On Sun, 20 Nov 2022 16:49:16 GMT, Markus KARG <d...@openjdk.org> wrote:
>>> @AlanBateman WDYT? >> >> The long standing and undocumented behavior is unfortunate. I don't think >> the 1-arg constructor is fixable while still allowing for lazy behavior. So >> I think the only thing we can do is document the existing behavior that a >> null element in the enumeration will lead to read or close throwing NPE at a >> later time. For the 2-arg constructor then we need to be cautious. This is >> very old API and we have to assume the phenomenon that is Hyrum's Law. You >> may be right that nobody can reasonably depend on the current behavior but I >> think it would need a survey to inform the discussion. I can't tell if this >> PR was intended to rehabilitate SequenceInputStream or just a reaction to >> seeing that it uses Vector. If the latter then you could just change it to: >> >> var list = new ArrayList<InputStream>(); >> list.add(s1); >> list.add(s2); >> e = Collections.enumeration(list); >> peekNextStream(); >> >> and preserve the existing behavior. >> >> While SequenceInputStream may be legacy I guess it would be interesting to >> get some idea of usage or usages that use something better in libraries. >> Maybe there is a case for InputStream to have a concat method for example. > > Indeed my intention solely is to get rid of `Vector`, so I am fine with > keeping a low profile and being full backwards compatible. I assume SIS is > seldomly used, so opening a can of warms is not worth it (I think). I am > perfectly happy with your proposal, but for the sake of brevity I'd rather go > with a shorter way of doing the pretty same: > ```this(Collections.enumeration(Arrays.asList(s1, s2))``` > Are we all fine with that? Otherwise I'd switch this PR back to Alan`s > proposal. N.B.: Regarding usage numbers, I will do a quick poll on Twitter. ------------- PR: https://git.openjdk.org/jdk/pull/11249
