[mycpp/runtime] Fix leak when using getline()

Python-2.7.13 configure now runs with no crashes or leaks!  Woohoo !!!

benchmarks/gc.sh

@@ -289,13 +289,22 @@ gc-run-oil() {
 
   local i=0
   for script in */*.sh; do
-    echo "=== $script"
+    case $script in
+      (build/clean.sh|build/common.sh|build/dev.sh)
+        # Top level does something!
+        echo "=== SKIP $script"
+        continue
+        ;;
+    esac
+
+    echo
+    echo "=== ($i) $script"
 
     # Just run the top level, which (hopefully) does nothing
     OIL_GC_STATS=1 OIL_GC_THRESHOLD=1000 OIL_GC_ON_EXIT=1 $bin $script
 
     i=$((i + 1))
-    if test $i -gt 3; then
+    if test $i -gt 60; then
       break
     fi
   done

mycpp/gc_mylib.cc

@@ -18,7 +18,7 @@ MutableStr* NewMutableStr(int cap) {
   return reinterpret_cast<MutableStr*>(NewStr(cap));
 }
 
-// NOTE: split_once() was in gc_mylib, and is likely not leaky
+// TODO: remove unnecessary rooting
 Tuple2<Str*, Str*> split_once(Str* s, Str* delim) {
   StackRoots _roots({&s, &delim});
 
@@ -72,6 +72,8 @@ Str* CFileLineReader::readline() {
   errno = 0;
   ssize_t len = getline(&line, &allocated_size, f_);
   if (len < 0) {
+    // man page says the buffer should be freed even if getline fails
+    free(line);
     if (errno != 0) {  // Unexpected error
       log("getline() error: %s", strerror(errno));
       throw Alloc<IOError>(errno);
@@ -80,9 +82,10 @@ Str* CFileLineReader::readline() {
     return kEmptyString;
   }
 
-  // TODO: Fix the leak here.
   // Note: getline() NUL terminates the buffer
-  return ::StrFromC(line, len);
+  Str* result = ::StrFromC(line, len);
+  free(line);
+  return result;
 }
 
 // Problem: most Str methods like index() and slice() COPY so they have a