@alexmiller: I’ve been thinking about building software to test new versions of Clojure against existing Clojure libraries to help detect breakages early. Is this something that would be useful?

Sure. I know Andy has something like this and there was in the past something like this inside Relevance before I was there

is this a bug? https://gist.github.com/mpenet/5231071c9f8217d13fd3

so... decomplecbot 😉

mpenet: https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/LazySeq.java#L34

looks like adding meta calls seq()

yup I noticed

this feels like a bug, right?

dunno

if not this should be documented

can't pass the same internal thunk to two different LazySeq instances, else it will be realized twice

(and lazy seq bodies are ^:once fns)

I suppose it can be rationalized...

it can be, but this is impl details imho, a lambda user would expect a new LazySeq instance with the metadata attached

I got a bug caused by this, and it took me a while to drill down to find the cause, and I don't think I'd be considered a casual user. as I said, either it should be changed or documented imho

I guess file a jira...my hunch is that it will be a doc clarification. I don't see how to implement it properly

*er change the impl to handle it

I thought this a bit and I think it has to be this way

Not sure where it could be doc'ed other than the metadata page

I guess this could be a special case where metadata is "set" and the same LazySeq instance returned, but this might be a bit hairy to implement, documenting it is what seems to be the least horrible here

that is also not how metadata is designed to work