Certain redo post-build failures would still mark a target as built.

If we failed because:
- target dir doesn't exist
- failed to copy from stdout
- failed to rename $3

We would correctly return error 209, but the target would still be
marked as having been built, so redo-ifchange would not try to build it
next time, beause we forgot to call sf.set_failed() in those cases.

minimal/do worked correctly.

Added a test to catch this in the future.

redo/builder.py

@@ -419,7 +419,8 @@ class _BuildJob(object):
                 sf.csum = None
                 sf.update_stamp()
                 sf.set_changed()
-        else:
+        # rv might have changed up above
+        if rv:
             helpers.unlink(self.tmpname)
             sf = self.sf
             sf.set_failed()

t/122-defaults-parent/.gitignore

@@ -0,0 +1 @@
+/*.log

t/122-defaults-parent/all.do

@@ -0,0 +1,22 @@
+rm -f x/shouldfail
+
+log=$PWD/$1.log
+rm -f "$log"
+
+expect_fail() {
+	local rv=$1
+	shift
+	if ("$@") >>$log 2>&1; then
+		cat "$log" >&2
+		echo "unexpected success:" "$@" >&2
+		return $rv
+	else
+		return 0
+	fi
+}
+
+cd inner
+expect_fail 11 redo ../x/shouldfail   # should fail
+expect_fail 12 redo-ifchange ../x/shouldfail  # should fail again
+
+exit 0

t/122-defaults-parent/clean.do

@@ -0,0 +1,2 @@
+rm -f *~ .*~ */*~ */.*~ *.tmp */*.tmp x/shouldfail *.log */*.log
+

t/122-defaults-parent/inner/default.do

@@ -0,0 +1,2 @@
+echo "inner/default.do: PWD=$PWD '$1' '$2' '$3'" >&2
+echo SUSPICIOUS