Merge branch 'develop'

.gitignore

@@ -8,6 +8,7 @@
 /config.log
 mpack-test-file
 /mpack-test-dir/
+/analysis/
 
 # folders
 /tags

.travis.yml

@@ -1,8 +1,5 @@
 language: c
 
-before_install:
-  - pip install --user cpp-coveralls
-
 addons:
   apt:
     packages:

Doxyfile

@@ -7,7 +7,6 @@ PROJECT_NUMBER = develop
 
 INPUT = \
     README.md \
-    src/mpack/mpack-platform.h \
     src/mpack/mpack-common.h \
     src/mpack/mpack-reader.h \
     src/mpack/mpack-writer.h \
@@ -22,8 +21,9 @@ STRIP_FROM_PATH = . ./src
 
 PREDEFINED = \
     inline= \
-    MPACK_INLINE= \
     MPACK_ALWAYS_INLINE= \
+    MPACK_INLINE= \
+    MPACK_INLINE_SPEED= \
     \
     MPACK_READER=1 \
     MPACK_WRITER=1 \

SConscript

@@ -24,7 +24,7 @@ prog = env.Program("mpack-test", srcs,
 if platform.machine() in ["i386", "x86_64"]:
     valgrind = "valgrind --leak-check=full --error-exitcode=1 "
     # travis version of valgrind is too old, doesn't support leak kinds
-    if "TRAVIS" not in env:
+    if "TRAVIS" not in env["ENV"]:
         valgrind = valgrind + "--show-leak-kinds=all --errors-for-leak-kinds=all "
         valgrind = valgrind + "--suppressions=tools/valgrind-suppressions "
 else:

SConstruct

@@ -19,12 +19,14 @@ env = Environment()
 conf = Configure(env, custom_tests = {'CheckFlags': CheckFlags})
 
 for x in os.environ.keys():
-    if x in ["CC", "CXX", "PATH", "TRAVIS", "TERM"] or x.startswith("CLANG_") or x.startswith("CCC_"):
+    if x in ["CC", "CXX"]:
         env[x] = os.environ[x]
+    if x in ["PATH", "TRAVIS", "TERM"] or x.startswith("CLANG_") or x.startswith("CCC_"):
+        env["ENV"][x] = os.environ[x]
 
 env.Append(CPPFLAGS = [
     "-Wall", "-Wextra", "-Werror",
-    "-Wconversion", "-Wno-sign-conversion", "-Wundef",
+    "-Wconversion", "-Wno-sign-conversion", "-Wundef", "-Wshadow",
     "-Isrc", "-Itest",
     "-DMPACK_SCONS=1",
     "-g",
@@ -34,7 +36,7 @@ env.Append(LINKFLAGS = [
     ])
 # Additional warning flags are passed in SConscript based on the language (C/C++)
 
-if 'TRAVIS' not in env:
+if 'TRAVIS' not in env["ENV"]:
     # Travis-CI currently uses Clang 3.4 which does not support this option,
     # and it also appears to be incompatible with other GCC options on Travis-CI
     env.Append(CPPFLAGS = ["-Wno-float-conversion"])
@@ -60,7 +62,7 @@ if hasOg:
     debugflags = ["-DDEBUG", "-Og"]
 else:
     debugflags = ["-DDEBUG", "-O0"]
-releaseflags = ["-Os"]
+releaseflags = ["-O2"]
 cflags = ["-std=c99"]
 
 gcovflags = []
@@ -95,7 +97,6 @@ def AddBuilds(variant_dir, cppflags, linkflags = []):
 
 # The default build, everything in debug. This is the build used
 # for code coverage measurement and static analysis.
-
 AddBuild("debug", allfeatures + allconfigs + debugflags + cflags + gcovflags, gcovflags)
 
 
@@ -105,6 +106,8 @@ if ARGUMENTS.get('more') or ARGUMENTS.get('all'):
     AddBuild("release", allfeatures + allconfigs + releaseflags + cflags)
     AddBuilds("embed", allfeatures + cflags)
     AddBuilds("noio", allfeatures + noioconfigs + cflags)
+    AddBuild("debug-size", ["-DMPACK_OPTIMIZE_FOR_SIZE=1"] + debugflags + allfeatures + allconfigs + cflags)
+    AddBuild("release-size", ["-Os"] + allfeatures + allconfigs + cflags)
 
 
 # Run "scons all=1" to run all builds. This is what the CI runs.
@@ -113,8 +116,6 @@ if ARGUMENTS.get('all'):
     # various release builds
     AddBuild("release-unopt", allfeatures + allconfigs + cflags + ["-O0"])
     AddBuild("release-fastmath", allfeatures + allconfigs + releaseflags + cflags + ["-ffast-math"])
-    AddBuild("release-speed", ["-DMPACK_OPTIMIZE_FOR_SIZE=0"] +
-            allfeatures + allconfigs + releaseflags + cflags)
     if conf.CheckFlags(ltoflags, ltoflags, "-flto"):
         AddBuild("release-lto", allfeatures + allconfigs + ltoflags + cflags, ltoflags)
 

src/mpack-config.h.sample

@@ -180,15 +180,14 @@
  */
 
 /**
- * Whether to optimize for size or speed. Optimizing for size causes
- * very few functions to be declared inline, and can save a couple
- * kilobytes of space in the resulting executable.
+ * Whether to optimize for size or speed.
+ *
+ * Optimizing for size simplifies some parsing and encoding algorithms
+ * at the expense of speed, and saves a few kilobytes of space in the
+ * resulting executable.
  *
  * This automatically detects -Os with GCC/Clang. Unfortunately there
  * doesn't seem to be a macro defined for /Os under MSVC.
- *
- * This feature is currently experimental and may be removed in a
- * future release.
  */
 #ifndef MPACK_OPTIMIZE_FOR_SIZE
 #ifdef __OPTIMIZE_SIZE__
@@ -199,32 +198,41 @@
 #endif
 
 /**
- * Stack space to use when initializing a reader or writer with a
- * stack-allocated buffer.
+ * Stack space in bytes to use when initializing a reader or writer
+ * with a stack-allocated buffer.
  */
 #ifndef MPACK_STACK_SIZE
 #define MPACK_STACK_SIZE 4096
 #endif
 
 /**
  * Buffer size to use for allocated buffers (such as for a file writer.)
+ *
+ * Starting with a single page and growing as needed seems to
+ * provide the best performance with minimal memory waste.
+ * Increasing this does not improve performance even when writing
+ * huge messages.
  */
 #ifndef MPACK_BUFFER_SIZE
-#define MPACK_BUFFER_SIZE 65536
+#define MPACK_BUFFER_SIZE 4096
 #endif
 
 /**
- * Number of nodes in each allocated node page.
+ * Size of an allocated node page in bytes.
+ *
+ * The children for a given compound element must be contiguous, so
+ * larger pages than this may be allocated as needed. (Safety checks
+ * exist to prevent malicious data from causing too large allocations.)
  *
- * Nodes are 16 bytes when compiled for a 32-bit architecture and
- * 24 bytes when compiled for a 64-bit architecture.
+ * Nodes are 12 bytes when compiled for a 32-bit architecture and
+ * 16 bytes when compiled for a 64-bit architecture.
  *
  * Using as many nodes fit in one memory page seems to provide the
  * best performance, and has very little waste when parsing small
  * messages.
  */
 #ifndef MPACK_NODE_PAGE_SIZE
-#define MPACK_NODE_PAGE_SIZE (4096 / sizeof(mpack_node_t))
+#define MPACK_NODE_PAGE_SIZE 4096
 #endif
 
 /**

src/mpack/mpack-common.c

@@ -311,11 +311,14 @@ mpack_error_t mpack_track_destroy(mpack_track_t* track, bool cancel) {
 
 
 /* The below code is from Bjoern Hoehrmann's Flexible and Economical */
-/* UTF-8 decoder, modified to support MPack inlining and add the mpack prefix. */
+/* UTF-8 decoder, modified to add the mpack prefix. */
 
 /* Copyright (c) 2008-2010 Bjoern Hoehrmann <bjoern@hoehrmann.de> */
 /* See http://bjoern.hoehrmann.de/utf-8/decoder/dfa/ for details. */
 
+#define MPACK_UTF8_ACCEPT 0
+#define MPACK_UTF8_REJECT 12
+
 static const uint8_t mpack_utf8d[] = {
   /* The first part of the table maps bytes to character classes that */
   /* to reduce the size of the transition table and create bitmasks. */
@@ -337,7 +340,25 @@ static const uint8_t mpack_utf8d[] = {
   12,36,12,12,12,12,12,12,12,12,12,12,
 };
 
-uint32_t mpack_utf8_decode(uint32_t* state, uint32_t* codep, uint8_t byte) {
+/**
+ * Parses one byte from a UTF-8 stream.
+ *
+ * Returns and sets state to:
+ *   - MPACK_UTF8_ACCEPT if the byte completes a valid unicode code point, placing it in codep
+ *   - MPACK_UTF8_REJECT if the byte is invalid UTF-8
+ *   - something else if more bytes are needed to form a valid character
+ *
+ * If more bytes are needed, this should be called again with the next byte
+ * in the string. state and codep should not be modified, since they will
+ * contain the partially read code point.
+ *
+ * The initial state should be set to MPACK_UTF8_ACCEPT before parsing a string.
+ *
+ * This does not accept any UTF-8 variant such as Modified UTF-8, CESU-8 or
+ * WTF-8. Overlong sequences and UTF-16 surrogates will be rejected. Only
+ * pure UTF-8 is accepted.
+ */
+static inline uint32_t mpack_utf8_decode(uint32_t* state, uint32_t* codep, uint8_t byte) {
   uint32_t type = mpack_utf8d[byte];
 
   *codep = (*state != MPACK_UTF8_ACCEPT) ?
@@ -352,7 +373,7 @@ uint32_t mpack_utf8_decode(uint32_t* state, uint32_t* codep, uint8_t byte) {
 
 
 
-bool mpack_utf8_check(char* str, size_t bytes) {
+bool mpack_utf8_check(const char* str, size_t bytes) {
     uint32_t state = MPACK_UTF8_ACCEPT;
     uint32_t codepoint;
     for (size_t i = 0; i < bytes; ++i)
@@ -361,7 +382,7 @@ bool mpack_utf8_check(char* str, size_t bytes) {
     return state == MPACK_UTF8_ACCEPT;
 }
 
-bool mpack_utf8_check_no_null(char* str, size_t bytes) {
+bool mpack_utf8_check_no_null(const char* str, size_t bytes) {
     uint32_t state = MPACK_UTF8_ACCEPT;
     uint32_t codepoint;
     for (size_t i = 0; i < bytes; ++i)
@@ -370,7 +391,7 @@ bool mpack_utf8_check_no_null(char* str, size_t bytes) {
     return state == MPACK_UTF8_ACCEPT;
 }
 
-bool mpack_str_check_no_null(char* str, size_t bytes) {
+bool mpack_str_check_no_null(const char* str, size_t bytes) {
     for (size_t i = 0; i < bytes; ++i)
         if (str[i] == '\0')
             return false;