Add feature for including openssl-sys

[Raniz85]

Added a new feature (openssl-static) that includes the openssl-sys crate
to fix static builds.

Renamed src/lib.rs to src/bindgen.rs and added a new lib.rs that
re-exports everything from bindgen.rs to prevent future bindgen runs
from reverting this change.

This fixes #25

[golddranks]

I think #25 was missing a reliable reproduction; here is one: https:/golddranks/pq_link_test

I built it against your fork of pq-sys, and indeed, adding the feature openssl-static fixes the build.

[Raniz85]

Any chance of merging this?

[weiznich]

@Raniz85 I got recently the rights to merge PR's in here, but I want to do a check of the current status of the crate first before merging anything. Unfortunately I haven't found the necessary time yet.

[weiznich]

While thinking a bit more about this PR: What's the expected use case for this? You mention fixing static builds, but this change on your own won't help you there much as you need to enable the corresponding feature anyway. Most people will likely use pq-sys through diesel that means that with this change you need to add pq-sys to your dependencies + enable the corresponding. feature. Without this feature you would need to add openssl-sys to your dependencies and be done. I must say that I don't really get what's then the advantage of having this feature here.

[Raniz85]

Without this feature you would need to add openssl-sys to your dependencies and be done.

It's been more than a year since I worked on this so my memory is hazy, but if I remember correctly, just adding a dependency on openssl-sys isn't enough, the dependency (or rather the extern crate openssl_sys) must come from pq-sys so that we can guarantee that openssl-sys is linked before pq-sys.

If the order is reversed, compilation will fail.

Read the linked issue and the issues referenced there, they contain quite a lot of information and debugging.

[golddranks]

I can confirm that there were certainly problems with the linking ordrer. What made them hard to deal with was that they were somewhat non-deterministic.

[Raniz85]

@weiznich Any progress on this? Either reject it or merge it please

[weiznich]

@Raniz85 I had no time to look into this yet, so there is no progress here. To be clear even to decide if this should be rejected or not I need time to understand the underlying problem. Unfortunately I cannot say when I will find the required time to look into this.

[weiznich]

Closed in favor of #46