Jukka Lehtosalo
Jukka Lehtosalo
Status update: @TH3CHARLie has replaced all old-style primitives. Unboxing/boxing and type checks still need to be migrated, and we need to do something about wrapper functions and various structs related...
I've done some analysis, and here are some of the most promising improvements: - [ ] Combine multiple simple ops into a single C statement (#749 covers some of this)...
These are the officially supported platforms for which we have some CI: * Python 3.7 and later * Windows, macOS, Linux * x86-64, 64-bit ARM (currently only Apple Silicon), and...
This should also work for `math.pi`, etc., that get translated to float literals (see python/mypy#15324).
I agree that we'll likely need to do something about the size of generated code pretty soon. Once we are able to compile much of mypy we should look at...
Yeah, we could do better here. C compilers are pretty good at optimizing away redundant checks, though, but at least simplifying the code could speed up compilation.
I'm open to a PR if it doesn't make the code significantly harder to read or maintain, and isn't too big. (This is something that I'd definitely like to improve...
The simplest way to implement might be to trick the C compiler into optimizing tail calls. We'd need to omit the generation of a traceback entry and possibly adjust the...
A simple peephole optimization pass could perhaps be sufficient for a good fraction of cases. An unbox immediately followed by a box could be replaced with a type check and...
Clarification: The simple approach discussed above also requires at least def-use information.