> Here is a patch. It moves warnings initialization after 
> io initialization (the latter being arguably more critical 
> than the former). Also adds two tests.

initstdio() and initmain() may emit warnings. Eg. issue #8766 (which now have a test in test_warnings).

It's possible to initialize warnings before initstdio() by rewriting more warnings functions in C (into _warnings). But I don't think that it would solve all issues: emit a warning may write text into stderr, and if stderr is not ready yet, it doesn't work anyway. So I'm ok to exchange initstdio() and warnings initialization: we may loose some warnings, but at least, the warnings machinery will be operational.

But Antoine, could you please add a comment to explain why warnings is initialized after initstdio()?