test validity of downloaded dict. Internally (in jni), add lots of
checks that format is correct and we don't run off end of ptr walking
truncated dict.
arrays into the jni, pass the full file paths in in addition to the
byte arrays. This isn't possible with the built-in dicts, but does
work for the downloaded ones (which are usually larger). This checkin
does the mmap and uses memcmp to verify that the bytes are the same as
passed in. Next step is to not pass the bytes when the path will do
and to actually use the mmap'd ptr.
Robots default to BasEnglish dict and humans to CollegeEng. Add new
per-game default for robot dict. Still need to deal with language
changes and non-English case in general.
object) rather than allocating a new array in the C heap -- for the
DAWG data of a dictionary. This can use up to 5% of the java heap for
huge dictionaries, but I'm hoping it fixes a problem reported by a
user of the large German dictionary that seems to involve allocation.
If I'm reading correctly, as long as I stay within 16M (24M or more on
newer devices) I'm sure to get my memory in the java world while it's
less a sure thing in the JNI world (where in addition linux's
aggressive overallocation is used, meaning I'll fail when I try to
swap in memory on write rather than get back NULL from malloc.)
number of words can be included. Changed to build dicts and linux to
open them. Android still needs to learn. Also, some of the tools in
dawg/ need to be fixed to read old-format (pre-utf8) .xwd files.
their games being drawn (no BoardActivity around) break the two
android-only callbacks out of UtilCtxt and into a new JNIUtils
interface that then requires new handing in C.
