patgen SIGSEGV on aarch64-linux

[Karl Berry]

    On RPi 2 the current values still crash.

As far as I understand it, the limitation is about available swap. It's
not necessarily related to any particular hardware or os. I expect
adding swap or zram will make it work.

However, since apparently it is hopeless, I've reduced the values back
to the original level (10/5 million for trie{,c}_size respectively).

Instead of hardwiring the huge array sizes, the better solution is
clearly start with a small value and provide a way (cmdline option
and/or config) to increase it at runtime.  Unfortunately there are half
a dozen or so arrays of triec_pointer involved, so writing this is not
the work of a moment.

Sorry Keno. Failing a patch for the above, you'll need to compile your
own patgen.
    
    annoying that the test suite is failing.

More than annoying, IMHO. We don't want the test suite to fail; that's a
much worse outcome than anything else, just leading to missing real
problems. --best, karl.