Merge pull request #5595 from jamessan/tsan-pie

tsan: Compile with -fPIE and updated clang

.travis.yml

@@ -9,9 +9,7 @@ env:
     # http://docs.travis-ci.com/user/speeding-up-the-build/#Paralellizing-your-build-on-one-VM
     - MAKE_CMD="make -j2"
     # Update PATH for pip.
-    - PATH="$(python2.7 -c 'import site; print(site.getuserbase())')/bin:$PATH"
-    # LLVM symbolizer path.
-    - LLVM_SYMBOLIZER="$(which llvm-symbolizer-3.4)"
+    - PATH="$(python2.7 -c 'import site; print(site.getuserbase())')/bin:/usr/lib/llvm-symbolizer-3.8/bin:$PATH"
     # Build directory for Neovim.
     - BUILD_DIR="$TRAVIS_BUILD_DIR/build"
     # Build directory for third-party dependencies.
@@ -36,9 +34,8 @@ env:
                          -DCMAKE_TOOLCHAIN_FILE=$TRAVIS_BUILD_DIR/cmake/i386-linux-gnu.toolchain.cmake"
     # Environment variables for Clang sanitizers.
     - ASAN_OPTIONS="detect_leaks=1:check_initialization_order=1:log_path=$LOG_DIR/asan"
-    - ASAN_SYMBOLIZER_PATH="$LLVM_SYMBOLIZER"
-    - TSAN_OPTIONS="external_symbolizer_path=$LLVM_SYMBOLIZER log_path=$LOG_DIR/tsan"
-    - UBSAN_OPTIONS="log_path=$LOG_DIR/ubsan"
+    - TSAN_OPTIONS="log_path=$LOG_DIR/tsan"
+    - UBSAN_OPTIONS="print_stacktrace=1 log_path=$LOG_DIR/ubsan"
     # Environment variables for Valgrind.
     - VALGRIND_LOG="$LOG_DIR/valgrind-%p.log"
     # Cache marker for third-party dependencies cache.
@@ -70,10 +67,10 @@ matrix:
       compiler: gcc-5 -m32
       env: BUILD_32BIT=ON
     - os: linux
-      compiler: clang
+      compiler: clang-3.8
       env: CLANG_SANITIZER=ASAN_UBSAN
     - os: linux
-      compiler: clang
+      compiler: clang-3.8
       env: CLANG_SANITIZER=TSAN
     - os: osx
       compiler: clang
@@ -96,11 +93,12 @@ addons:
       # TODO: Remove PPA when Travis gets Python >=3.3.
       - deadsnakes
       - ubuntu-toolchain-r-test
+      - llvm-toolchain-precise-3.8
     packages:
       - autoconf
       - automake
       - build-essential
-      - clang-3.4
+      - clang-3.8
       - cmake
       - g++-5-multilib
       - g++-multilib
@@ -109,7 +107,7 @@ addons:
       - gdb
       - libc6-dev-i386
       - libtool
-      - llvm-3.4-dev
+      - llvm-3.8-dev
       - pkg-config
       - python3.3-dev
       - unzip

src/nvim/CMakeLists.txt

@@ -326,6 +326,7 @@ elseif(CLANG_TSAN)
   message(STATUS "Enabling Clang thread sanitizer for nvim.")
   set_property(TARGET nvim APPEND_STRING PROPERTY COMPILE_FLAGS "-DEXITFREE ")
   set_property(TARGET nvim APPEND_STRING PROPERTY COMPILE_FLAGS "-fsanitize=thread ")
+  set_property(TARGET nvim APPEND_STRING PROPERTY COMPILE_FLAGS "-fPIE ")
   set_property(TARGET nvim APPEND_STRING PROPERTY LINK_FLAGS "-fsanitize=thread ")
 endif()
 

src/nvim/globals.h

@@ -788,8 +788,28 @@ EXTERN int vr_lines_changed INIT(= 0);      /* #Lines changed by "gR" so far */
 /// Encoding used when 'fencs' is set to "default"
 EXTERN char_u *fenc_default INIT(= NULL);
 
-// To speed up BYTELEN() we keep a table with the byte lengths for utf-8
-EXTERN char utf8len_tab[256];
+// To speed up BYTELEN(); keep a lookup table to quickly get the length in
+// bytes of a UTF-8 character from the first byte of a UTF-8 string.  Bytes
+// which are illegal when used as the first byte have a 1.  The NUL byte has
+// length 1.
+EXTERN char utf8len_tab[256] INIT(= {
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
+  3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
+  4, 4, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6, 1, 1,
+});
 
 # if defined(USE_ICONV) && defined(DYNAMIC_ICONV)
 /* Pointers to functions and variables to be loaded at runtime */

src/nvim/mbyte.c

@@ -68,24 +68,6 @@ struct interval {
 # include "unicode_tables.generated.h"
 #endif
 
-/*
- * Lookup table to quickly get the length in bytes of a UTF-8 character from
- * the first byte of a UTF-8 string.
- * Bytes which are illegal when used as the first byte have a 1.
- * The NUL byte has length 1.
- */
-char utf8len_tab[256] =
-{
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
-  2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,
-  3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,6,6,1,1,
-};
-
 /*
  * Like utf8len_tab above, but using a zero for illegal lead bytes.
  */

src/nvim/msgpack_rpc/helpers.c

@@ -326,7 +326,9 @@ void msgpack_rpc_from_string(String result, msgpack_packer *res)
   FUNC_ATTR_NONNULL_ARG(2)
 {
   msgpack_pack_str(res, result.size);
-  msgpack_pack_str_body(res, result.data, result.size);
+  if (result.size > 0) {
+    msgpack_pack_str_body(res, result.data, result.size);
+  }
 }
 
 typedef struct {

src/nvim/shada.c

@@ -1575,7 +1575,9 @@ static char *shada_filename(const char *file)
     do { \
       const String s_ = (s); \
       msgpack_pack_bin(spacker, s_.size); \
-      msgpack_pack_bin_body(spacker, s_.data, s_.size); \
+      if (s_.size > 0) { \
+        msgpack_pack_bin_body(spacker, s_.data, s_.size); \
+      } \
     } while (0)
 
 /// Write single ShaDa entry