core: Jim_SetVariable() now frees value on errormaster-next

If the value has a zero reference count, free it in the error path.
Otherwise callers can't do: Jim_SetVariable(..., Jim_New...())
in case the object is leaked in the error path.

Fix various callers that were previously freeing the object in the
error path, and add a test to loop.test to show that this was not.

Signed-off-by: Steve Bennett <steveb@workware.net.au>

1 files changed, 9 insertions, 0 deletions

diff --git a/tests/loop.test b/tests/loop.test
index c6144e0..43fe562 100644
--- a/tests/loop.test
+++ b/tests/loop.test
@@ -152,6 +152,15 @@ test loop-2.8 {modify loop var} {
     set a
 } {1 2 3 4 5}
 
+# Previously this would leak memory (configure --maintainer)
+test loop-2.9 {fail to set loop var} -body {
+    set i 1
+    loop i(x) 1 6 {
+        incr y
+    }
+    set y
+} -returnCodes error -result {can't set "i(x)": variable isn't array}
+
 testreport
 break