summaryrefslogtreecommitdiff
path: root/linguistic/source/gciterator.cxx
diff options
context:
space:
mode:
Diffstat (limited to 'linguistic/source/gciterator.cxx')
-rw-r--r--linguistic/source/gciterator.cxx19
1 files changed, 8 insertions, 11 deletions
diff --git a/linguistic/source/gciterator.cxx b/linguistic/source/gciterator.cxx
index 0b284d660abc..d0c51f90b464 100644
--- a/linguistic/source/gciterator.cxx
+++ b/linguistic/source/gciterator.cxx
@@ -417,10 +417,9 @@ void GrammarCheckingIterator::ProcessResult(
text::TextMarkupDescriptor * pDescriptors = aDescriptors.getArray();
// at pos 0 .. nErrors-1 -> all grammar errors
- for (sal_Int32 i = 0; i < nErrors; ++i)
+ for (const linguistic2::SingleProofreadingError &rError : rRes.aErrors)
{
- const linguistic2::SingleProofreadingError &rError = rRes.aErrors[i];
- text::TextMarkupDescriptor &rDesc = aDescriptors[i];
+ text::TextMarkupDescriptor &rDesc = *pDescriptors++;
rDesc.nType = rError.nErrorType;
rDesc.nOffset = rError.nErrorStart;
@@ -455,9 +454,9 @@ void GrammarCheckingIterator::ProcessResult(
// at pos nErrors -> sentence markup
// nSentenceLength: includes the white-spaces following the sentence end...
const sal_Int32 nSentenceLength = rRes.nStartOfNextSentencePosition - rRes.nStartOfSentencePosition;
- pDescriptors[ nErrors ].nType = text::TextMarkupType::SENTENCE;
- pDescriptors[ nErrors ].nOffset = rRes.nStartOfSentencePosition;
- pDescriptors[ nErrors ].nLength = nSentenceLength;
+ pDescriptors->nType = text::TextMarkupType::SENTENCE;
+ pDescriptors->nOffset = rRes.nStartOfSentencePosition;
+ pDescriptors->nLength = nSentenceLength;
xMulti->commitMultiTextMarkup( aDescriptors ) ;
}
@@ -1081,20 +1080,18 @@ void GrammarCheckingIterator::GetConfiguredGCSvcs_Impl()
uno::Reference< container::XNameAccess > xNA( GetUpdateAccess(), uno::UNO_QUERY_THROW );
xNA.set( xNA->getByName( "GrammarCheckerList" ), uno::UNO_QUERY_THROW );
const uno::Sequence< OUString > aElementNames( xNA->getElementNames() );
- const OUString *pElementNames = aElementNames.getConstArray();
- sal_Int32 nLen = aElementNames.getLength();
- for (sal_Int32 i = 0; i < nLen; ++i)
+ for (const OUString& rElementName : aElementNames)
{
uno::Sequence< OUString > aImplNames;
- uno::Any aTmp( xNA->getByName( pElementNames[i] ) );
+ uno::Any aTmp( xNA->getByName( rElementName ) );
if (aTmp >>= aImplNames)
{
if (aImplNames.hasElements())
{
// only the first entry is used, there should be only one grammar checker per language
const OUString aImplName( aImplNames[0] );
- const LanguageType nLang = LanguageTag::convertToLanguageType( pElementNames[i] );
+ const LanguageType nLang = LanguageTag::convertToLanguageType( rElementName );
aTmpGCImplNamesByLang[ nLang ] = aImplName;
}
}