On Fri, Jul 11, 2025 at 12:09:48AM +0200, Étienne Mollier wrote: > 0. if I trust Helmut Grohne, I cannot empty the entropy pool; > maybe I can just make it more strained, so the behavior may > be unreproducible to some extent: for instance I have not > managed to reproduce the behavior straight on my laptop;
I would be quite surprised if the entropy pool was the problem. I remember the time when several packages (mostly cryptography-related) failed their test suites because of this. I found haveged to be the perfect workaround for that, and I always installed it in all my autobuilders, but I stopped using it two years ago when the kernel had a new implementation of /dev/random making haveged not necessary anymore (maybe this is what Helmut might have told you about the entropy pool). Moreover, as a test, I've built scilab 20 times with haveged installed, and the failure rate is more or less the same. I think the entropy pool is unlikely to be the problem. (But of course I'm glad that you are considering all possible scenarios). I guess that at some point we should probably tell upstream about this. Thanks.
