treewide: remove -march=native

[thiagokokada]

Description of changes

-march=native breaks purity, so let's remove it.

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 22.11 Release Notes (or backporting 22.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

[PedroHLC]

-march=native breaks purity, so let's remove it.

Not only that, the generated code will expect everybody that downloads it from caches to have the same processor as the builder, we don't use server's CPUs, so probably it makes performance even worse.

[thiagokokada]

-march=native breaks purity, so let's remove it.

Not only that, the generated code will expect everybody that downloads it from caches to have the same processor as the builder, we don't use server's CPUs, so probably it makes performance even worse.

From a performance stand point -march=native may increase performance even considering that Hydra != host machine, because we use very conservative options right now: https://github.com/NixOS/nixpkgs/pull/202526/files#diff-e32d2e727a496a1da28a670f938cf5b76322be57c4f86b1d5a6f29eb2c5dead6

But of course, it is impossible to know since it would need tests.

[NickCao]

These impure flags should have been handled by the cc-wrapper:

nixpkgs/pkgs/build-support/cc-wrapper/cc-wrapper.sh

Lines 108 to 121 in 20193b6

	# Clear march/mtune=native -- they bring impurity.
	if [ "$NIX_ENFORCE_NO_NATIVE_@suffixSalt@" = 1 ]; then
	kept=()
	# Old bash empty array hack
	for p in ${params+"${params[@]}"}; do
	if [[ "$p" = -m*=native ]]; then
	skip "$p"
	else
	kept+=("$p")
	fi
	done
	# Old bash empty array hack
	params=(${kept+"${kept[@]}"})
	fi

[thiagokokada]

These impure flags should have been handled by the cc-wrapper:

nixpkgs/pkgs/build-support/cc-wrapper/cc-wrapper.sh

Lines 108 to 121 in 20193b6

	# Clear march/mtune=native -- they bring impurity.
	if [ "$NIX_ENFORCE_NO_NATIVE_@suffixSalt@" = 1 ]; then
	kept=()
	# Old bash empty array hack
	for p in ${params+"${params[@]}"}; do
	if [[ "$p" = -m*=native ]]; then
	skip "$p"
	else
	kept+=("$p")
	fi
	done
	# Old bash empty array hack
	params=(${kept+"${kept[@]}"})
	fi

Nice to know, but I think it is still worth to remove those just to make sure that nobody will follow the same.

[PedroHLC]

Result of nixpkgs-review pr 202687 run on x86_64-linux 1

2 packages built:

• glava
• opentrack

[github-actions]

Successfully created backport PR #202693 for release-22.11.

[zhaofengli]

These impure flags should have been handled by the cc-wrapper

I wasn't aware of this behavior until it was mentioned on Matrix today. It affects usage in an impure mkShell environment as well, which can be surprising/undesirable. I opened #203675 so we can discuss about this.