GM self-tests: always run all tests (don't stop at first failure)

gm/tests/run.sh

Index: gm/tests/run.sh
===================================================================
--- gm/tests/run.sh	(revision 8878)
+++ gm/tests/run.sh	(working copy)
@@ -25,6 +25,8 @@
 OUTPUT_EXPECTED_SUBDIR=output-expected
 CONFIGS="--config 8888 565"
 
+ENCOUNTERED_ANY_ERRORS=0
+
 # Compare contents of all files within directories $1 and $2,
 # EXCEPT for any dotfiles.
 # If there are any differences, a description is written to stdout and
@@ -38,7 +40,7 @@
   diff -r --exclude=.* $1 $2
   if [ $? != 0 ]; then
     echo "failed in: compare_directories $1 $2"
-    exit 1
+    ENCOUNTERED_ANY_ERRORS=1

[borenet, 2013/04/26 16:04:09]

Alternatively, you could add error messages to this variable:
ERRORS="${ERRORS}\n${NEW_ERROR_MSG}"

And test if it's empty at the end.

[epoger, 2013/04/26 16:10:55]

The question is, what would I put into $ERRORS that would be a useful summary?

I suppose I could pipe (or tee) the output of diff to $ERRORS, but I'm not sure
whether that output would make sense without the context provided by the output
generated as it goes.

I think it's best to just tell the human it failed, and trust her to scroll up
and see what happened, for now.  Maybe we can improve that later...

[borenet, 2013/04/26 16:12:43]

Fair enough.

   fi
 }
 
@@ -171,4 +173,9 @@
 # Ignore some error types (including ExpectationsMismatch)
 gm_test "--ignoreErrorTypes ExpectationsMismatch NoGpuContext --verbose --hierarchy --match selftest1 $CONFIGS -r $GM_INPUTS/json/different-pixels.json" "$GM_OUTPUTS/ignore-expectations-mismatch"
 
-echo "All tests passed."
+if [ $ENCOUNTERED_ANY_ERRORS == 0 ]; then
+  echo "All tests passed."
+  exit 0
+else
+  exit 1
+fi