Last night, after coming home from a party, I turned on my computer and Deadwood crashed (alas, without a stack trace or any useful info) during Windows start up. I tried to reproduce the bug last night and this morning without success; I set up some stress tests and ran Deadwood both with debugging symbols turned on, turned off, with Valgrind, etc. Deadwood looks good in Valgrind (no errors with compiled with -DVALGRIND_NOERRORS and no leaks), and I have not been able to reproduce the crash.
I'm thinking at this point that maybe I used an older version of Deadwood without the "don't reference a null pointer" fix from yesterday, even though the crashed program had a recent timestamp. I've done some hardening of all of the new compression code, have been completely unable to reproduce this crash, so I made sure to recompile Deadwood to use the latest version with the crash fix (and I can verify the crash fix works when stress-testing it), and will run this version of Deadwood for about a day and a half. If I don't get any crashes again by tomorrow around noon, I will make the HEAD snapshot Deadwood 2.4.04 in the early afternoon and see if the one crash I saw last night doesn't pop up again.