Full Abseil + RE2 support (#173)

Add RE2 support + full Abseil Related cleanups: - Simpler Docker container - Single Makefile - Support for -flto - SDK proto upgrade to v26.1 - Updated build docs - More/independent examples - Remove protobuf CI check Signed-off-by: Martijn Stevenson <[email protected]>
proxy-wasm · Oct 25, 2024 · 9dc95ae · 9dc95ae
1 parent 6b3dc93
commit 9dc95ae
Show file tree

Hide file tree

Showing 20 changed files with 17,981 additions and 17,848 deletions.
diff --git a/.github/workflows/cpp.yml b/.github/workflows/cpp.yml
@@ -60,29 +60,6 @@ jobs:
         export PATH=$PATH:$(go env GOPATH)/bin
         addlicense -ignore="**/BUILD" -check .
 
-  protobuf:
-    runs-on: ubuntu-22.04
-
-    steps:
-    - uses: actions/checkout@v2
-
-    - name: Install protobuf v3.9.1
-      run: |
-        git clone https://github.com/protocolbuffers/protobuf
-        cd protobuf
-        git checkout v3.9.1
-        ./autogen.sh
-        ./configure
-        make
-        sudo make install
-        sudo ldconfig
-
-    - name: Re-generate and verify protobuf artifacts
-      run: |
-        rm *.pb.{cc,h}
-        make protobuf
-        git diff --exit-code -G "(^[^ /])|(^\s+[^\*])" *.pb.{cc,h}
-
   build:
     runs-on: ubuntu-22.04
 

diff --git a/Dockerfile-sdk b/Dockerfile-sdk
@@ -1,7 +1,6 @@
 FROM ubuntu:noble
 
-COPY ./sdk_container.sh /
-COPY ./build_wasm.sh /
+COPY *.sh /
 COPY *.cc *.h *.js *.proto Makefile* /sdk/
 
 RUN ./sdk_container.sh
diff --git a/Makefile b/Makefile
@@ -1,27 +1,57 @@
-CPP_API ?= .
+ifdef NO_CONTEXT
+  CPP_CONTEXT_LIB =
+else
+  CPP_CONTEXT_LIB = ${PROXY_WASM_CPP_SDK}/proxy_wasm_intrinsics.cc
+endif
 
-all: proxy_wasm_intrinsics.pb.h proxy_wasm_intrinsics_lite.pb.h struct_lite.pb.h ${CPP_API}/libprotobuf.a ${CPP_API}/libprotobuf-lite.a
+PROTOBUF ?= none
+ifeq ($(PROTOBUF), full)
+  PROTO_DEPS := protobuf
+  PROTO_OPTS := -DPROXY_WASM_PROTOBUF_FULL=1 \
+		${PROXY_WASM_CPP_SDK}/proxy_wasm_intrinsics.pb.cc
+else ifeq ($(PROTOBUF), lite)
+  PROTO_DEPS := protobuf-lite
+  PROTO_OPTS := -DPROXY_WASM_PROTOBUF_LITE=1 \
+		${PROXY_WASM_CPP_SDK}/proxy_wasm_intrinsics_lite.pb.cc \
+		${PROXY_WASM_CPP_SDK}/struct_lite.pb.cc
+else
+  PROTO_DEPS :=
+  PROTO_OPTS :=
+endif
 
-protobuf: proxy_wasm_intrinsics.pb.h proxy_wasm_intrinsics_lite.pb.h struct_lite.pb.h
+# Provide a list of libraries that the wasm depends on (absl_*, re2, etc).
+WASM_DEPS ?=
 
-proxy_wasm_intrinsics.pb.h: proxy_wasm_intrinsics.proto
-	protoc --cpp_out=. proxy_wasm_intrinsics.proto
+# Determine dependency link options.
+# NOTE: Strip out -pthread which RE2 claims to need...
+PKG_CONFIG ?= pkg-config
+PKG_CONFIG_PATH = ${EMSDK}/upstream/emscripten/cache/sysroot/lib/pkgconfig
+WASM_LIBS = $(shell $(PKG_CONFIG) $(WASM_DEPS) $(PROTO_DEPS) \
+	  --with-path=$(PKG_CONFIG_PATH) --libs | sed -e 's/-pthread //g')
 
-proxy_wasm_intrinsics_lite.pb.h struct_lite.pb.h: proxy_wasm_intrinsics_lite.proto
-	protoc --cpp_out=. -I. proxy_wasm_intrinsics_lite.proto
-	protoc --cpp_out=. struct_lite.proto
+debug-deps:
+	# WASM_DEPS : ${WASM_DEPS}
+	# WASM_LIBS : ${WASM_LIBS}
+	# PROTO_DEPS: ${PROTO_DEPS}
+	# PROTO_OPTS: ${PROTO_OPTS}
 
-${CPP_API}/libprotobuf.a ${CPP_API}/libprotobuf-lite.a:
-	rm -rf protobuf-wasm \
-	&& git clone https://github.com/protocolbuffers/protobuf protobuf-wasm \
-	&& cd protobuf-wasm \
-	&& git checkout v3.9.1 \
-	&& ./autogen.sh \
-	&& emconfigure ./configure --disable-shared CXXFLAGS="-O3 -flto" \
-	&& emmake make \
-	&& cd .. \
-	&& cp protobuf-wasm/src/.libs/libprotobuf-lite.a ${CPP_API}/libprotobuf-lite.a \
-	&& cp protobuf-wasm/src/.libs/libprotobuf.a ${CPP_API}/libprotobuf.a
+# TODO(mpwarres): Add Emscripten stack/heap size params in PR#174.
+%.wasm %.wat: %.cc
+	em++ --no-entry -sSTANDALONE_WASM -sEXPORTED_FUNCTIONS=_malloc \
+		--std=c++17 -O3 -flto \
+		--js-library ${PROXY_WASM_CPP_SDK}/proxy_wasm_intrinsics.js \
+		-I${PROXY_WASM_CPP_SDK} \
+		${CPP_CONTEXT_LIB} \
+		${PROTO_OPTS} \
+		${WASM_LIBS} \
+		$*.cc -o $*.wasm
 
 clean:
-	rm -f proxy_wasm_intrinsics.pb.h proxy_wasm_intrinsics_lite.pb.h struct_lite.pb.h ${CPP_API}/libprotobuf.a ${CPP_API}/libprotobuf-lite.a
+	rm *.wasm
+
+# NOTE: How to regenerate .pb.h and .pb.cc files for a protobuf update:
+# - download + extract protobuf release (currently v26.1)
+# - regenerate:
+#   ./bin/protoc --cpp_out=../ -I../ -Iinclude/google/protobuf/ ../struct_lite.proto
+#   ./bin/protoc --cpp_out=../ -I../ -Iinclude/google/protobuf/ ../proxy_wasm_intrinsics_lite.proto
+#   ./bin/protoc --cpp_out=../ -I../ -Iinclude/google/protobuf/ ../proxy_wasm_intrinsics.proto
diff --git a/Makefile.base b/Makefile.base
diff --git a/Makefile.base_lite b/Makefile.base_lite
diff --git a/bazel/defs.bzl b/bazel/defs.bzl
@@ -16,10 +16,11 @@ load("@emsdk//emscripten_toolchain:wasm_rules.bzl", "wasm_cc_binary")
 load("@rules_cc//cc:defs.bzl", "cc_binary")
 
 def _optimized_wasm_cc_binary_transition_impl(settings, attr):
-    # TODO(PiotrSikora): Add -flto to copts/linkopts when fixed in emsdk.
-    # See: https://github.com/emscripten-core/emsdk/issues/971
+    # Define STANDALONE_WASM at compile time as well as link time (below).
+    # This fixes Abseil by not including Emscripten JS stacktraces + symbolization.
+    # TODO(martijneken): Remove after Abseil stops using this define.
     return {
-        "//command_line_option:copt": ["-O3"],
+        "//command_line_option:copt": ["-O3", "-flto", "-DSTANDALONE_WASM"],
         "//command_line_option:cxxopt": [],
         "//command_line_option:linkopt": [],
         "//command_line_option:collect_code_coverage": False,
@@ -102,6 +103,8 @@ def proxy_wasm_cc_binary(
     )
 
     wasm_cc_binary(
+        standalone = True,
+        threads = "off",
         name = "wasm_" + name,
         cc_target = ":proxy_wasm_" + name.rstrip(".wasm"),
         tags = tags + [

diff --git a/bazel/repositories.bzl b/bazel/repositories.bzl
@@ -31,3 +31,19 @@ def proxy_wasm_cpp_sdk_repositories():
         strip_prefix = "protobuf-3.17.3",
         url = "https://github.com/protocolbuffers/protobuf/releases/download/v3.17.3/protobuf-all-3.17.3.tar.gz",
     )
+
+    maybe(
+        http_archive,
+        name = "com_google_absl",
+        sha256 = "95e90be7c3643e658670e0dd3c1b27092349c34b632c6e795686355f67eca89f",
+        strip_prefix = "abseil-cpp-20240722.0",
+        urls = ["https://github.com/abseil/abseil-cpp/archive/20240722.0.zip"],
+    )
+
+    maybe(
+        http_archive,
+        name = "com_google_re2",
+        sha256 = "18cf85922e27fad3ed9c96a27733037da445f35eb1a2744c306a37c6d11e95c4",
+        strip_prefix = "re2-2023-07-01",
+        url = "https://github.com/google/re2/archive/2023-07-01.tar.gz",
+    )
diff --git a/build_wasm.sh b/build_wasm.sh
@@ -16,4 +16,4 @@
 
 source /root/emsdk/emsdk_env.sh
 export PATH=/usr/local/bin:$PATH
-make
+make "$@"
diff --git a/docs/building.md b/docs/building.md
@@ -1,8 +1,8 @@
 # Proxy-Wasm C++ SDK Build Instructions
 
 The C++ SDK has dependencies on specific versions of the C++ WebAssembly
-toolchain [Emscripten](https://emscripten.org) and the protobuf library,
-therefore use of a Docker image is recommended.
+toolchain [Emscripten](https://emscripten.org). Use of a Docker image is
+recommended to achieve repeatable builds and save work.
 
 ## Docker
 
@@ -11,7 +11,7 @@ A Dockerfile for the C++ SDK is provided in [Dockerfile-sdk](../Dockerfile-sdk).
 It can built in this repository's root directory by:
 
 ```bash
-docker build -t wasmsdk:v2 -f Dockerfile-sdk .
+docker build -t wasmsdk:v3 -f Dockerfile-sdk .
 ```
 
 The docker image can be used for compiling C++ plugin code into Wasm modules.
@@ -23,12 +23,10 @@ Create a directory with your source files and a Makefile:
 ```makefile
 PROXY_WASM_CPP_SDK=/sdk
 
-all: myproject.wasm
-
-include ${PROXY_WASM_CPP_SDK}/Makefile.base_lite
+include ${PROXY_WASM_CPP_SDK}/Makefile
 ```
 
-Source file (myproject.cc):
+Create a C++ source file (myproject.cc):
 
 ```c++
 #include <string>
@@ -57,62 +55,40 @@ void ExampleContext::onDone() { logInfo("onDone " + std::to_string(id())); }
 
 ### Compiling with the Docker build image
 
-Run docker:
-
-```bash
-docker run -v $PWD:/work -w /work  wasmsdk:v2 /build_wasm.sh
-```
-
-### Caching the standard libraries
-
-The first time that emscripten runs it will generate the standard libraries.  To
-cache these in the docker image, after the first successful compilation (e.g
-myproject.cc above), commit the image with the standard libraries:
+Run docker to build wasm, using a target with a .wasm suffix:
 
 ```bash
-docker commit `docker ps -l | grep wasmsdk:v2 | awk '{print $1}'` wasmsdk:v2
+docker run -v $PWD:/work -w /work wasmsdk:v3 /build_wasm.sh myproject.wasm
 ```
 
-This will save time on subsequent compiles.
+You can specify wasm dependencies via these Makefile variables:
 
-### Using Abseil from the Docker image
+-   PROTOBUF = {full, lite, none}
+-   WASM_DEPS = list of libraries
 
-Abseil (optionally) is built in /root/abseil and can be used. Note that the
-Abseil containers (e.g. `absl::flat_hash_set`) exercise many syscalls which are
-not supported. Consequentially individual files should be pulled in which are
-relatively self contained (e.g. `strings`). Example customized Makefile:
+For example:
 
 ```makefile
 PROXY_WASM_CPP_SDK=/sdk
-CPP_API:=${PROXY_WASM_CPP_SDK}
-CPP_CONTEXT_LIB = ${CPP_API}/proxy_wasm_intrinsics.cc
-ABSL = /root/abseil-cpp
-ABSL_CPP = ${ABSL}/absl/strings/str_cat.cc ${ABSL}/absl/strings/str_split.cc ${ABSL}/absl/strings/numbers.cc ${ABSL}/absl/strings/ascii.cc
 
-all: plugin.wasm
+PROTOBUF=lite
+WASM_DEPS=re2 absl_strings
 
-%.wasm %.wat: %.cc ${CPP_API}/proxy_wasm_intrinsics.h ${CPP_API}/proxy_wasm_enums.h ${CPP_API}/proxy_wasm_externs.h ${CPP_API}/proxy_wasm_api.h ${CPP_API}/proxy_wasm_intrinsics.js ${CPP_CONTEXT_LIB}
-        ls /root
-                em++ --no-entry -s EXPORTED_FUNCTIONS=['_malloc'] --std=c++17 -O3 -flto -I${CPP_API} -I${CPP_API}/google/protobuf -I/usr/local/include -I${ABSL} --js-library ${CPP_API}/proxy_wasm_intrinsics.js ${ABSL_CPP} $*.cc ${CPP_API}/proxy_wasm_intrinsics.pb.cc ${CPP_CONTEXT_LIB} ${CPP_API}/libprotobuf.a -o $*.wasm
+include ${PROXY_WASM_CPP_SDK}/Makefile
 ```
 
-Precompiled Abseil libraries are also available, so the above can also be done
-as:
-
-```makefile
-PROXY_WASM_CPP_SDK=/sdk
-CPP_API:=${PROXY_WASM_CPP_SDK}
-CPP_CONTEXT_LIB = ${CPP_API}/proxy_wasm_intrinsics.cc
-ABSL = /root/abseil-cpp
-ABSL_LIBS = ${ABSL}/absl/strings/libabsl_strings.a ${ABSL}/absl/strings/libabsl_strings_internal.a  ${ABSL}/absl/strings/libabsl_str_format_internal.a
+### Caching the standard libraries
 
-all: plugin.wasm
+The first time that emscripten runs it will generate the standard libraries.  To
+cache these in the docker image, after the first successful compilation (e.g
+myproject.cc above), commit the image with the standard libraries:
 
-%.wasm %.wat: %.cc ${CPP_API}/proxy_wasm_intrinsics.h ${CPP_API}/proxy_wasm_enums.h ${CPP_API}/proxy_wasm_externs.h ${CPP_API}/proxy_wasm_api.h ${CPP_API}/proxy_wasm_intrinsics.js ${CPP_CONTEXT_LIB}
-        ls /root
-                em++ --no-entry -s EXPORTED_FUNCTIONS=['_malloc'] --std=c++17 -O3 -flto -I${CPP_API} -I${CPP_API}/google/protobuf -I/usr/local/include -I${ABSL} --js-library ${CPP_API}/proxy_wasm_intrinsics.js  $*.cc ${CPP_API}/proxy_wasm_intrinsics.pb.cc ${CPP_CONTEXT_LIB} ${CPP_API}/libprotobuf.a ${ABSL_LIBS} -o $*.wasm
+```bash
+docker commit `docker ps -l | grep wasmsdk:v3 | awk '{print $1}'` wasmsdk:v3
 ```
 
+This will save time on subsequent compiles.
+
 ### Ownership of the resulting .wasm files
 
 The compiled files may be owned by root.  To chown them, add the follow lines to
@@ -124,48 +100,32 @@ PROXY_WASM_CPP_SDK=/sdk
 all: myproject.wasm
   chown ${uid}.${gid} $^
 
-include ${PROXY_WASM_CPP_SDK}/Makefile.base_lite
+include ${PROXY_WASM_CPP_SDK}/Makefile
 ```
 
 Invocation file (e.g. build.sh):
 
 ```bash
 #!/bin/bash
-docker run -e uid="$(id -u)" -e gid="$(id -g)" -v $PWD:/work -w /work wasmsdk:v2 /build_wasm.sh
+docker run -e uid="$(id -u)" -e gid="$(id -g)" -v $PWD:/work -w /work wasmsdk:v3 /build_wasm.sh
 ```
 
 ## Dependencies for building Wasm modules:
 
 If you do not wish to use the Docker image, the dependencies can be installed by
-script (sdk\_container.sh), or by hand.
-
-### protobuf v3.9.1
-
-You must install the version of protobuf on your build system that matches the
-libprotobuf.a files (without any patches) so that the generated code matches the
-.a library.  Currently this is based on tag v3.9.1 of
-https://github.com/protocolbuffers/protobuf.
-
-```bash
-git clone https://github.com/protocolbuffers/protobuf
-cd protobuf
-git checkout v3.9.1
-git submodule update --init --recursive
-./autogen.sh
-./configure
-make
-make check
-sudo make install
-```
+script (sdk\_container.sh), or by hand. First you need Emscripten to build
+everything else. Then you can build other wasm-compatible libraries, such as
+protobuf, abseil, and RE2.
 
 ### Emscripten
 
+Version 3.1.67 is known to work:
+
 ```bash
 git clone https://github.com/emscripten-core/emsdk.git
 cd emsdk
 git checkout 3.1.67
-
-./emsdk install 3.1.67
+./emsdk install --shallow 3.1.67
 ./emsdk activate 3.1.67
 
 source ./emsdk_env.sh
@@ -178,26 +138,3 @@ It is possible later versions will work, e.g.
 ./emsdk install latest
 ./emsdk activate latest
 ```
-
-However 3.1.67 is known to work.
-
-### Rebuilding the libprotobuf.a files
-
-To build the protobuf static libraries for use in your proxy-wasm wasm module,
-if you need them, you may use Emscripten in the same way sdk\_container.sh does
-it:
-
-```bash
-git clone https://github.com/protocolbuffers/protobuf protobuf-wasm
-cd protobuf-wasm
-git checkout v3.9.1
-git submodule update --init --recursive
-./autogen.sh
-emconfigure ./configure --disable-shared CXXFLAGS="-O3 -flto"
-emmake make
-cd ..
-cp protobuf-wasm/src/.libs/libprotobuf-lite.a ${CPP_API}/libprotobuf-lite.a
-cp protobuf-wasm/src/.libs/libprotobuf.a ${CPP_API}/libprotobuf.a
-```
-
-Note: ensure /usr/local/bin is in your path.