Yang, Bo

> the example you gave does not seem particularly likely. Additionally, by this definition, any change of a return type - even narrowing it - would be a breaking change....

> >we will have to define a stable ABI for distributing libraries, not just the source code > > This seems extremely unlikely, ever; bytecode will continue to change as...

I think "runtime backward compatibility" is more realistic than binary, because currently all inferred type parameters do not change the runtime behavior. Reified type parameters do change the runtime behavior....

> Yes, there are indeed ways the typechecker can emit errors where it previously did not. These callsites can be restored by explicitly typing out the previously inferred types. If...

Yes. The problem is not only about changes introduced by HHVM, but also changes introduced by third party libraries, for users who don't have a monorepo.

Where did you find the error log?

Since we are now creating the Apt repository for universal deb packages, now we have a chance to revisit the convention for package names, versions and [suites](https://wiki.debian.org/DebianRepository/Format#Suite). ## Goals &...

According to a private conversation with @fredemmott, the reason why the package names are different is to support installing both `hhvm` and `hhvm-nightly` on the same machine. Unfortunately the ability...

Note that the versioning in RPM 4.10.0 or higher also supports tilde: https://rpm.org/ticket/56

I encountered a CMake error when finding TBB, and created a PR to fix it in https://github.com/facebook/hhvm/pull/9069