Hi, I'm working on TAP5-2688, which is about the TypeCoercer picking the "wrong" CoercionTuple. Because JSONArray became a real Collection, its String->JSONArray CoercionTuple is picked over the Object->List CoercionTuple.
https://issues.apache.org/jira/browse/TAP5-2688 I've created a patch and added it to the ticket, but I wanted to have some feedback before committing and pushing. The patch contains: - Preferring exact targetType matches over cheaper (less intermediate) coercions. - Add an explicit String->Collection CoercionTuple in tapestry-json What's up to discussion IMO is the general question about cost vs. quality. Previously "cost" was favored over quality because no quality checks existed except the exact source and target type checks. The patch I've added to the ticket now prefers quality over cost. That means that the first coercion of a value will check all coercion tuples if a better match with an exact targetType is found. The only problem I see with this approach is costlier first checks (it's cached afterward), and maybe a more costly way is chosen even if a cheaper "good enough" CoercionTuple would suffice. Possible changes: - Restrict intermediate lookups if a "good enough" CoercionTuple is found. A one or two-step coercion should be preferable over a 5+ coercion. But what would be a good number? A maximum of one or two additional level checks sounds good to me. What do you think? Cheers, Ben
