Set the same env vars as cargo both when invoking rustc and when invoking a cargo_build_script

[illicitonion]

Cargo sets this list of env vars when either invoking rustc, or cargo runing or cargo testing a binary. We should be setting them all when we invoke rustc.

Cargo additionally sets this list of env vars when running (not when compiling) a build script. We should be setting them all when we invoke a build script. We should also be setting all of the rustc env vars when we invoke a build script.

Currently we have two lists of env vars:

This list is used for when we invoke rustc:

rules_rust/rust/private/rustc.bzl

Lines 57 to 85 in e744b93

	def _get_rustc_env(ctx, toolchain):
	"""Gathers rustc environment variables
	Args:
	ctx (ctx): The current target's rule context object
	toolchain (rust_toolchain): The current target's rust toolchain context
	Returns:
	dict: Rustc environment variables
	"""
	version = ctx.attr.version if hasattr(ctx.attr, "version") else "0.0.0"
	major, minor, patch = version.split(".", 2)
	if "-" in patch:
	patch, pre = patch.split("-", 1)
	else:
	pre = ""
	return {
	"CARGO_CFG_TARGET_ARCH": toolchain.target_arch,
	"CARGO_CFG_TARGET_OS": toolchain.os,
	"CARGO_PKG_AUTHORS": "",
	"CARGO_PKG_DESCRIPTION": "",
	"CARGO_PKG_HOMEPAGE": "",
	"CARGO_PKG_NAME": ctx.label.name,
	"CARGO_PKG_VERSION": version,
	"CARGO_PKG_VERSION_MAJOR": major,
	"CARGO_PKG_VERSION_MINOR": minor,
	"CARGO_PKG_VERSION_PATCH": patch,
	"CARGO_PKG_VERSION_PRE": pre,
	}

This list is used when we invoke an already-compiled build-script (note that we use the above list when compiling the build script):

rules_rust/cargo/cargo_build_script.bzl

Lines 79 to 132 in e744b93

	env = dict(ctx.configuration.default_shell_env)
	env.update({
	"CARGO_MANIFEST_DIR": manifest_dir,
	"CARGO_PKG_NAME": crate_name,
	"HOST": toolchain.exec_triple,
	"OPT_LEVEL": compilation_mode_opt_level,
	# This isn't exactly right, but Bazel doesn't have exact views of "debug" and "release", so...
	"PROFILE": {"dbg": "debug", "fastbuild": "debug", "opt": "release"}.get(ctx.var["COMPILATION_MODE"], "unknown"),
	"RUSTC": toolchain.rustc.path,
	"TARGET": toolchain.target_triple,
	# OUT_DIR is set by the runner itself, rather than on the action.
	})
	if ctx.attr.version:
	version = ctx.attr.version.split("+")[0].split(".")
	patch = version[2].split("-") if len(version) > 2 else [""]
	env["CARGO_PKG_VERSION_MAJOR"] = version[0]
	env["CARGO_PKG_VERSION_MINOR"] = version[1] if len(version) > 1 else ""
	env["CARGO_PKG_VERSION_PATCH"] = patch[0]
	env["CARGO_PKG_VERSION_PRE"] = patch[1] if len(patch) > 1 else ""
	env["CARGO_PKG_VERSION"] = ctx.attr.version
	# Pull in env vars which may be required for the cc_toolchain to work (e.g. on OSX, the SDK version).
	# We hope that the linker env is sufficient for the whole cc_toolchain.
	cc_toolchain, feature_configuration = find_cc_toolchain(ctx)
	_, _, linker_env = get_linker_and_args(ctx, cc_toolchain, feature_configuration, None)
	env.update(**linker_env)
	# MSVC requires INCLUDE to be set
	cc_env = get_cc_compile_env(cc_toolchain, feature_configuration)
	include = cc_env.get("INCLUDE")
	if include:
	env["INCLUDE"] = include
	if cc_toolchain:
	toolchain_tools.append(cc_toolchain.all_files)
	cc_executable = cc_toolchain.compiler_executable
	if cc_executable:
	env["CC"] = cc_executable
	ar_executable = cc_toolchain.ar_executable
	if ar_executable:
	env["AR"] = ar_executable
	for f in ctx.attr.crate_features:
	env["CARGO_FEATURE_" + name_to_crate_name(f).upper()] = "1"
	env.update(expand_locations(
	ctx,
	ctx.attr.build_script_env,
	getattr(ctx.attr, "data", []) +
	getattr(ctx.attr, "compile_data", []),
	))

Both of these lists are currently incomplete, as we've generally added stuff as-needed. Additionally, the rustc_env list contains some vars which should actually only be in the build_script list (e.g. CARGO_CFG_TARGET_ARCH and CARGO_CFG_TARGET_OS).

We should:

1. Make the build script env vars be a superset of the rustc ones, by calling _get_rustc_env from _build_script_impl. This will both help keep the lists in sync, and avoid duplicating things like the processing of version numbers.
2. Move any build_script-specific env vars out of _get_rustc_env
3. (Optional): Add more env vars pro-actively, rather than reactively when someone finds some third-party code doesn't build.

[bbstilson]

Related to:

Add more env vars pro-actively, rather than reactively when someone finds some third-party code doesn't build.

I've hit an issue compiling a sqlx query! macro as the CARGO env var is not set by rules_rust and is required by the macro

Any interim suggestions about how to get around this would be appreciated 🙂

[illicitonion]

$CARGO is definitely a tricky one. In the short term, you can probably add an annotation setting build_script_data to @rules_rust//rust/toolchain:current_cargo_files and setting build_script_env to {"CARGO": "$(execpath @rules_rust//rust/toolchain:current_cargo_files)"}. But in general, any time a build script tries to use cargo directly, it's likely to be doing something that won't be exactly compatible with how Bazel runs things.