A global library init function?

[bagder]

(Discussion forked off a sub-thread in #261)

I think there's case for adding global library init and free functions to do global (single) initializations and then the corresponding cleanups.

OpenSSL has an initialization that's only necessary to do once per process, not once per context as we do now. It is even discouraged to get called without using the OpenSSL mutex locks when using threads, so it is also a reliability concern.

[mcmanus]

this can be done quietly for the user though, right? through init_once and ref counting?

…

On Mon, Jan 23, 2017 at 11:59 PM, Daniel Stenberg ***@***.***> wrote: (Discussion forked off a sub-thread <#261 (comment)> in #261 <#261>) I think there's case for adding global library init and free functions to do global (single) initializations and then the corresponding cleanups. OpenSSL has an initialization that's only necessary to do once per process, not once per context as we do now. It is even discouraged to get called without using the OpenSSL mutex locks <https://wiki.openssl.org/index.php/Library_Initialization> when using threads, so it is also a reliability concern. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#266>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAP5s4cq0tcO8rgHpzkmuNVz5sOBlEFiks5rVMBBgaJpZM4LrI3F> .

[bagder]

It can only be done hidden in a safe manner if protected by a mutex though, and to protect with a mutex we'd need to add a pthread (or equivalent) requirement...

[weinrank]

Mh, is it useful to do this quietly?

I think we should involve the user for these steps.
For example when initializing the usrsctp library, it is possible to set a UDP encapsulation port.
IMHO this should the configurable by the user.

[mcmanus]

thread lib seems like an ok requirement to me in 2017.. if you really don't want it you can build an atomic spinlock for arm and x86 easily enough for this 1 case and have an architectural dep instead.

…

On Tue, Jan 24, 2017 at 8:06 AM, Daniel Stenberg ***@***.***> wrote: It can only be done hidden in a safe manner if protected by a mutex though, and to protect with a mutex we'd need to add a pthread (or equivalent) requirement... — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#266 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAP5s4lDT2y7nGmFZ7To0gDELW1dgPdIks5rVTJ-gaJpZM4LrI3F> .

[mcmanus]

the problem with explicit global init is that it doesn't necessarily embed well into other libraries.. the library can't really tell if it has been init'd already/concurrently what not..

…

On Tue, Jan 24, 2017 at 8:20 AM, Felix Weinrank ***@***.***> wrote: Mh, is it useful to do this quietly? I think we should involve the user for these steps. For example when initializing the usrsctp library, it is possible to set a UDP encapsulation port. IMHO this should the configurable by the user. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#266 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAP5s9iWA-YgNT-J8XqiwvyJL2-su3eNks5rVTXSgaJpZM4LrI3F> .

[bagder]

It should also be noted that modern OpenSSL (>= 1.1.0) already depends on pthreads itself, so it isn't that far fetched I think.

[bagder]

If we have it mutex protected, we can have it done silently if not already done explicitly by the user. So we can in fact have it both ways. If we want to.