Try to restore some build system conventions in a low-impact way

.github/workflows/main.yml

@@ -13,44 +13,45 @@ jobs:
       with:
         submodules: true
 
-    - name: Install clang (Windows)
+    - name: Install LLVM tools (Windows)
       shell: bash
       run: |
-          curl -fsSLO https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0/LLVM-10.0.0-win64.exe
-          7z x LLVM-10.0.0-win64.exe -y -o"llvm"
-          echo "$(pwd)/llvm/bin" >> $GITHUB_PATH
-          echo "WASM_AR=$(pwd)/llvm/bin/llvm-ar.exe" >> $GITHUB_ENV
+        curl -fsSLO https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0/LLVM-10.0.0-win64.exe
+        7z x LLVM-10.0.0-win64.exe -y -o"llvm"
+        echo "$(pwd)/llvm/bin" >> $GITHUB_PATH
+        echo "CC=$(pwd)/llvm/bin/clang.exe" >> $GITHUB_ENV
+        echo "AR=$(pwd)/llvm/bin/llvm-ar.exe" >> $GITHUB_ENV
+        echo "NM=$(pwd)/llvm/bin/llvm-nm.exe" >> $GITHUB_ENV
       if: matrix.os == 'windows-latest'
 
-    - name: Install llvm-nm (Windows)
+    - name: Override llvm-nm with one from rustup (Windows)
       run: |
         rustup update stable
         rustup default stable
         rustup component add llvm-tools-preview
+        echo "NM=$(rustc --print sysroot|sed 's|C:|/c|'|sed 's|\\|/|g')/lib/rustlib/x86_64-pc-windows-msvc/bin/llvm-nm.exe" >> $GITHUB_ENV
       if: matrix.os == 'windows-latest'
 
-    - name: Register llvm-nm in environment (Windows)
-      shell: bash
-      run: |
-        echo "WASM_NM=$(rustc --print sysroot|sed 's|C:|/c|'|sed 's|\\|/|g')/lib/rustlib/x86_64-pc-windows-msvc/bin/llvm-nm.exe" >> $GITHUB_ENV
-      if: matrix.os == 'windows-latest'
-
-    - name: Install clang (MacOS)
+    - name: Install LLVM tools (MacOS)
       shell: bash
       run: |
         curl -sSfL https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0/clang+llvm-10.0.0-x86_64-apple-darwin.tar.xz | tar xJf -
         export CLANG_DIR=`pwd`/clang+llvm-10.0.0-x86_64-apple-darwin/bin
         echo "$CLANG_DIR" >> $GITHUB_PATH
-        echo "WASM_CC=$CLANG_DIR/clang" >> $GITHUB_ENV
+        echo "CC=$CLANG_DIR/clang" >> $GITHUB_ENV
+        echo "AR=$CLANG_DIR/llvm-ar" >> $GITHUB_ENV
+        echo "NM=$CLANG_DIR/llvm-nm" >> $GITHUB_ENV
       if: matrix.os == 'macos-latest'
 
-    - name: Install clang (Linux)
+    - name: Install LLVM tools (Linux)
       shell: bash
       run: |
         curl -sSfL https://github.com/llvm/llvm-project/releases/download/llvmorg-10.0.0/clang+llvm-10.0.0-x86_64-linux-gnu-ubuntu-18.04.tar.xz | tar xJf -
         export CLANG_DIR=`pwd`/clang+llvm-10.0.0-x86_64-linux-gnu-ubuntu-18.04/bin
         echo "$CLANG_DIR" >> $GITHUB_PATH
-        echo "WASM_CC=$CLANG_DIR/clang" >> $GITHUB_ENV
+        echo "CC=$CLANG_DIR/clang" >> $GITHUB_ENV
+        echo "AR=$CLANG_DIR/llvm-ar" >> $GITHUB_ENV
+        echo "NM=$CLANG_DIR/llvm-nm" >> $GITHUB_ENV
       if: matrix.os == 'ubuntu-latest'
 
     - name: Build libc

Makefile

@@ -1,11 +1,21 @@
 # These variables are specifically meant to be overridable via the make
 # command-line.
-WASM_CC ?= clang
-WASM_NM ?= $(patsubst %clang,%llvm-nm,$(filter-out ccache sccache,$(WASM_CC)))
-WASM_AR ?= $(patsubst %clang,%llvm-ar,$(filter-out ccache sccache,$(WASM_CC)))
-WASM_CFLAGS ?= -O2 -DNDEBUG
-# The directory where we build the sysroot.
+CC ?= clang
+NM ?= $(patsubst %clang,%llvm-nm,$(filter-out ccache sccache,$(CC)))
+AR ?= $(patsubst %clang,%llvm-ar,$(filter-out ccache sccache,$(CC)))
+EXTRA_CFLAGS ?= -O2 -DNDEBUG
+# These variables describe the locations of various files and
+# directories to install.
+#
+# `SYSROOT` is just used to define the defaults, and in the install
+# rule. That means you should not override `SYSROOT_*` if you are going
+# to run `make install`, but if you don't need to, you are free to move
+# the `SYSROOT_*` directories around freely, not preserving their
+# default locations relative `SYSROOT`.
 SYSROOT ?= $(CURDIR)/sysroot
+SYSROOT_LIB ?= $(SYSROOT)/lib/$(MULTIARCH_TRIPLE)
+SYSROOT_INC ?= $(SYSROOT)/include
+SYSROOT_SHARE ?= $(SYSROOT)/share/$(MULTIARCH_TRIPLE)
 # A directory to install to for "make install".
 INSTALL_DIR ?= /usr/local
 # single or posix
@@ -184,8 +194,10 @@ LIBC_TOP_HALF_ALL_SOURCES = \
     $(LIBC_TOP_HALF_MUSL_SOURCES) \
     $(shell find $(LIBC_TOP_HALF_SOURCES) -name \*.c)
 
+# Add any extra flags
+CFLAGS = $(EXTRA_CFLAGS)
 # Set the target.
-CFLAGS = $(WASM_CFLAGS) --target=$(TARGET_TRIPLE)
+CFLAGS += --target=$(TARGET_TRIPLE)
 # WebAssembly floating-point match doesn't trap.
 # TODO: Add -fno-signaling-nans when the compiler supports it.
 CFLAGS += -fno-trapping-math
@@ -252,12 +264,6 @@ LIBWASI_EMULATED_GETPID_OBJS = $(call objs,$(LIBWASI_EMULATED_GETPID_SOURCES))
 LIBWASI_EMULATED_SIGNAL_OBJS = $(call objs,$(LIBWASI_EMULATED_SIGNAL_SOURCES))
 LIBWASI_EMULATED_SIGNAL_MUSL_OBJS = $(call objs,$(LIBWASI_EMULATED_SIGNAL_MUSL_SOURCES))
 
-# These variables describe the locations of various files and
-# directories in the generated sysroot tree.
-SYSROOT_LIB := $(SYSROOT)/lib/$(MULTIARCH_TRIPLE)
-SYSROOT_INC = $(SYSROOT)/include
-SYSROOT_SHARE = $(SYSROOT)/share/$(MULTIARCH_TRIPLE)
-
 # Files from musl's include directory that we don't want to install in the
 # sysroot's include directory.
 MUSL_OMIT_HEADERS :=
@@ -361,13 +367,13 @@ $(SYSROOT_LIB)/libwasi-emulated-signal.a: $(LIBWASI_EMULATED_SIGNAL_OBJS) $(LIBW
 %.a:
 	@mkdir -p "$(@D)"
 	# On Windows, the commandline for the ar invocation got too long, so it needs to be split up.
-	$(WASM_AR) crs $@ $(wordlist 1, 199, $^)
-	$(WASM_AR) crs $@ $(wordlist 200, 399, $^)
-	$(WASM_AR) crs $@ $(wordlist 400, 599, $^)
-	$(WASM_AR) crs $@ $(wordlist 600, 799, $^)
+	$(AR) crs $@ $(wordlist 1, 199, $^)
+	$(AR) crs $@ $(wordlist 200, 399, $^)
+	$(AR) crs $@ $(wordlist 400, 599, $^)
+	$(AR) crs $@ $(wordlist 600, 799, $^)
 	# This might eventually overflow again, but at least it'll do so in a loud way instead of
 	# silently dropping the tail.
-	$(WASM_AR) crs $@ $(wordlist 800, 100000, $^)
+	$(AR) crs $@ $(wordlist 800, 100000, $^)
 
 $(MUSL_PRINTSCAN_OBJS): CFLAGS += \
 	    -D__wasilibc_printscan_no_long_double \
@@ -382,15 +388,15 @@ $(LIBWASI_EMULATED_SIGNAL_MUSL_OBJS): CFLAGS += \
 
 $(OBJDIR)/%.long-double.o: $(CURDIR)/%.c include_dirs
 	@mkdir -p "$(@D)"
-	$(WASM_CC) $(CFLAGS) -MD -MP -o $@ -c $<
+	$(CC) $(CFLAGS) -MD -MP -o $@ -c $<
 
 $(OBJDIR)/%.no-floating-point.o: $(CURDIR)/%.c include_dirs
 	@mkdir -p "$(@D)"
-	$(WASM_CC) $(CFLAGS) -MD -MP -o $@ -c $<
+	$(CC) $(CFLAGS) -MD -MP -o $@ -c $<
 
 $(OBJDIR)/%.o: $(CURDIR)/%.c include_dirs
 	@mkdir -p "$(@D)"
-	$(WASM_CC) $(CFLAGS) -MD -MP -o $@ -c $<
+	$(CC) $(CFLAGS) -MD -MP -o $@ -c $<
 
 -include $(shell find $(OBJDIR) -name \*.d)
 
@@ -420,7 +426,7 @@ $(LIBWASI_EMULATED_PROCESS_CLOCKS_OBJS): CFLAGS += \
     -I$(LIBC_BOTTOM_HALF_CLOUDLIBC_SRC)
 
 include_dirs:
-	$(RM) -r "$(SYSROOT)"
+	$(RM) -r "$(SYSROOT_INC)"
 
 	#
 	# Install the include files.
@@ -450,7 +456,7 @@ startup_files: include_dirs
 	#
 	@mkdir -p "$(OBJDIR)"
 	cd "$(OBJDIR)" && \
-	$(WASM_CC) $(CFLAGS) -c $(LIBC_BOTTOM_HALF_CRT_SOURCES) -MD -MP && \
+	$(CC) $(CFLAGS) -c $(LIBC_BOTTOM_HALF_CRT_SOURCES) -MD -MP && \
 	mkdir -p "$(SYSROOT_LIB)" && \
 	mv *.o "$(SYSROOT_LIB)"
 
@@ -468,11 +474,15 @@ finish: startup_files libc
 	# Create empty placeholder libraries.
 	#
 	for name in m rt pthread crypt util xnet resolv dl; do \
-	    $(WASM_AR) crs "$(SYSROOT_LIB)/lib$${name}.a"; \
+	    $(AR) crs "$(SYSROOT_LIB)/lib$${name}.a"; \
 	done
 
 	#
-	# The build succeeded! The generated sysroot is in $(SYSROOT).
+	# The build succeeded! The generated files are in
+	#
+	# lib:     $(SYSROOT_LIB)
+	# include: $(SYSROOT_INC)
+	# share:   $(SYSROOT_SHARE)
 	#
 
 # The check for defined and undefined symbols expects there to be a heap
@@ -495,9 +505,9 @@ check-symbols: startup_files libc
 	@# LLVM PR40497, which is fixed in 9.0, but not in 8.0.
 	@# Ignore certain llvm builtin symbols such as those starting with __mul
 	@# since these dependencies can vary between llvm versions.
-	"$(WASM_NM)" --defined-only "$(SYSROOT_LIB)"/libc.a "$(SYSROOT_LIB)"/libwasi-emulated-*.a "$(SYSROOT_LIB)"/*.o \
+	"$(NM)" --defined-only "$(SYSROOT_LIB)"/libc.a "$(SYSROOT_LIB)"/libwasi-emulated-*.a "$(SYSROOT_LIB)"/*.o \
 	    |grep ' [[:upper:]] ' |sed 's/.* [[:upper:]] //' |LC_ALL=C sort > "$(SYSROOT_SHARE)/defined-symbols.txt"
-	for undef_sym in $$("$(WASM_NM)" --undefined-only "$(SYSROOT_LIB)"/libc.a "$(SYSROOT_LIB)"/libc-*.a "$(SYSROOT_LIB)"/*.o \
+	for undef_sym in $$("$(NM)" --undefined-only "$(SYSROOT_LIB)"/libc.a "$(SYSROOT_LIB)"/libc-*.a "$(SYSROOT_LIB)"/*.o \
 	    |grep ' U ' |sed 's/.* U //' |LC_ALL=C sort |uniq); do \
 	    grep -q '\<'$$undef_sym'\>' "$(SYSROOT_SHARE)/defined-symbols.txt" || echo $$undef_sym; \
 	done | grep -v "^__mul" > "$(SYSROOT_SHARE)/undefined-symbols.txt"
@@ -507,16 +517,16 @@ check-symbols: startup_files libc
 	#
 	# Generate a test file that includes all public header files.
 	#
-	cd "$(SYSROOT)" && \
-	  for header in $$(find include -type f -not -name mman.h -not -name signal.h -not -name times.h -not -name resource.h |grep -v /bits/); do \
-	      echo '#include <'$$header'>' | sed 's/include\///' ; \
-	done |LC_ALL=C sort >share/$(MULTIARCH_TRIPLE)/include-all.c ; \
+	cd "$(SYSROOT_INC)" && \
+	  for header in $$(find . -type f -not -name mman.h -not -name signal.h -not -name times.h -not -name resource.h |grep -v /bits/); do \
+	      echo '#include <'$$header'>' | sed 's/\.\///' ; \
+	done |LC_ALL=C sort >$(SYSROOT_SHARE)/include-all.c ; \
 	cd - >/dev/null
 
 	#
 	# Test that it compiles.
 	#
-	$(WASM_CC) $(CFLAGS) -fsyntax-only "$(SYSROOT_SHARE)/include-all.c" -Wno-\#warnings
+	$(CC) $(CFLAGS) -fsyntax-only "$(SYSROOT_SHARE)/include-all.c" -Wno-\#warnings
 
 	#
 	# Collect all the predefined macros, except for compiler version macros
@@ -532,7 +542,7 @@ check-symbols: startup_files libc
 	@#
 	@# TODO: Undefine __FLOAT128__ for now since it's not in clang 8.0.
 	@# TODO: Filter out __FLT16_* for now, as not all versions of clang have these.
-	$(WASM_CC) $(CFLAGS) "$(SYSROOT_SHARE)/include-all.c" \
+	$(CC) $(CFLAGS) "$(SYSROOT_SHARE)/include-all.c" \
 	    -isystem $(SYSROOT_INC) \
 	    -std=gnu17 \
 	    -E -dM -Wno-\#warnings \
@@ -558,8 +568,25 @@ check-symbols: startup_files libc
 	# This ignores whitespace because on Windows the output has CRLF line endings.
 	diff -wur "$(CURDIR)/expected/$(MULTIARCH_TRIPLE)" "$(SYSROOT_SHARE)"
 
+NORMAL_SYSROOT := \
+	$(findstring $(SYSROOT)/lib,$(SYSROOT_LIB)) \
+	$(findstring $(SYSROOT)/include,$(SYSROOT_INC)) \
+	$(findstring $(SYSROOT)/share,$(SYSROOT_SHARE))
+
+# If the 3 dirs are not within the standard locations, the install rule
+# is undefined. This protects the user from doing something silly by
+# mistake.
+ifeq ($(SYSROOT)/lib $(SYSROOT)/include $(SYSROOT)/share,$(strip $(NORMAL_SYSROOT)))
 install: finish
 	mkdir -p "$(INSTALL_DIR)"
 	cp -r "$(SYSROOT)/lib" "$(SYSROOT)/share" "$(SYSROOT)/include" "$(INSTALL_DIR)"
+.PHONY: install
+endif
+
+clean:
+	$(RM) -r "$(OBJDIR)"
+	$(RM) -r "$(SYSROOT_INC)"
+	$(RM) -r "$(SYSROOT_LIB)"
+	$(RM) -r "$(SYSROOT_SHARE)"
 
-.PHONY: default startup_files libc finish install include_dirs
+.PHONY: default startup_files libc finish include_dirs clean

README.md

@@ -21,9 +21,9 @@ To build a WASI sysroot from source, obtain a WebAssembly-supporting C compiler
 and then run:
 
 ```sh
-make WASM_CC=/path/to/clang/with/wasm/support \
-     WASM_AR=/path/to/llvm-ar \
-     WASM_NM=/path/to/llvm-nm
+make CC=/path/to/clang/with/wasm/support \
+     AR=/path/to/llvm-ar \
+     NM=/path/to/llvm-nm
 ```
 
 This makes a directory called "sysroot", by default. See the top of the Makefile