Fix: CRYSTAL_LOAD_DEBUG_INFO=1 fails with -Dexecution_context

[ysbaddaden]

Loading the debug info relies on File and goes through the event loop (even if reading files is blocking) which in turn requires a fiber scheduler (event loops are tied to a scheduler).

They used to be lazily instantiated, but with execution contexts it must be explicit. We thus have to load the debug info after creating the default execution context.

[HertzDevil]

On Windows apparently this is not early enough: ConsoleUtils.@@reader_thread creates an execution context before the default context is initialized, and then the runtime breaks because STDERR is not initialized either

[straight-shoota]

I presume we should still merge this, though?

[ysbaddaden]

I don't have any issue with CRYSTAL_LOAD_DEBUG_INFO=1 on Windows with this branch. I guess we can merge this fix.

I'll propose a few changes for Windows in a separate branch (no need for the isolated context + we should control when to start the thread).

EDIT: I eventually reproduced 🤦 I'll still push a fix as explained above

[crysbot]

Successfully created backport PR for release/1.16:

• [Backport release/1.16] Fix: CRYSTAL_LOAD_DEBUG_INFO=1 fails with -Dexecution_context #15715