Modify Travis to only run the linter on modified lines

.travis.yml

@@ -36,7 +36,7 @@ matrix:
       compiler: clang
       env: COMPILER=clang++
     - env: NAME="CPP-LINT"
-      script: DIFF=`git diff --name-only master HEAD` && if [ "$DIFF" != "" ]; then python scripts/cpplint.py $DIFF; fi
+      script: scripts/run_lint.sh master HEAD
 
 script:
   - make -C src minisat2-download

scripts/run_lint.sh

@@ -0,0 +1,90 @@
+#!/bin/bash
+
+set -e
+
+if [[ "$#" -ne 2 ]]
+then
+    echo "Script for running the CPP linter only on modified lines."
+    echo "Requires two arguments the start and the end"
+    echo "start - a git reference that marks the first commit whose changes to consider"
+    echo "end - a git reference that marks the last commit whose changes to consider"
+
+    exit 1
+fi
+
+git_start=$1
+git_end=$2
+
+# Get the list of files that have changed
+diff_files=`git diff --name-only $git_start $git_end`
+
+# Build a filter that will filter the blame output
+# to only include lines that come from one of the relevant_commits
+# We do this by making the blame tool output the same hash for all
+# lines that are too old.
+blame_grep_filter=`git rev-parse "$git_start"`
+
+# Build a regex for finding the line number of a given line inside blame
+# First matches the 40 digit hash of the commi
+# Then match an arbitary length number that represents the line in the original file
+# Finally matches (and groups) another arbitary length digit which is the
+# line in the final file
+regex="[0-9a-f]{40} [0-9]+ ([0-9]+)"
+
+# We only split on lines or otherwise the git  blame output is nonsense
+IFS=$'\n'
+
+are_errors=0
+
+for file in $diff_files; do
+  # We build another grep filter the output of the linting script
+  lint_grep_filter="^("
+
+  # Include line 0 errors (e.g. copyright)
+  lint_grep_filter+=$file
+  lint_grep_filter+=":0"
+
+  # We first filter only the lines that start with a commit hash
+  # Then we filter out the ones that come from the start commit
+  modified_lines=`git blame $git_start..$git_end --line-porcelain $file | grep -E "^[0-9a-f]{40}" | { grep -v "$blame_grep_filter" || true; }`
+
+  # For each modified line we find the line number
+  for line in $modified_lines; do
+
+    # Use the above regex to match the line number
+    if [[ $line =~ $regex ]]
+    then
+      # Some bash magic to get the first group from the regex (the line number)
+      LINENUM="${BASH_REMATCH[1]}"
+
+      # The format from the linting script is filepath:linenum: [error type]
+      # So we build the first bit to filter out relevant lines
+      LINE_FILTER=$file:$LINENUM
+
+      # Add the line filter on to the grep expression as we want
+      # lines that match any of the line filters
+      lint_grep_filter+="|"
+      lint_grep_filter+=$LINE_FILTER
+    fi
+  done
+
+  # Add the closing bracket
+  lint_grep_filter+=")"
+
+  # Run the linting script and filter by the filter we've build
+  # of all the modified lines
+  # The errors from the linter go to STDERR so must be redirected to STDOUT
+  result=`python scripts/cpplint.py $file 2>&1 | grep -E "$lint_grep_filter"`
+
+  # Providing some errors were relevant we print them out
+  if [ "$result" ]
+  then
+    are_errors=1
+    (>&2 echo "$result")
+  fi
+done
+
+unset IFS
+
+# Return an error code if errors are found
+exit $are_errors