Regression test doesn't cover flat-nodes

[pnorman]

The Liechtenstein extract is ill-suited for this, but we have no test coverage of flat nodes.

One potential issue specific to flat-nodes is that it's a file and it could have different behavior if there's a previous flat nodes file with the same name.

[alex85k]

And also the flat-nodes test will be very slow on most machines...

By the way, current default max ID is > 3000000000 and default reserved space is 1.5 times smaller.

[pnorman]

Flat nodes is not particularly slow, even on one of my more normal machines.

[apmon]

flat nodes is probably a case where you want to do unit testing rather than integration testing. That way you need to write much less of the file than you would even if you import just liechtenstein.

node-persistent-cache-reader.c has some (primitive) stand alone code to test/benchmark some aspects of the flat nodes sub system. One could presumably expand that to do some proper unit testing of the likely things that could (or have) gone wrong with the flat nodes system. You will still need ~20GB of space to hold the flat node file. But you presumably don't have to write to all of it, and so if you have a modern filesystem that can create a large (sparse) file quickly, tests shouldn't be too bad.

[pnorman]

Yes - we've added a unit test for the specific problem we found on our branch.