summaryrefslogtreecommitdiff
path: root/unotools
diff options
context:
space:
mode:
authorTor Lillqvist <tml@collabora.com>2018-05-11 01:38:58 +0300
committerMike Kaganski <mike.kaganski@collabora.com>2018-05-11 13:45:24 +0200
commit2ff121f298e64ff204621220622472fe697d599d (patch)
tree02d0c8efbc99d3899ebdf047e08a4ae3939a6a1e /unotools
parentf619c9b48a414072d35b831c08c393f3611027b6 (diff)
Avoid gengal hanging in an --enable-dbgutil build on Windows
With a newer C++ debug runtime (in an --enable-dbgutil build), passing an invalid locale name causes an attempt to display an error dialog. Which does not even show up, at least for me, but instead the process (gengal, at least) just hangs. Which is far from ideal. Passing a POSIX-style locale name to the std::locale constructor on Windows is a bit odd, but apparently in the normal C++ runtime it "just" causes an exception to be thrown, that boost catches (see the loadable(std::string name) in boost's libs\locale\src\std\std_backend.cpp), and then instead uses the Windows style locale name it knows how to construct. (Why it even tries the POSIX style name on Windows I can't understand.) Actually it isn't just the locale name part "en_US" of a locale like "en_US.UTF-8" that is problematic, but also the encoding part, "UTF-8". The Microsoft C/C++ library does not support UTF-8 locales. The error message that our own report hook catches says: "f:\dd\vctools\crt\crtw32\stdcpp\xmbtowc.c(89) : Assertion failed: ploc->_Mbcurmax == 1 || ploc->_Mbcurmax == 2". Clearly in a UTF-8 locale (perhaps one that boost internally constructs?) the maximum bytes per character will be more than 2. With a debug C++ runtime, we need to avoid the error dialog, and just ignore the error. So we install an own CRT error report hook that ignores the error for the duration of the locale construcion. Change-Id: Ia2ca994f03d1ce94ce7e9d7607f204c320ab8f2d Reviewed-on: https://gerrit.libreoffice.org/54110 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
Diffstat (limited to 'unotools')
-rw-r--r--unotools/source/i18n/resmgr.cxx70
1 files changed, 69 insertions, 1 deletions
diff --git a/unotools/source/i18n/resmgr.cxx b/unotools/source/i18n/resmgr.cxx
index 9e0b8f02577b..f756575bc5bb 100644
--- a/unotools/source/i18n/resmgr.cxx
+++ b/unotools/source/i18n/resmgr.cxx
@@ -1,4 +1,4 @@
-/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; fill-column: 100 -*- */
/*
* This file is part of the LibreOffice project.
*
@@ -68,6 +68,13 @@
#include <unordered_map>
#include <memory>
+#if defined(_WIN32) && defined(DBG_UTIL)
+#include <o3tl/char16_t2wchar_t.hxx>
+#include <prewin.h>
+#include <crtdbg.h>
+#include <postwin.h>
+#endif
+
namespace
{
OUString createFromUtf8(const char* data, size_t size)
@@ -100,6 +107,26 @@ namespace
}
}
+#if defined(_WIN32) && defined(DBG_UTIL)
+static int IgnoringCrtReportHook(int reportType, wchar_t *message, int * /* returnValue */)
+{
+ OUString sType;
+ if (reportType == _CRT_WARN)
+ sType = "WARN";
+ else if (reportType == _CRT_ERROR)
+ sType = "ERROR";
+ else if (reportType == _CRT_ASSERT)
+ sType = "ASSERT";
+ else
+ sType = "?(" + OUString::number(reportType) + ")";
+
+ SAL_WARN("unotools.i18n", "CRT Report Hook: " << sType << ": " << OUString(o3tl::toU(message)));
+
+ return TRUE;
+}
+#endif
+
+
namespace Translate
{
std::locale Create(const sal_Char* pPrefixName, const LanguageTag& rLocale)
@@ -123,7 +150,48 @@ namespace Translate
bindtextdomain(pPrefixName, sPath.getStr());
#endif
gen.add_messages_domain(pPrefixName);
+
+#if defined(_WIN32) && defined(DBG_UTIL)
+ // With a newer C++ debug runtime (in an --enable-dbgutil build), passing an invalid locale
+ // name causes an attempt to display an error dialog. Which does not even show up, at least
+ // for me, but instead the process (gengal, at least) just hangs. Which is far from ideal.
+
+ // Passing a POSIX-style locale name to the std::locale constructor on Windows is a bit odd,
+ // but apparently in the normal C++ runtime it "just" causes an exception to be thrown, that
+ // boost catches (see the loadable(std::string name) in boost's
+ // libs\locale\src\std\std_backend.cpp), and then instead uses the Windows style locale name
+ // it knows how to construct. (Why does it even try the POSIX style name I can't
+ // understand.)
+
+ // Actually it isn't just the locale name part "en_US" of a locale like "en_US.UTF-8" that
+ // is problematic, but also the encoding part, "UTF-8". The Microsoft C/C++ library does not
+ // support UTF-8 locales. The error message that our own report hook catches says:
+ // "f:\dd\vctools\crt\crtw32\stdcpp\xmbtowc.c(89) : Assertion failed: ploc->_Mbcurmax == 1
+ // || ploc->_Mbcurmax == 2". Clearly in a UTF-8 locale (perhaps one that boost internally
+ // constructs?) the maximum bytes per character will be more than 2.
+
+ // With a debug C++ runtime, we need to avoid the error dialog, and just ignore the error.
+
+ struct CrtSetReportHook
+ {
+ int mnCrtSetReportHookSucceeded;
+
+ CrtSetReportHook()
+ {
+ mnCrtSetReportHookSucceeded = _CrtSetReportHookW2(_CRT_RPTHOOK_INSTALL, IgnoringCrtReportHook);
+ }
+
+ ~CrtSetReportHook()
+ {
+ if (mnCrtSetReportHookSucceeded >= 0)
+ _CrtSetReportHookW2(_CRT_RPTHOOK_REMOVE, IgnoringCrtReportHook);
+ }
+ } aHook;
+
+#endif
+
std::locale aRet(gen(sIdentifier.getStr()));
+
aCache[sUnique] = aRet;
return aRet;
}