James E Keenan via RT wrote: >Now, there's a lot of analysis packed into that commit message -- >but I'm not sure that all of it is reflected in the tests. Yes. The commit message is mostly a rearrangement of comments I made on IRC while we were discussing the issue, setting out the background for the behaviour. It's not well structured to provide context for the tests; in fact it's not well structured at all. Furthermore, the form of the test code suggests some misunderstanding of what I said. I contrasted "foo($x{bla})" against "\$x{bla}". "foo(\$x{bla})", which has both kinds of lvalueness (but not applied to the same subexpression) is not a coherent choice of test expression. I approve of adding tests for this part of the language, but we should do better than this patch. Better test code can be readily be constructed: there are a bunch of additional nearby cases that should be covered. The commit message doesn't need to explain the semantics at the level of detail that was attempted. -zeframThread Previous | Thread Next