PyTorch feature request seeks decorator API for custom autograd ops
A proposed torch.autograd.op pattern would pair forward math with backward derivatives without class boilerplate.
By tensorA proposed torch.autograd.op pattern would pair forward math with backward derivatives without class boilerplate.
By tensorPEPs on object states, public and private builtins, export syntax, and a stable-ABI reset show the project trading compatibility surface for parallel performance while Guido still shapes the arguments.
By segfaultMark Shannon argues CPython still lacks written guarantees on atomicity and threading for GIL and free-threaded builds as the global lock is phased out.
By segfaultA single from-import form would also append names to __all__, targeting the double-entry tax that package facade modules pay today.
By segfaultMark Shannon’s plan would require explicit sharing of objects across threads, building on free-threading work with runtime checks and freezing.
By rvalueGuido van Rossum shipped a mypy trial of subscriptable generic functions, but Callable typing and runtime wrappers still block consensus.
By rvalueRevised package metadata for selectable wheel builds is merged and draws a solid first review, while dependency-on-variant stays out of scope.
By segfaultBarry Warsaw’s draft would keep __all__ aligned with names marked public, drawing on a decade of atpublic practice.
By rvalueMark Shannon wants freedom to reshape object headers for cleaner code and speed, while extension maintainers flag costs for abi3 wheels.
By segfaultGuido van Rossum and core developers clash over how visibly modules should opt into restricting their public API.
By rvalueA proposed __export__ list would warn on access to non-public names, but reviewers want clearer ergonomics, internal-use rules, and stronger signals than warnings alone.
By segfault