fdo#81272 Speed up break iterators

Change-Id: Idcd7de83ff7e551b350db0bf80e2ac78baa67686
Reviewed-on: https://gerrit.libreoffice.org/11374
Reviewed-by: Caolán McNamara <caolanm@redhat.com>
Tested-by: Caolán McNamara <caolanm@redhat.com>

3 files changed, 22 insertions, 1 deletions

diff --git a/i18npool/inc/xdictionary.hxx b/i18npool/inc/xdictionary.hxx
index 57f5238ac853..f7c0110d39fd 100644
--- a/i18npool/inc/xdictionary.hxx
+++ b/i18npool/inc/xdictionary.hxx
@@ -76,6 +76,8 @@ public:
 
 private:
     WordBreakCache cache[CACHE_MAX];
+    OUString segmentCachedString;
+    Boundary segmentCachedBoundary;
 
     bool        seekSegment(const OUString& rText, sal_Int32 pos, Boundary& boundary);
     WordBreakCache& getCache(const sal_Unicode *text, Boundary& boundary);
diff --git a/i18npool/source/breakiterator/breakiterator_unicode.cxx b/i18npool/source/breakiterator/breakiterator_unicode.cxx
index 5270b1db5bad..658d4ff861ea 100644
--- a/i18npool/source/breakiterator/breakiterator_unicode.cxx
+++ b/i18npool/source/breakiterator/breakiterator_unicode.cxx
@@ -185,7 +185,7 @@ void SAL_CALL BreakIterator_Unicode::loadICUBreakIterator(const com::sun::star::
         }
     }
 
-    if (newBreak || !icuBI->aICUText.equals(rText))
+    if (newBreak || icuBI->aICUText.pData != rText.pData)
     {
         // UChar != sal_Unicode in MinGW
         const UChar *pText = reinterpret_cast<const UChar *>(rText.getStr());
diff --git a/i18npool/source/breakiterator/xdictionary.cxx b/i18npool/source/breakiterator/xdictionary.cxx
index 1067e33bcecc..95e1cc957f86 100644
--- a/i18npool/source/breakiterator/xdictionary.cxx
+++ b/i18npool/source/breakiterator/xdictionary.cxx
@@ -271,6 +271,21 @@ bool xdictionary::seekSegment(const OUString &rText, sal_Int32 pos,
     Boundary& segBoundary)
 {
     sal_Int32 indexUtf16;
+
+    if (segmentCachedString.pData != rText.pData) {
+        // Cache the passed text so we can avoid regenerating the segment if it's the same
+        // (pData is refcounted and assigning the OUString references it, which ensures that
+        // the object is the same if we get the same pointer back later)
+        segmentCachedString = rText;
+    } else {
+        // If pos is within the cached boundary, use that boundary
+        if (pos >= segmentCachedBoundary.startPos && pos <= segmentCachedBoundary.endPos) {
+            indexUtf16 = segmentCachedBoundary.startPos;
+            rText.iterateCodePoints(&indexUtf16, 1);
+            return segmentCachedBoundary.endPos > indexUtf16;
+        }
+    }
+
     segBoundary.endPos = segBoundary.startPos = pos;
 
     indexUtf16 = pos;
@@ -293,6 +308,10 @@ bool xdictionary::seekSegment(const OUString &rText, sal_Int32 pos,
             break;
     }
 
+    // Cache the calculated boundary
+    segmentCachedBoundary.startPos = segBoundary.startPos;
+    segmentCachedBoundary.endPos = segBoundary.endPos;
+
     indexUtf16 = segBoundary.startPos;
     rText.iterateCodePoints(&indexUtf16, 1);
     return segBoundary.endPos > indexUtf16;