The tone difference is certainly stark! The bug report feels like a process is in play, while the mail has a tone of a rant in a closed room.

I often take it for granted the compiler is always right (and it darn well just about is), but bugs like this look exactly like mis-cast magic to me. Learning how the compiler makes decisions seems like a rabbit hole, but may be fascinating enough to be worth it. (And think about being a maintainer of such a complex beast!)

"Here's a small (auto-)reduced testcase for this specific issue"

Do you happen to know how that is done? How is a test case "auto"-reduced? I've never done that before.

For example http://embed.cs.utah.edu/creduce/ ; the slides from a presentation in 2012 are interesting: http://embed.cs.utah.edu/creduce/pldi12_talk.pdf

This is awesome, thanks mate!

If you find this topic interesting, look up everything written by http://blog.regehr.org/

Thank you!

For D we use DustMite (disclaimer: I'm the author). It also works to some extent with C/C++ (thanks to the languages' syntax being similar enough).

https://github.com/CyberShadow/DustMite/wiki

The LLVM project recommends delta:

http://delta.tigris.org/

> The bad compiler versions are 4.5.0 (when debug_insn came in) to 4.8.3 and 4.9.0 and 4.9.1.

I think it's a case of "bugs happen". They fixed this one pretty quickly.