Accept null dlhandle and support custom dlopen flags

[chapulina]

After digging through #30, I was able to get ign-sensors to work with only this change, without touching the local / global variables. It seems to me that this change is backwards compatible, because it only tries the nullptr after trying the usual method first. Update: this PR also allows changing setting to global.

We know no previous use cases were getting a null handle because we haven't run into that assert. So I think this is safe to go into ign-plugin1.

Unfortunately, I have no idea what leads us into this situation, so I'm not able to write a test case for it 😕

Let me know what you think, @ahcorde .

Here's the ign-sensors PR that uses this: gazebosim/gz-sensors#90

[codecov]

Codecov Report

Merging #35 (0317f4c) into ign-plugin1 (a80598b) will decrease coverage by 1.18%.
The diff coverage is 30.00%.

@@               Coverage Diff               @@
##           ign-plugin1      #35      +/-   ##
===============================================
- Coverage        99.82%   98.64%   -1.19%     
===============================================
  Files               15       15              
  Lines              584      592       +8     
===============================================
+ Hits               583      584       +1     
- Misses               1        8       +7     

Impacted Files	Coverage Δ
loader/src/Loader.cc	96.75% <30.00%> (-3.25%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a80598b...54e2981. Read the comment docs.

[ahcorde]

I made a cimment about this here gazebosim/gz-sensors#90 (comment)

[chapulina]

I made a cimment about this here

@ahcorde , I think the current version of this PR addresses those test failures. This PR adds support for RTLD_GLOBAL just like #30 does, with the advantage of:

• not requiring a version bump
• allowing the flexibility of not applying the change to all libraries
• allowing the flexibility of restoring the RTLD_LOCAL behaviour in the future if we want

Let me know what you think.

[chapulina]

I started writing some tests using RLTD_GLOBAL in 54e2981 and it's not looking good. Tests crash on shutdown, crash when loading libraries depending on the flag combination.

I'm afraid that we get this in, start using RTLD_GLOBAL, and start seeing runtime failures throughout the stack.

[chapulina]

Removing beta label, we won't have time to wrap this up before code freeze. Let's retarget at Ignition-F.

[chapulina]

Let's table this idea for now, tweaking the flags is probably not the best way to go about this.