lawyer to read them to you.
@johnmyleswhite If this were true, every song, movie, and piece of literature would be constantly exposed to significant copyright risk due to their authors listening to music, watching movies, and reading books.
Copyright is built around the assumption authors are exposed to pre-existing works
@johnmyleswhite This isn't true at all, like not even a little bit.
The requirements of copyright don't care how a work was derived. Something like clean room can be an effective defense against a copyright suit, but lack of clean rooming isn't anywhere close to legally damning.
@valigo The answer turns out to be you need both. The standard is not a home for every esoteric requirement. Arguably the problem is introducing std::unordered_map at all.
std::map shows the "shape" of a map in C++, all other requirements should be met by third parties.
@valigo The original Swiss Tables talk addresses this exact point. They thought reference stability was false requirement, and it turns out it wasn't. The Swiss Table migration at Google was painful because they had to sort out all the locations reference stability was required.
@deepinthebuild BMIs are best understood as AST serializations of the contents of the interface. With headers you got the AST from reparsing the header in each consumer, for modules you need to manually instruct the compiler to perform the serialization.
@deepinthebuild Yes, entirely sensible. In the same way you can include the iostream header in a C++20 file, even though the standard library it corresponds to was compiled with C++17.
@deepinthebuild Ie, the same thing you need to do to make `import std` work with multiple consumers with different C++ standards or other BMI-incompatible conditions.
You build the BMI for the consumer, you re-use BMIs if consumers are inter-compatible.
@deepinthebuild What CMake and Xmake do, detect the incompatibility in the consumer and generate a build rule to create a compatible BMI from the interface unit.
@batatalal@yammvik CMake wasn't developed for or in collaboration with the labs, it was developed for an NIH project and stayed in the medical imaging world for most of its early life.
@tony914_@oxcrows Sure, but as long as it doesn't reproduce and distribute code in ways that violate copyright law, as long as it doesn't produce a copy, there's nothing illegal.
In the US that means a violating reproduction must be substantial, non-transformative, and not covered by fair use.
@mu_chrinovic I couldn't disagree more. The Kaleidoscope and ToyLang tutorials for LLVM/MLIR respectively are excellent and get you to a complete optimizing backend. This lets language designers focus on the language, not the pipeline.
New exploit: “xor dword [0xf80c2094], 1<<22”
Unlocks CPU microcode, the platform security processor, system management mode, and every internal processor register, all at once, on 100 million AMD CPUs. As far as I can tell can’t be fixed.
https://t.co/sgAfneFSsf