RCOCOA-2301 Remove the emulated InterprocessMutex File cache

[tgoyne]

This was added to reduce the number of file descriptors used back when we opened a separate SharedGroup on each thread. We now normally only ever open one DB per file per process at a time, so sharing files between them is no longer useful and it's been the source of a few bugs.

Fixes realm/realm-swift#8507 by just removing the code containing the assertion entirely.

[coveralls-official]

Pull Request Test Coverage Report for Build thomas.goyne_254

Details

• 19 of 19 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.02%) to 92.03%

---

Totals
Change from base Build 2176:	0.02%
Covered Lines:	244702
Relevant Lines:	265893

---

💛 - Coveralls

[kiburtse]

How would close affect this? Assertion suggests that the uid for the path has changed somehow. Locking of the static is pretty straight forward here, so something is very fishy. This refactoring is nice, but I feel like it's hiding probably the real problem. Please, add at least the same strong check into mapping where we also use get_file_uid. @ironage what do you think?

[tgoyne]

My hypothesis is that the assertion is failing due to the map having an expired shared_ptr, not due to the path changing. This code should be deleted regardless of if it's correct or not, so I don't really see a reason to try to figure out what exactly is broken.

[tgoyne]

I think the uses of get_unique_id() in the encryption code should probably go away too.

[ironage]

Removing all uses of get_unique_id() sounds like a good idea to me. Issues like #7162 have broken my trust in it and I wouldn't be surprised if there is some other corner case causing issues. Maybe we can just use the file path instead?

[tgoyne]

Almost everything does just use the file path and if you break things by deleting a file while it's open and creating a new file at the same path, well that's your problem to deal with. The encryption stuff originally used the inode for performance reasons and not correctness, but those global maps also have no reason to exist any more. They were global because that's the only way to get data to signal handlers, and thankfully we no longer use signal handlers.

[ironage]

Nice reduction in complexity. LGTM.