STREL-966 Fix RNA-seq realignment with mismatches

RNA-Seq reads pinned to fixed start position by an intron-exon boundary,
and containing a leading insertion, were being incorrectly processed
when the reads also contained a mismatch, leading to an exception.
Adding the read start position to the alignment processing step should
fix the issue.

CHANGELOG.md

@@ -1,3 +1,9 @@
+## Unreleased
+
+### Fixed
+- Fix realignment error when processing certain RNA-seq inputs (STREL-966)
+  - Recent changes to read realignment/scoring introduced an edge case which fails on RNA-Seq reads from certain aligners - specifically when an insertion immediately follows an intron gap. The issue is now fixed.
+
 ## v2.9.5 - 2018-06-25
 
 This is a minor bugfix update from v2.9.4.

src/c++/lib/starling_common/starling_read_align.cpp

@@ -1884,7 +1884,10 @@ getCandidateAlignments(
         }
         if (isRecomputeRequired)
         {
-            cal = make_start_pos_alignment(cal.al.pos, 0, cal.al.is_fwd_strand, rseg.read_size(), validIndels);
+            // Convert the input alignment path to differentiate sequence match (=) and mismatch (X) from
+            // 'alignment match' (M). This is required for mismatch processing to work correctly.
+            const pos_t readStartPos(unalignedPrefixSize(cal.al.path));
+            cal = make_start_pos_alignment(cal.al.pos, readStartPos, cal.al.is_fwd_strand, rseg.read_size(), validIndels);
         }
     }