+1, looks good. Presumably this happened on reversal to megamorphic so setObject was linked and it didn’t properly handle this case… the crux of the matter, if I understand it correctly is that once the property is set, the NEEDS_DECLARATION flag needs to be removed, right?
I presume the fast-linked (non-megamorphic) paths handle this correctly already, otherwise we would’ve seen the bug on first iteration, not on 17th :-)
Post by Sundararajan Athijegannathan
Please review http://cr.openjdk.java.net/~sundar/8200215/webrev.00/