[Issue 16590] Wrong di generation for ref methods

[via Digitalmars-d-bugs]

https://issues.dlang.org/show_bug.cgi?id=16590

--- Comment #11 from Chris Wright <dhasenan at gmail.com> ---
https://github.com/dlang/dmd/pull/6382 addresses by including function bodies.

The causes for the header and json issues are separate, and with json output,
you can at least construct the mangled version and demangle to get the
necessary information.

It's suboptimal to include function bodies for these functions in general. With
voldemort types, it's required; however, in all other cases, the compiler could
identify the actual type in use and report that. That can be handled by moving
header generation to follow semantic3 (currently it's just after parsing).
However, that would require filtering out certain symbols that are added during
semantic (for instance, __xdtor and a set of anonymous RTInfo!(type) fields).

--