Catch errors in writing the package cache

[filmor]

For some odd reason, from time to time I get (persistent) eaccess errors in tab2file. I still have to investigate, where these come from, but I think the error should be non-fatal.

[ferd]

This seems reasonable. I have concerns that on the first build, this just causes other errors down the road rather than failing early.

I'd say that the warning is a very good addition. I'm not sure that aborting after it is not the best decision though. How does it behave when you use it and encounter these errors?

[filmor]

It's a really odd issue and I will have to recompile ets.erl with a bit of debug information to find out what is happening. I see that it has a fake eaccess that it produces whenever the initial deletion doesn't (seem to) go through here https://github.com/erlang/otp/blob/master/lib/stdlib/src/ets.erl#L822. Maybe it's a latency issue or something of that sort, my ~ is on a Windows share (nothing I can do about that ... :/)

[ferd]

Sounds good. Will merge. I figure it will work alright to keep functionality when an index is already there, and otherwise the warning will be good to figure out another failure.