orthanc: Core/Toolbox.cpp comparison

comparison Core/Toolbox.cpp @ 3333:2a38e00a0638

Size of the Orthanc static binaries are reduced by compressing ICU data

author	Sebastien Jodogne <s.jodogne@gmail.com>
date	Wed, 27 Mar 2019 15:25:18 +0100
parents	f0c92ecd09c8
children	a2862d4484f5

comparison

equal deleted inserted replaced

-:79178122842c
+:2a38e00a0638
 #  include <uuid/uuid.h>
 #endif
 }
+#if defined(ORTHANC_STATIC_ICU)
+#  if (ORTHANC_STATIC_ICU == 1 && ORTHANC_ENABLE_LOCALE == 1)
+#    include <EmbeddedResources.h>
+#    include <unicode/udata.h>
+#    include <unicode/uloc.h>
+#    include "Compression/GzipCompressor.h"
+static std::string  globalIcuData_;
+extern "C"
+{
+// This is dummy content for the "icudt58_dat" (resp. "icudt63_dat")
+// global variable from the autogenerated "icudt58l_dat.c"
+// (resp. "icudt63l_dat.c") file that contains a huge C array. In
+// Orthanc, this array is compressed using gzip and attached as a
+// resource, then uncompressed during the launch of Orthanc by
+// static function "InitializeIcu()".
+struct
+{
+double bogus;
+uint8_t *bytes;
+} U_ICUDATA_ENTRY_POINT = { 0.0, NULL };
+}
+#  endif
+#endif
 namespace Orthanc
 {
 void Toolbox::LinesIterator::FindEndOfLine()
 {
 // http://dicom.nema.org/medical/dicom/current/output/chtml/part03/sect_C.12.html#sect_C.12.1.1.2
 std::string Toolbox::ConvertToUtf8(const std::string& source,
 Encoding sourceEncoding,
 bool hasCodeExtensions)
 {
+#if ORTHANC_STATIC_ICU == 1
+if (globalIcuData_.empty())
+{
+throw OrthancException(ErrorCode_BadSequenceOfCalls,
+"Call Toolbox::InitializeGlobalLocale()");
+}
+#endif
 // The "::skip" flag makes boost skip invalid UTF-8
 // characters. This can occur in badly-encoded DICOM files.
 try
 {
 #if ORTHANC_ENABLE_LOCALE == 1
 std::string Toolbox::ConvertFromUtf8(const std::string& source,
 Encoding targetEncoding)
 {
+#if ORTHANC_STATIC_ICU == 1
+if (globalIcuData_.empty())
+{
+throw OrthancException(ErrorCode_BadSequenceOfCalls,
+"Call Toolbox::InitializeGlobalLocale()");
+}
+#endif
 // The "::skip" flag makes boost skip invalid UTF-8
 // characters. This can occur in badly-encoded DICOM files.
 try
 {
 {
 }
 return (globalLocale_.get() != NULL);
 }
+static void InitializeIcu()
+{
+#if ORTHANC_STATIC_ICU == 1
+if (globalIcuData_.empty())
+{
+LOG(INFO) << "Setting up the ICU common data";
+GzipCompressor compressor;
+compressor.Uncompress(globalIcuData_,
+EmbeddedResources::GetFileResourceBuffer(EmbeddedResources::LIBICU_DATA),
+EmbeddedResources::GetFileResourceSize(EmbeddedResources::LIBICU_DATA));
+std::string md5;
+Toolbox::ComputeMD5(md5, globalIcuData_);
+if (md5 != ORTHANC_ICU_DATA_MD5 ||
+globalIcuData_.empty())
+{
+throw OrthancException(ErrorCode_InternalError,
+"Cannot decode the ICU common data");
+}
+// "ICU data is designed to be 16-aligned"
+// http://userguide.icu-project.org/icudata#TOC-Alignment
+{
+static const size_t ALIGN = 16;
+UErrorCode status = U_ZERO_ERROR;
+if (reinterpret_cast<intptr_t>(globalIcuData_.c_str()) % ALIGN == 0)
+{
+// Data is already properly aligned
+udata_setCommonData(globalIcuData_.c_str(), &status);
+}
+else
+{
+std::string aligned;
+aligned.resize(globalIcuData_.size() + ALIGN - 1);
+intptr_t offset = reinterpret_cast<intptr_t>(aligned.c_str()) % ALIGN;
+if (offset != 0)
+{
+offset = ALIGN - offset;
+}
+if (offset + globalIcuData_.size() > aligned.size())
+{
+throw OrthancException(ErrorCode_InternalError, "Cannot align on 16-bytes boundary");
+}
+// We don't use "memcpy()", as it expects its data to be aligned
+const uint8_t* p = reinterpret_cast<uint8_t*>(&globalIcuData_[0]);
+uint8_t* q = reinterpret_cast<uint8_t*>(&aligned[0]) + offset;
+for (size_t i = 0; i < globalIcuData_.size(); i++, p++, q++)
+{
+*q = *p;
+}
+globalIcuData_.swap(aligned);
+const uint8_t* data = reinterpret_cast<const uint8_t*>(globalIcuData_.c_str()) + offset;
+if (reinterpret_cast<intptr_t>(data) % ALIGN != 0)
+{
+throw OrthancException(ErrorCode_InternalError, "Cannot align on 16-bytes boundary");
+}
+else
+{
+udata_setCommonData(data, &status);
+}
+}
+if (status != U_ZERO_ERROR)
+{
+throw OrthancException(ErrorCode_InternalError, "Cannot initialize ICU");
+}
+}
+if (Toolbox::DetectEndianness() != Endianness_Little)
+{
+// TODO - The data table must be swapped (uint16_t)
+throw OrthancException(ErrorCode_NotImplemented);
+}
+// "First-use of ICU from a single thread before the
+// multi-threaded use of ICU begins", to make sure everything is
+// properly initialized (should not be mandatory in our
+// case). We let boost handle calls to "u_init()" and "u_cleanup()".
+// http://userguide.icu-project.org/design#TOC-ICU-Initialization-and-Termination
+uloc_getDefault();
+}
+#endif
+}
 void Toolbox::InitializeGlobalLocale(const char* locale)
 {
+InitializeIcu();
 // Make Orthanc use English, United States locale
 // Linux: use "en_US.UTF-8"
 // Windows: use ""
 // Wine: use NULL
 }
 std::string Toolbox::ToUpperCaseWithAccents(const std::string& source)
 {
-if (globalLocale_.get() == NULL)
+bool error = (globalLocale_.get() == NULL);
+#if ORTHANC_STATIC_ICU == 1
+if (globalIcuData_.empty())
+{
+error = true;
+}
+#endif
+if (error)
 {
 throw OrthancException(ErrorCode_BadSequenceOfCalls,
 "No global locale was set, call Toolbox::InitializeGlobalLocale()");
 }

Mercurial > hg > orthanc

comparison Core/Toolbox.cpp @ 3333:2a38e00a0638