Similar to #318290 but for documentation.
All instances of the old pattern were found using `rg '\}\/bin\/' -tmd`.
Some instances were deliberately left as-is, either because it does not
make sense to convert them, or because they were in release notes.
Restore the correct behaviour prior to commit bb71feaec1
("lib.functionArgs: save a function call on every lookup")
while preserving its performance gain with best efforts.
The test asserted the exact stderr of `nix-instantiate` when coercing a
missing path. That wording is implementation-defined and subject to
change: it differs between CppNix and Lix, and it can change between CppNix
releases too.
Only check that evaluation fails and that the error names the missing
path.
Make `useGccNG` the default for every NetBSD platform, as Cygwin's already
is, rather than opting in one example system. NetBSD 11.0 is only supported
with the split set, so this comes straight after the version bump: the
monolithic `gcc` is not kept working for it.
NetBSD is low tier and has to use GCC rather than LLVM, so it is a good
place to dogfood the split set.
With the fixes that follow, GCC 16 builds `hello`, `zlib`, `bash`,
`perl`, `python3` and the NetBSD libc for it.
`x86_64-netbsd-llvm` is unaffected, since `useLLVM` takes precedence.
Assisted-by: Claude Code (Claude Opus 5.5)
retarged links to the nixpkgs manual
introduce invisible anchors in the nixos manual so visitors of old references see the migration message
removed meta narration such as for 'basic' types
descriptions for every type was copied verbatim
Releases of cuda-bindings before 13.4 are licensed under the NVIDIA
Software License for NVIDIA CUDA Python (cuda_bindings/LICENSE), which
permits redistribution subject to its distribution requirements.
Add lib.licenses.nvidiaCudaPython and accept it in
allowUnfreeCudaPredicate, so that packages depending on cuda-bindings,
such as python3Packages.transformer-engine, evaluate with the CUDA
predicate instead of requiring allowUnfree.
Assisted-by: Claude Code (Claude Opus 5.5)
The nvidia-cutlass-dsl and nvidia-cutlass-dsl-libs-base wheels ship the
NVIDIA Software License Agreement for the NVIDIA CUTLASS DSLs (identical
to EULA.txt in NVIDIA/cutlass), not a generic redistributable license.
It permits redistributing only Python source files incorporated into an
application (section 1.1d) and otherwise forbids distribution (section
2.2), so the binary wheels are not redistributable.
Add lib.licenses.nvidiaCutlassDsl and accept it in
allowUnfreeCudaPredicate, so that packages depending on the CUTLASS DSL,
such as python3Packages.transformer-engine, evaluate with the CUDA
predicate instead of requiring allowUnfree.
Assisted-by: Claude Code (Claude Opus 5.5)
