Index: icu/trunk/source/layout/IndicReordering.cpp
===================================================================
--- icu/trunk/source/layout/IndicReordering.cpp	(revision 25772)
+++ icu/trunk/source/layout/IndicReordering.cpp	(revision 26090)
@@ -126,4 +126,8 @@
     FeatureMask fSMFeatures;
 
+    LEUnicode   fPreBaseConsonant;
+    LEUnicode   fPreBaseVirama;
+    le_int32    fPBCIndex;
+    FeatureMask fPBCFeatures;
 
     void saveMatra(LEUnicode matra, le_int32 matraIndex, IndicClassTable::CharClass matraClass)
@@ -172,5 +176,6 @@
           fMatraFeatures(0), fMPreOutIndex(-1), fMPreFixups(mpreFixups),
           fVMabove(0), fVMpost(0), fVMIndex(0), fVMFeatures(0),
-          fSMabove(0), fSMbelow(0), fSMIndex(0), fSMFeatures(0)
+          fSMabove(0), fSMbelow(0), fSMIndex(0), fSMFeatures(0),
+          fPreBaseConsonant(0), fPreBaseVirama(0), fPBCIndex(0), fPBCFeatures(0)
     {
         // nothing else to do...
@@ -191,4 +196,6 @@
         fVMabove = fVMpost  = 0;
         fSMabove = fSMbelow = 0;
+
+        fPreBaseConsonant = fPreBaseVirama = 0;
     }
 
@@ -386,4 +393,12 @@
     }
 
+    void notePreBaseConsonant(le_uint32 index,LEUnicode PBConsonant, LEUnicode PBVirama, FeatureMask features)
+    {
+        fPBCIndex = index;
+        fPreBaseConsonant = PBConsonant;
+        fPreBaseVirama = PBVirama;
+        fPBCFeatures = features;
+    }
+
     void noteBaseConsonant()
     {
@@ -465,4 +480,20 @@
     }
     
+    void writePreBaseConsonant()
+    {
+        // The TDIL spec says that consonant + virama + RRA should produce a rakar in Malayalam.  However,
+        // it seems that almost none of the fonts for Malayalam are set up to handle this.
+        // So, we're going to force the issue here by using the rakar as defined with RA in most fonts.
+
+        if (fPreBaseConsonant == 0x0d31) { // RRA
+            fPreBaseConsonant = 0x0d30; // RA
+        }
+        
+        if (fPreBaseConsonant != 0) {
+            writeChar(fPreBaseConsonant, fPBCIndex, fPBCFeatures);
+            writeChar(fPreBaseVirama,fPBCIndex-1,fPBCFeatures);
+        }
+    }
+
     le_int32 getOutputIndex()
     {
@@ -723,4 +754,5 @@
             }
 
+            
             IndicClassTable::CharClass charClass = CC_RESERVED;
             IndicClassTable::CharClass nextClass = CC_RESERVED;
@@ -730,7 +762,9 @@
             le_bool  seenVattu = FALSE;
             le_bool  seenBelowBaseForm = FALSE;
+            le_bool  seenPreBaseForm = FALSE;
             le_bool  hasNukta = FALSE;
             le_bool  hasBelowBaseForm = FALSE;
             le_bool  hasPostBaseForm = FALSE;
+            le_bool  hasPreBaseForm = FALSE;
 
             if (postBase < markStart && classTable->isNukta(chars[postBase])) {
@@ -746,12 +780,20 @@
                 hasBelowBaseForm = IndicClassTable::hasBelowBaseForm(charClass) && !hasNukta;
                 hasPostBaseForm  = IndicClassTable::hasPostBaseForm(charClass)  && !hasNukta;
+                hasPreBaseForm = IndicClassTable::hasPreBaseForm(charClass) && !hasNukta;
 
                 if (IndicClassTable::isConsonant(charClass)) {
                     if (postBaseLimit == 0 || seenVattu ||
                         (baseConsonant > baseLimit && !classTable->isVirama(chars[baseConsonant - 1])) ||
-                        !(hasBelowBaseForm || hasPostBaseForm)) {
+                        !(hasBelowBaseForm || hasPostBaseForm || hasPreBaseForm)) {
                         break;
                     }
 
+                    // Note any pre-base consonants
+                    if ( baseConsonant == lastConsonant && lastConsonant > 0 && 
+                         hasPreBaseForm && classTable->isVirama(chars[baseConsonant - 1])) {
+                        output.notePreBaseConsonant(lastConsonant,chars[lastConsonant],chars[lastConsonant-1],tagArray2);
+                        seenPreBaseForm = TRUE;
+   
+                    }
                     // consonants with nuktas are never vattus
                     seenVattu = IndicClassTable::isVattu(charClass) && !hasNukta;
@@ -786,10 +828,12 @@
 
             // write any pre-base consonants
+            output.writePreBaseConsonant();
+
             le_bool supressVattu = TRUE;
 
             for (i = baseLimit; i < baseConsonant; i += 1) {
                 LEUnicode ch = chars[i];
-                // Don't put 'blwf' on first consonant.
-                FeatureMask features = (i == baseLimit? tagArray2 : tagArray1);
+                // Don't put 'pstf' or 'blwf' on anything before the base consonant.
+                FeatureMask features = tagArray1 & ~( pstfFeatureMask | blwfFeatureMask );
 
                 charClass = classTable->getCharClass(ch);
@@ -842,5 +886,5 @@
 
             // write below-base consonants
-            if (baseConsonant != lastConsonant) {
+            if (baseConsonant != lastConsonant && !seenPreBaseForm) {
                 for (i = bcSpan + 1; i < postBase; i += 1) {
                     output.writeChar(chars[i], i, tagArray1);
@@ -872,5 +916,5 @@
             // write post-base consonants
             // FIXME: does this put the right tags on post-base consonants?
-            if (baseConsonant != lastConsonant) {
+            if (baseConsonant != lastConsonant && !seenPreBaseForm) {
                 if (postBase <= lastConsonant) {
                     for (i = postBase; i <= lastConsonant; i += 1) {