3.10.2 changes for master

[hannesm]

This embeds the changes of 3.10.2 for master, namely #1218 and #1209. Superseeds #1216.

[hannesm]

The CI faliure shows that some generated main.ml now initialize the random device multiple times (there are multiple random devices) -- this is not good. This needs to be investigated and fixed before merge!

[TheLortex]

For information, the random devices involved are:

• in Mirage_impl_stack.direct_stackv4: by default it's rng ~time ~mclock ()
• in Mirage_impl_conduit and the test config: default_random

[hannesm]

Thanks for looking into this @TheLortex. My intuition is that we should use default_random (to avoid multiple RNG devices), as done in 4102b23.

There's still a question about flexibility and what the default should be, in case a user wants a custom "default time" / "default clock" / "default random", how would they pass it through -- I'm not sure where (and when) to discuss this. OTOH from a security perspective, since entropy is (with mirage-crypto-rng) only fed to a single RNG, we should ensure that Mirage only ever generates a single "real RNG".

[hannesm]

IMHO good to be merged.

[TheLortex]

CI-wise, this is looking good. The mirage-skeleton PR mirage/mirage-skeleton#299 fixes the remaining errors.

[hannesm]

@TheLortex great. Feel free to merge then. For mirage-skeleton, I used to rebase & force-push the dev branch (to have the few commits ahead of master clearly visible, and easily merge them to master one day mirage@master is out).

[dinosaure]

Thanks both!