Damus
Se7enZ · 3d
Can see missing the issue in review, especially wrt to the huge PR diff. But this should have been integration tested - both for entropy and proper library symbols. Dusty's article is great. Hindsig...
Sjors profile picture
Entropy is hard to test. You'd have to have to run it on (slow) real hardware billions of times to find a 32 bit collision. In a simulator you don't expect to use the real TRNG.

Checking library symbols is not a typical thing for an integration test, though it might have been good (in hindsight) to run a whitelist of symbols you want to see. Bitcoin Core does some of that:
https://github.com/bitcoin/bitcoin/blob/master/contrib/guix/symbol-check.py
1
Se7enZ · 3d
What about bitwise collision counting as a signal? You're right - I wouldn't have thought to check for symbols prior to seeing the fix. Hindsight.
Matt Corallo · 2d
All the more reason for defensive coding in RNGs. Sadly this isn’t the norm (does anyone do it?) but hopefully it can become it… nostr:nevent1qqstxve7raenw9yuacxftz0jtakmxtxpxa3lkst6ga8367lzgly33cq4q9vun