Kaydet (Commit) 57bbe564 authored tarafından Stephan Bergmann's avatar Stephan Bergmann

More loplugin:cstylecast: idlc

auto-rewrite with <https://gerrit.libreoffice.org/#/c/47798/> "Enable
loplugin:cstylecast for some more cases" plus
solenv/clang-format/reformat-formatted-files

Change-Id: I1cb9e69b8138cb5bb63f18231018f166b67b3072
üst ac489a4d
......@@ -37,7 +37,7 @@ public:
AstDeclaration* addDeclaration(AstDeclaration* pDecl);
sal_uInt32 nMembers() const
{ return (sal_uInt32)(m_declarations.size()); }
{ return static_cast<sal_uInt32>(m_declarations.size()); }
DeclList::const_iterator getIteratorBegin() const
{ return m_declarations.begin(); }
DeclList::const_iterator getIteratorEnd() const
......
......@@ -146,51 +146,51 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_ushort:
if (ev->u.usval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.usval;
ev->u.sval = static_cast<sal_Int16>(ev->u.usval);
ev->et = ET_short;
return true;
case ET_long:
if (ev->u.lval < SAL_MIN_INT16 || ev->u.lval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.lval;
ev->u.sval = static_cast<sal_Int16>(ev->u.lval);
ev->et = ET_short;
return true;
case ET_ulong:
if (ev->u.ulval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.ulval;
ev->u.sval = static_cast<sal_Int16>(ev->u.ulval);
ev->et = ET_short;
return true;
case ET_hyper:
if (ev->u.hval < SAL_MIN_INT16 || ev->u.hval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.hval;
ev->u.sval = static_cast<sal_Int16>(ev->u.hval);
ev->et = ET_short;
return true;
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.uhval;
ev->u.sval = static_cast<sal_Int16>(ev->u.uhval);
ev->et = ET_short;
return true;
case ET_boolean:
ev->u.sval = (sal_Int16)ev->u.bval;
ev->u.sval = static_cast<sal_Int16>(ev->u.bval);
ev->et = ET_short;
return true;
case ET_float:
if (ev->u.fval < SAL_MIN_INT16 || ev->u.fval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.fval;
ev->u.sval = static_cast<sal_Int16>(ev->u.fval);
ev->et = ET_short;
return true;
case ET_double:
if (ev->u.dval < SAL_MIN_INT16 || ev->u.dval > SAL_MAX_INT16)
return false;
ev->u.sval = (sal_Int16)ev->u.dval;
ev->u.sval = static_cast<sal_Int16>(ev->u.dval);
ev->et = ET_short;
return true;
case ET_byte:
ev->u.sval = (sal_Int16)ev->u.byval;
ev->u.sval = static_cast<sal_Int16>(ev->u.byval);
ev->et = ET_short;
return true;
default:
......@@ -203,7 +203,7 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_short:
if (ev->u.sval < 0)
return false;
ev->u.usval = (sal_uInt16)ev->u.sval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.sval);
ev->et = ET_ushort;
return true;
case ET_ushort:
......@@ -211,45 +211,45 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_long:
if (ev->u.lval < 0 || ev->u.lval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.lval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.lval);
ev->et = ET_ushort;
return true;
case ET_ulong:
if (ev->u.ulval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.ulval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.ulval);
ev->et = ET_ushort;
return true;
case ET_hyper:
if (ev->u.hval < 0 || ev->u.hval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.hval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.hval);
ev->et = ET_ushort;
return true;
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.uhval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.uhval);
ev->et = ET_ushort;
return true;
case ET_boolean:
ev->u.usval = (sal_uInt16)ev->u.bval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.bval);
ev->et = ET_short;
return true;
case ET_float:
if (ev->u.fval < 0.0 || ev->u.fval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.fval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.fval);
ev->et = ET_short;
return true;
case ET_double:
if (ev->u.dval < 0.0 || ev->u.dval > SAL_MAX_UINT16)
return false;
ev->u.usval = (sal_uInt16)ev->u.dval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.dval);
ev->et = ET_short;
return true;
case ET_byte:
ev->u.usval = (sal_uInt16)ev->u.byval;
ev->u.usval = static_cast<sal_uInt16>(ev->u.byval);
ev->et = ET_ushort;
return true;
default:
......@@ -260,11 +260,11 @@ coerce_value(AstExprValue *ev, ExprType t)
switch (ev->et)
{
case ET_short:
ev->u.lval = (sal_Int32)ev->u.sval;
ev->u.lval = static_cast<sal_Int32>(ev->u.sval);
ev->et = ET_long;
return true;
case ET_ushort:
ev->u.lval = (sal_Int32)ev->u.usval;
ev->u.lval = static_cast<sal_Int32>(ev->u.usval);
ev->et = ET_long;
return true;
case ET_long:
......@@ -272,39 +272,39 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_ulong:
if (ev->u.ulval > SAL_MAX_INT32)
return false;
ev->u.lval = (sal_Int32)ev->u.ulval;
ev->u.lval = static_cast<sal_Int32>(ev->u.ulval);
ev->et = ET_long;
return true;
case ET_hyper:
if (ev->u.hval < SAL_MIN_INT32 || ev->u.hval > SAL_MAX_INT32)
return false;
ev->u.lval = (sal_Int32)ev->u.hval;
ev->u.lval = static_cast<sal_Int32>(ev->u.hval);
ev->et = ET_long;
return true;
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_INT32)
return false;
ev->u.lval = (sal_Int32)ev->u.uhval;
ev->u.lval = static_cast<sal_Int32>(ev->u.uhval);
ev->et = ET_long;
return true;
case ET_boolean:
ev->u.lval = (sal_Int32)ev->u.bval;
ev->u.lval = static_cast<sal_Int32>(ev->u.bval);
ev->et = ET_long;
return true;
case ET_float:
if (ev->u.fval < SAL_MIN_INT32 || ev->u.fval > SAL_MAX_INT32)
return false;
ev->u.lval = (sal_Int32)ev->u.fval;
ev->u.lval = static_cast<sal_Int32>(ev->u.fval);
ev->et = ET_long;
return true;
case ET_double:
if (ev->u.dval < SAL_MIN_INT32 || ev->u.dval > SAL_MAX_INT32)
return false;
ev->u.lval = (sal_Int32)ev->u.dval;
ev->u.lval = static_cast<sal_Int32>(ev->u.dval);
ev->et = ET_long;
return true;
case ET_byte:
ev->u.lval = (sal_Int32) ev->u.byval;
ev->u.lval = static_cast<sal_Int32>(ev->u.byval);
ev->et = ET_long;
return true;
default:
......@@ -317,17 +317,17 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_short:
if (ev->u.sval < 0)
return false;
ev->u.ulval = (sal_uInt32)ev->u.sval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.sval);
ev->et = ET_ulong;
return true;
case ET_ushort:
ev->u.ulval = (sal_uInt32)ev->u.usval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.usval);
ev->et = ET_ulong;
return true;
case ET_long:
if (ev->u.lval < 0)
return false;
ev->u.ulval = (sal_uInt32)ev->u.lval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.lval);
ev->et = ET_ulong;
return true;
case ET_ulong:
......@@ -335,33 +335,33 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_hyper:
if (ev->u.hval < 0 || ev->u.hval > SAL_MAX_UINT32)
return false;
ev->u.lval = (sal_uInt32)ev->u.hval;
ev->u.lval = static_cast<sal_uInt32>(ev->u.hval);
ev->et = ET_ulong;
return true;
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_UINT32)
return false;
ev->u.ulval = (sal_uInt32)ev->u.uhval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.uhval);
ev->et = ET_ulong;
return true;
case ET_boolean:
ev->u.ulval = (sal_uInt32)ev->u.bval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.bval);
ev->et = ET_ulong;
return true;
case ET_float:
if (ev->u.fval < 0.0 || ev->u.fval > SAL_MAX_UINT32)
return false;
ev->u.ulval = (sal_uInt32)ev->u.fval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.fval);
ev->et = ET_ulong;
return true;
case ET_double:
if (ev->u.dval < 0.0 || ev->u.dval > SAL_MAX_UINT32)
return false;
ev->u.ulval = (sal_uInt32)ev->u.dval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.dval);
ev->et = ET_ulong;
return true;
case ET_byte:
ev->u.ulval = (sal_uInt32)ev->u.byval;
ev->u.ulval = static_cast<sal_uInt32>(ev->u.byval);
ev->et = ET_ulong;
return true;
default:
......@@ -372,19 +372,19 @@ coerce_value(AstExprValue *ev, ExprType t)
switch (ev->et)
{
case ET_short:
ev->u.hval = (sal_Int64)ev->u.sval;
ev->u.hval = static_cast<sal_Int64>(ev->u.sval);
ev->et = ET_hyper;
return true;
case ET_ushort:
ev->u.hval = (sal_Int64)ev->u.usval;
ev->u.hval = static_cast<sal_Int64>(ev->u.usval);
ev->et = ET_hyper;
return true;
case ET_long:
ev->u.hval = (sal_Int64)ev->u.lval;
ev->u.hval = static_cast<sal_Int64>(ev->u.lval);
ev->et = ET_hyper;
return true;
case ET_ulong:
ev->u.hval = (sal_Int64)ev->u.ulval;
ev->u.hval = static_cast<sal_Int64>(ev->u.ulval);
ev->et = ET_hyper;
return true;
case ET_hyper:
......@@ -392,27 +392,27 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_INT64)
return false;
ev->u.hval = (sal_Int64)ev->u.uhval;
ev->u.hval = static_cast<sal_Int64>(ev->u.uhval);
ev->et = ET_long;
return true;
case ET_boolean:
ev->u.hval = (sal_Int64)ev->u.bval;
ev->u.hval = static_cast<sal_Int64>(ev->u.bval);
ev->et = ET_hyper;
return true;
case ET_float:
if (ev->u.fval < SAL_MIN_INT64 || ev->u.fval > SAL_MAX_INT64)
return false;
ev->u.hval = (sal_Int64)ev->u.fval;
ev->u.hval = static_cast<sal_Int64>(ev->u.fval);
ev->et = ET_hyper;
return true;
case ET_double:
if (ev->u.dval < SAL_MIN_INT64 || ev->u.dval > SAL_MAX_INT64)
return false;
ev->u.hval = (sal_Int64)ev->u.dval;
ev->u.hval = static_cast<sal_Int64>(ev->u.dval);
ev->et = ET_hyper;
return true;
case ET_byte:
ev->u.hval = (sal_Int64)ev->u.byval;
ev->u.hval = static_cast<sal_Int64>(ev->u.byval);
ev->et = ET_hyper;
return true;
default:
......@@ -425,49 +425,49 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_short:
if (ev->u.sval < 0)
return false;
ev->u.uhval = (sal_uInt64)ev->u.sval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.sval);
ev->et = ET_uhyper;
return true;
case ET_ushort:
ev->u.uhval = (sal_uInt64)ev->u.usval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.usval);
ev->et = ET_uhyper;
return true;
case ET_long:
if (ev->u.lval < 0)
return false;
ev->u.uhval = (sal_uInt64)ev->u.lval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.lval);
ev->et = ET_uhyper;
return true;
case ET_ulong:
ev->u.uhval = (sal_uInt64)ev->u.ulval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.ulval);
ev->et = ET_uhyper;
return true;
case ET_hyper:
if (ev->u.hval < 0)
return false;
ev->u.uhval = (sal_uInt64)ev->u.hval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.hval);
ev->et = ET_uhyper;
return true;
case ET_uhyper:
return true;
case ET_boolean:
ev->u.uhval = (sal_uInt64)ev->u.bval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.bval);
ev->et = ET_uhyper;
return true;
case ET_float:
if (ev->u.fval < 0.0 || ev->u.fval > SAL_MAX_UINT64)
return false;
ev->u.uhval = (sal_uInt64)ev->u.fval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.fval);
ev->et = ET_uhyper;
return true;
case ET_double:
if (ev->u.dval < 0.0 || ev->u.dval > SAL_MAX_UINT64)
return false;
ev->u.uhval = (sal_uInt64)ev->u.dval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.dval);
ev->et = ET_uhyper;
return true;
case ET_byte:
ev->u.uhval = (sal_uInt64)ev->u.byval;
ev->u.uhval = static_cast<sal_uInt64>(ev->u.byval);
ev->et = ET_uhyper;
return true;
default:
......@@ -523,29 +523,29 @@ coerce_value(AstExprValue *ev, ExprType t)
switch (ev->et)
{
case ET_short:
ev->u.fval = (float)ev->u.sval;
ev->u.fval = static_cast<float>(ev->u.sval);
ev->et = ET_float;
return true;
case ET_ushort:
ev->u.fval = (float)ev->u.usval;
ev->u.fval = static_cast<float>(ev->u.usval);
ev->et = ET_float;
return true;
case ET_long:
ev->u.fval = (float)ev->u.lval;
ev->u.fval = static_cast<float>(ev->u.lval);
ev->et = ET_float;
return true;
case ET_ulong:
ev->u.fval = (float)ev->u.ulval;
ev->u.fval = static_cast<float>(ev->u.ulval);
ev->et = ET_float;
return true;
case ET_hyper:
ev->u.fval = (float)ev->u.hval;
ev->u.fval = static_cast<float>(ev->u.hval);
ev->et = ET_float;
return true;
case ET_uhyper:
if ((float)ev->u.ulval > FLT_MAX)
if (static_cast<float>(ev->u.ulval) > FLT_MAX)
return false;
ev->u.fval = (float)ev->u.ulval;
ev->u.fval = static_cast<float>(ev->u.ulval);
ev->et = ET_float;
return true;
case ET_boolean:
......@@ -555,13 +555,13 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_float:
return true;
case ET_double:
if ((float)ev->u.dval > FLT_MAX || (float)ev->u.dval < -FLT_MAX)
if (static_cast<float>(ev->u.dval) > FLT_MAX || static_cast<float>(ev->u.dval) < -FLT_MAX)
return false;
ev->u.fval = (float)ev->u.dval;
ev->u.fval = static_cast<float>(ev->u.dval);
ev->et = ET_float;
return true;
case ET_byte:
ev->u.fval = (float)ev->u.byval;
ev->u.fval = static_cast<float>(ev->u.byval);
ev->et = ET_float;
return true;
default:
......@@ -572,29 +572,29 @@ coerce_value(AstExprValue *ev, ExprType t)
switch (ev->et)
{
case ET_short:
ev->u.dval = (double)ev->u.sval;
ev->u.dval = static_cast<double>(ev->u.sval);
ev->et = ET_double;
return true;
case ET_ushort:
ev->u.dval = (double)ev->u.usval;
ev->u.dval = static_cast<double>(ev->u.usval);
ev->et = ET_double;
return true;
case ET_long:
ev->u.dval = (double)ev->u.lval;
ev->u.dval = static_cast<double>(ev->u.lval);
ev->et = ET_double;
return true;
case ET_ulong:
ev->u.dval = (double)ev->u.ulval;
ev->u.dval = static_cast<double>(ev->u.ulval);
ev->et = ET_double;
return true;
case ET_hyper:
ev->u.dval = (double)ev->u.hval;
ev->u.dval = static_cast<double>(ev->u.hval);
ev->et = ET_double;
return true;
case ET_uhyper:
if (ev->u.dval > FLT_MAX || ev->u.dval < -FLT_MAX)
return false;
ev->u.dval = (double)ev->u.ulval;
ev->u.dval = static_cast<double>(ev->u.ulval);
ev->et = ET_double;
return true;
case ET_boolean:
......@@ -602,13 +602,13 @@ coerce_value(AstExprValue *ev, ExprType t)
ev->et = ET_double;
return true;
case ET_float:
ev->u.dval = (double)ev->u.fval;
ev->u.dval = static_cast<double>(ev->u.fval);
ev->et = ET_double;
return true;
case ET_double:
return true;
case ET_byte:
ev->u.dval = (double)ev->u.byval;
ev->u.dval = static_cast<double>(ev->u.byval);
ev->et = ET_double;
return true;
default:
......@@ -621,37 +621,37 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_short:
if (ev->u.sval < SAL_MIN_INT8 || ev->u.sval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char)ev->u.sval;
ev->u.byval = static_cast<unsigned char>(ev->u.sval);
ev->et = ET_byte;
return true;
case ET_ushort:
if (ev->u.usval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char)ev->u.usval;
ev->u.byval = static_cast<unsigned char>(ev->u.usval);
ev->et = ET_byte;
return true;
case ET_long:
if (ev->u.lval < SAL_MIN_INT8 || ev->u.lval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.lval;
ev->u.byval = static_cast<unsigned char>(ev->u.lval);
ev->et = ET_byte;
return true;
case ET_ulong:
if (ev->u.ulval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.ulval;
ev->u.byval = static_cast<unsigned char>(ev->u.ulval);
ev->et = ET_byte;
return true;
case ET_hyper:
if (ev->u.hval < SAL_MIN_INT8 || ev->u.hval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.hval;
ev->u.byval = static_cast<unsigned char>(ev->u.hval);
ev->et = ET_byte;
return true;
case ET_uhyper:
if (ev->u.uhval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.uhval;
ev->u.byval = static_cast<unsigned char>(ev->u.uhval);
ev->et = ET_byte;
return true;
case ET_boolean:
......@@ -661,13 +661,13 @@ coerce_value(AstExprValue *ev, ExprType t)
case ET_float:
if (ev->u.fval < SAL_MIN_INT8 || ev->u.fval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.fval;
ev->u.byval = static_cast<unsigned char>(ev->u.fval);
ev->et = ET_byte;
return true;
case ET_double:
if (ev->u.dval < SAL_MIN_INT8 || ev->u.dval > SAL_MAX_UINT8)
return false;
ev->u.byval = (unsigned char) ev->u.dval;
ev->u.byval = static_cast<unsigned char>(ev->u.dval);
ev->et = ET_byte;
return true;
case ET_byte:
......
......@@ -41,7 +41,7 @@ bool AstOperation::isVariadic() const {
bool AstOperation::dumpBlob(typereg::Writer & rBlob, sal_uInt16 index)
{
sal_uInt16 nParam = getNodeCount(NT_parameter);
sal_uInt16 nExcep = (sal_uInt16)m_exceptions.size();
sal_uInt16 nExcep = static_cast<sal_uInt16>(m_exceptions.size());
OUString returnTypeName;
if (m_pReturnType == nullptr) {
......
......@@ -136,7 +136,7 @@ OString makeTempName(const OString& prefix)
#if defined(SAL_W32) || defined(SAL_UNX)
OSL_ASSERT( sizeof(tmpFilePattern) >
(size_t) ( tmpPath.getLength()
static_cast<size_t>( tmpPath.getLength()
+ RTL_CONSTASCII_LENGTH( PATH_SEPARATOR )
+ prefix.getLength()
+ RTL_CONSTASCII_LENGTH( "XXXXXX") ) );
......@@ -329,7 +329,7 @@ sal_Int32 compileFile(const OString * pathname)
nullptr, startDir.pData, nullptr, 0, &hProcess );
oslProcessInfo hInfo;
hInfo.Size = (sal_uInt32)(sizeof(oslProcessInfo));
hInfo.Size = sal_uInt32(sizeof(oslProcessInfo));
if (osl_getProcessInfo(hProcess, osl_Process_EXITCODE, &hInfo)
!= osl_Process_E_None)
{
......
......@@ -76,7 +76,7 @@ SAL_IMPLEMENT_MAIN_WITH_ARGS(argc, argv)
if ( options.verbose() )
{
fprintf( stdout, "%s: compiling %i source files ... \n",
options.getProgramName().getStr(), (int)files.size() );
options.getProgramName().getStr(), static_cast<int>(files.size()) );
fflush( stdout );
}
for (std::vector< OString >::const_iterator i(files.begin());
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment