Fix issue #4

[23Skidoo]

This still doesn't completely free the users from the need to think, but should make it harder to get messages silently dropped.

Resolves #4.

[arybczak]

To be fair, if you want to get rid of finalizers then the optimal solution IMO is to rewrite functions that construct loggers to be alloca-like, e.g. withBulkStdoutLogger :: (Logger -> IO r) -> IO r, or withPgLogger :: ConnectionSourceM IO -> (Logger -> IO r) -> IO r. Then you can guarantee that waitForLogger is always run when Logger goes out of scope. This won't prevent anyone from using Logger that was already finalized by returning it outside the scope in which it's valid, but that's fine as you can do the same with alloca, so simply mentioning this in the documentation should be sufficient.

[23Skidoo]

@arybczak Thanks, that's a good idea, I'll implement it.

This won't prevent anyone from using Logger that was already finalized by returning it outside the scope in which it's valid

This can sort of be prevented by marking the logger as closed and raising an exception when it's used outside of scope. There will still be the issue of main thread exiting before the child threads, but that's just something clients need to be aware of.

[23Skidoo]

Implemented new logger creation API as suggested by @arybczak, but not logger invalidation (yet).

[23Skidoo]

Logger invalidation now implemented.

[jonathanjouty]

Followup with ElasticSearch logging

[jonathanjouty]

Followup with ElasticSearch logging

[jonathanjouty]

I've had a quick look and Travis CI is happy, so I'm not going to dig deeper right now.

@arybczak did you want to review changes? If not, @23Skidoo feel free to merge.

[23Skidoo]

I think I'll merge this now and address @arybczak's potential future comments later if/when he decides to review.