Randall S. Becker

Just a note from me. `time_t` is not `unsigned long` on some platforms. It is just long on NonStop, as an example. This should not really matter much in this...

Weren't we thinking of some mechanism to allow 32-bit builds to work post-2038? The default `time_t` will not allow for that.

> Yes, but the memory leak is not really a big issue, since it will all be statically reachable memory, because even if the provider is unloaded later, the shared...

[like] Randall Becker reacted to your message: ________________________________ From: Richard Levitte ***@***.***> Sent: Thursday, October 26, 2023 11:16:47 AM To: openssl/openssl ***@***.***> Cc: Randall Becker ***@***.***>; Comment ***@***.***> Subject: Re:...

> In addition to issues with dynamic libraries, `atexit` is hazardous in multi-threaded applications. `atexit` runs when the main thread returns from `main`, but there's no guarantee that other threads...

> > That's what #17483 tries to do, in a cooperative manner, and also to allow multiple "init" from the same component (providers typically don't know how many times their...

I have issues with continuing to use `atexit()` which has been problematic and the reason for this PR. This will cause further breakages.

> Kurt Roeckx wrote: > I think the cleanup function was introduced at the same time as the atexit, and there were 1 or more other cleanup functions before that...

> > The use of `atexit` changed at 3.3 with new test introductions and changed existing behaviour of OpenSSL regardless of what was there "in the beginning". This broke some...

> > Another question is how to implement a key generation for keys stored in a token but that is IMO a different topic. > > Indeed storing generate keys...