Remove -Xcheckinit from the default option set #85
Conversation
There was a problem hiding this comment.
Thanks. I do feel bad that this goes against Rob's wishes in #10 but I'm inclined to follow the compiler engineers' advice that this isn't intended for production :)
|
|
||
| ### Changed | ||
|
|
||
| * [#85](https://github.com/typelevel/sbt-tpolecat/pull/85) - `-Xcheckinit` was removed from the default option set. After discussion on issues [#10](https://github.com/typelevel/sbt-tpolecat/issues/10) and [#83](https://github.com/typelevel/sbt-tpolecat/issues/83) it has become apparent that although `-Xcheckinit` offers a lot of value for some users, it creates some very tricky problems for users that are writing async code. The initialization checking code this option produces introduces `@volatile` variables that can disguise memory visibility issues. |
There was a problem hiding this comment.
Technically, as long as you test and publish your code with -Xcheckinit then you won't have hidden memory visibility bugs, just performance issues. It's only a problem if you use it for testing, but then publish without it (removing all the extra memory barriers).
There was a problem hiding this comment.
Yeah that is true, I wonder how it affects things when (for example) a user specifies -Xcheckinit and has inlining enabled. Seems like there are a lot of possible interactions here depending upon the permutations of user vs library-specified options.
Resolves #83