When our chainstate activation sequence fails and we're running the GUI, we ask the user via a question box whether or not they'd like to re-run the sequence with fReindex = true. Fun fact: despite the fact that this sequence is wrapped in a while loop, we only run it twice at most.

If they answer yes, the sequence will always fail, because a chainstate will have already been initialized in the original run and InitializeChainstate will throw a std::logic_error:

This always reproduces. To reproduce:

gdb ./src/qt/bitcoin-qt
(gdb) b InitializeChainstate
(gdb) # add breakpoint for the "if (!fLoaded && !ShutdownRequested()) {" line
(gdb) r  # should stop at the breakpoint for InitializeChainstate
(gdb) c  # to advance to the breakpoint for the "if (!fLoaded && !ShutdownRequested()) {" line
(gdb) set fLoaded=false
(gdb) c  # should stop at the breakpoint for InitializeChainstate
(gdb) n  # step through InitializeChainstate to see that it throws

We expect the re-run of the chainstate activation sequence to not always fail.