Use TARGET_CC env variable to enforce clang when building C code

Makefile

@@ -12,9 +12,9 @@ CUSTOM_RUSTFLAGS := --cfg debug_assertions
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
-# Tweak this to change the clang version to use for building C code
-LLVM_SUFFIX := -16
-CLANG := clang$(LLVM_SUFFIX)
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set, a single contract will be built instead of all contracts
 CONTRACT :=
 # By default, we would clean build/{release,debug} folder first, in case old
@@ -28,7 +28,6 @@ export CUSTOM_RUSTFLAGS
 export TOP
 export CARGO_ARGS
 export MODE
-export LLVM_SUFFIX
 export CLANG
 export BUILD_DIR
 

README.md

@@ -39,6 +39,7 @@ Generally speaking, the directory structure, is simply a standard Rust workspace
 * `deps`: All git submodules should go here.
 * `tests`: Top level contract tests. Typically one would want to build full CKB transactions including the smart contracts in development, then run them in CKB's verifier for assurance of behaviors.
 * `docker`: Sample docker files for reproducible build. You might or might not need this in your project.
+* `scripts`: Utility scripts that will be leveraged by makefiles.
 * `Makefile`: Top, workspace level makefile for firing up commands.
 
 Note for the sample here, I am not locking Rust toolchain versions, since I believe latest stable Rust is good enough to build smart contracts. However for individual projects one is working on, it might or might not make sense to lock Rust toolchain version for stability.
@@ -61,6 +62,7 @@ $ make build CUSTOM_RUSTFLAGS=""          # release build without debug assertio
 $ make build CARGO_ARGS="--verbose"       # release build with `--verbose` attached to cargo command, you can use other arguments accepted by cargo
 $ make build CONTRACT=minimal-log         # build a single contract
 $ make build CLEAN_BUILD_DIR_FIRST=false  # keep old untouched binaries
+$ make build CLANG=clang-17               # use a specific clang version to build C code
 ```
 
 You can also combine all the arguments here, suppose in a previous build you have build all the binaries, now you only want to build minimal-log binary. Doing `make build CONTRACT=minimal-log` will erase other binaries, you can do `make build CONTRACT=minimal-log CLEAN_BUILD_DIR_FIRST=false` to both build the minimal-log binary, and also keep the old ones untouched.

contracts/atomics-without-a/Cargo.toml

@@ -5,7 +5,7 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"
 log = { version = "0.4.20", default-features = false }
 bytes = { version = "1.5.0", default-features = false }
 

contracts/atomics-without-a/Makefile

@@ -15,6 +15,9 @@ FULL_RUSTFLAGS := -C target-feature=+zba,+zbb,+zbc,+zbs,-a $(CUSTOM_RUSTFLAGS)
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set to some value, the generated binaries will be copied over
 BUILD_DIR :=
 # Generated binaries to copy. By convention, a Rust crate's directory name will
@@ -30,7 +33,7 @@ endif
 default: build test
 
 build:
-	RUSTFLAGS="$(FULL_RUSTFLAGS)" \
+	RUSTFLAGS="$(FULL_RUSTFLAGS)" TARGET_CC="$(CLANG)" \
 		cargo build --target=riscv64imac-unknown-none-elf $(MODE_ARGS) $(CARGO_ARGS)
 	@set -eu; \
 	if [ "x$(BUILD_DIR)" != "x" ]; then \

contracts/legacy-c-dependency/Cargo.toml

@@ -5,5 +5,5 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"
 blake2b-rs = "0.2.0"

contracts/legacy-c-dependency/Makefile

@@ -15,6 +15,9 @@ FULL_RUSTFLAGS := -C target-feature=+zba,+zbb,+zbc,+zbs $(CUSTOM_RUSTFLAGS)
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set to some value, the generated binaries will be copied over
 BUILD_DIR :=
 # Generated binaries to copy. By convention, a Rust crate's directory name will
@@ -23,12 +26,8 @@ BUILD_DIR :=
 # more than one binary is supported here.
 BINARIES := $(notdir $(shell pwd))
 
-# For dependencies that are still using GNU-style convention, we would need
-# to setup certain environment variables, so they can be built via clang
-LLVM_SUFFIX := -16
-CLANG := clang$(LLVM_SUFFIX)
-
-TARGET_CC := $(CLANG)
+# Some older crates might not be parpared to be built against clang, we would
+# need to override CFLAGS to prepare them.
 TARGET_CFLAGS := --target=riscv64 -march=rv64imc_zba_zbb_zbc_zbs \
 	-nostdinc -nostdlib \
 	-I $(TOP)deps/ckb-c-stdlib/libc -DCKB_DECLARATION_ONLY
@@ -40,8 +39,8 @@ endif
 default: build test
 
 build:
-	RUSTFLAGS="$(FULL_RUSTFLAGS)" \
-		TARGET_CC="$(TARGET_CC)" TARGET_CFLAGS="$(TARGET_CFLAGS)" \
+	RUSTFLAGS="$(FULL_RUSTFLAGS)" TARGET_CC="$(CLANG)" \
+		TARGET_CFLAGS="$(TARGET_CFLAGS)" \
 		cargo build --target=riscv64imac-unknown-none-elf $(MODE_ARGS) $(CARGO_ARGS)
 	@set -eu; \
 	if [ "x$(BUILD_DIR)" != "x" ]; then \

contracts/loads-of-hashes/Cargo.toml

@@ -5,7 +5,7 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"
 big-cell-fetcher = { path = "../../crates/big-cell-fetcher" }
 big-witness-hasher = { path = "../../crates/big-witness-hasher" }
 blake2b-ref = "0.3.1"

contracts/loads-of-hashes/Makefile

@@ -15,6 +15,9 @@ FULL_RUSTFLAGS := -C target-feature=+zba,+zbb,+zbc,+zbs $(CUSTOM_RUSTFLAGS)
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set to some value, the generated binaries will be copied over
 BUILD_DIR :=
 # Generated binaries to copy. By convention, a Rust crate's directory name will
@@ -30,7 +33,7 @@ endif
 default: build test
 
 build:
-	RUSTFLAGS="$(FULL_RUSTFLAGS)" \
+	RUSTFLAGS="$(FULL_RUSTFLAGS)" TARGET_CC="$(CLANG)" \
 		cargo build --target=riscv64imac-unknown-none-elf $(MODE_ARGS) $(CARGO_ARGS)
 	@set -eu; \
 	if [ "x$(BUILD_DIR)" != "x" ]; then \

contracts/minimal-log/Cargo.toml

@@ -5,4 +5,4 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"

contracts/minimal-log/Makefile

@@ -15,6 +15,9 @@ FULL_RUSTFLAGS := -C target-feature=+zba,+zbb,+zbc,+zbs $(CUSTOM_RUSTFLAGS)
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set to some value, the generated binaries will be copied over
 BUILD_DIR :=
 # Generated binaries to copy. By convention, a Rust crate's directory name will
@@ -30,7 +33,7 @@ endif
 default: build test
 
 build:
-	RUSTFLAGS="$(FULL_RUSTFLAGS)" \
+	RUSTFLAGS="$(FULL_RUSTFLAGS)" TARGET_CC="$(CLANG)" \
 		cargo build --target=riscv64imac-unknown-none-elf $(MODE_ARGS) $(CARGO_ARGS)
 	@set -eu; \
 	if [ "x$(BUILD_DIR)" != "x" ]; then \

contracts/stack-reorder/Cargo.toml

@@ -5,7 +5,7 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"
 
 [build-dependencies]
 cc = "1.0"

contracts/stack-reorder/Makefile

@@ -16,6 +16,9 @@ FULL_RUSTFLAGS := -C target-feature=+zba,+zbb,+zbc,+zbs $(CUSTOM_RUSTFLAGS) \
 # stdout in unit tests
 CARGO_ARGS :=
 MODE := release
+# Tweak this to change the clang version to use for building C code. By default
+# we use a bash script with somes heuristics to find clang in current system.
+CLANG := $(shell $(TOP)/scripts/find_clang)
 # When this is set to some value, the generated binaries will be copied over
 BUILD_DIR :=
 # Generated binaries to copy. By convention, a Rust crate's directory name will
@@ -31,7 +34,7 @@ endif
 default: build test
 
 build:
-	RUSTFLAGS="$(FULL_RUSTFLAGS)" \
+	RUSTFLAGS="$(FULL_RUSTFLAGS)" TARGET_CC="$(CLANG)" \
 		cargo build --target=riscv64imac-unknown-none-elf $(MODE_ARGS) $(CARGO_ARGS)
 	@set -eu; \
 	if [ "x$(BUILD_DIR)" != "x" ]; then \

contracts/stack-reorder/build.rs

@@ -2,17 +2,16 @@ fn main() {
     println!("cargo:rerun-if-changed=bootloader.S");
     println!("cargo:rerun-if-changed=ld_interface.ld");
 
-    let clang = match std::env::var_os("CLANG") {
-        Some(val) => val,
-        None => "clang-16".into(),
-    };
-
     let target_arch = std::env::var("CARGO_CFG_TARGET_ARCH").unwrap();
     if target_arch == "riscv64" {
-        cc::Build::new()
+        let mut build = cc::Build::new();
+        assert!(
+            build.get_compiler().is_like_clang(),
+            "Clang must be used as the compiler!"
+        );
+        build
             .file("bootloader.S")
             .static_flag(true)
-            .compiler(clang)
             .no_default_flags(true)
             .flag("--target=riscv64")
             .flag("-march=rv64imc_zba_zbb_zbc_zbs")

crates/big-cell-fetcher/Cargo.toml

@@ -5,7 +5,7 @@ version = "0.1.0"
 edition = "2021"
 
 [dependencies]
-ckb-std = { version = "0.15.0", features = [ "build-with-clang" ] }
+ckb-std = "0.15.0"
 
 # Supporting native tests powered by ckb-x64-simulator
 [target.'cfg(all(target_arch = "x86_64", unix))'.dependencies.ckb-std]

crates/big-witness-hasher/build.rs

@@ -1,19 +1,18 @@
 fn main() {
     println!("cargo:rerun-if-changed=c.c");
 
-    let clang = match std::env::var_os("CLANG") {
-        Some(val) => val,
-        None => "clang-16".into(),
-    };
-
     let target_arch = std::env::var("CARGO_CFG_TARGET_ARCH").unwrap();
     if target_arch == "riscv64" {
-        cc::Build::new()
+        let mut build = cc::Build::new();
+        assert!(
+            build.get_compiler().is_like_clang(),
+            "Clang must be used as the compiler!"
+        );
+        build
             .file("c.c")
             .static_flag(true)
             .include("../../deps/ckb-c-stdlib")
             .include("../../deps/ckb-c-stdlib/libc")
-            .compiler(clang)
             .no_default_flags(true)
             .flag("--target=riscv64")
             .flag("-march=rv64imc_zba_zbb_zbc_zbs")

scripts/find_clang

@@ -0,0 +1,31 @@
+#!/usr/bin/env bash
+#
+# An utility script used to find a binary of clang 16+
+
+if [[ -n "${CLANG}" ]]; then
+  echo "${CLANG}"
+  exit 0
+fi
+
+CANDIDATES=("clang" "clang-16" "clang-17")
+
+BREW_PREFIX=$(brew --prefix 2> /dev/null)
+if [[ -n "${BREW_PREFIX}" ]]; then
+  CANDIDATES+=(
+    "${BREW_PREFIX}/opt/llvm/bin/clang"
+    "${BREW_PREFIX}/opt/llvm@16/bin/clang"
+    "${BREW_PREFIX}/opt/llvm@17/bin/clang"
+  )
+fi
+
+for candidate in ${CANDIDATES[@]}; do
+  OUTPUT=$($candidate -dumpversion 2> /dev/null | cut -d'.' -f 1)
+
+  if [[ $((OUTPUT)) -ge 16 ]]; then
+    echo "$candidate"
+    exit 0
+  fi
+done
+
+>&2 echo "Cannot find clang of version 16+!"
+exit 1