RC2 fixes

[goatpig]

• Fix blockxor file copy: DONE
• Windows build
• Implement txhint collision handling: DONE

[tomrobbo]

I built rc2_fixes + my three PRs to check they still work together. The DB loads, new blocks come in from misaligned offsets, and after a new block it only scans that block. Two things got in the way:

1. Startup died right after "Enabling zero-conf tracking" with an empty "BDM thread failed". gdb on the BDM thread:

   loadZeroConfMempool -> ParsedTx::setTx -> Tx::unserialize -> TxCalcLength (throw)
   

   One of the mempool txs saved by my previous build (5b98eca) doesn't parse on this one. --clear-mempool didn't help - the entries get parsed before the flag is checked, so it dies first. Moving the zeroconf db aside got it running. I've kept the file if you want it.

2. --rescan aborts straight away:

   BDM thread failed: unexpected db
   

   resetHistoryDatabases() asks for TXOUTS etc. before they've been opened.

One more thing I noticed: TxHintCollision is caught in the zc parser and broadcast paths but not in the scanner - the getDBKeyForHash call at BlockchainScanner.cpp:1062 has no try/catch. That's the path my first-scan crash (#788) came through, so a collision during a wallet scan would presumably still take the DB down. I couldn't re-test it with --rescan broken, short of a full rebuild.

[goatpig]

Startup died right after "Enabling zero-conf tracking" with an empty "BDM thread failed". gdb on the BDM thread:

It would help me if you still had the mempool file. I'd would like to watch the tx in the debugger. If you don't or can't provide the file, could you provide a full backtrace from a debug build instead?

--rescan aborts straight away:

It's likely trying to run the db as DB_BARE, which is current default. Try to give it an explicit --db-type=DB_FULL argument along with --rescan

he getDBKeyForHash call at BlockchainScanner.cpp:1062 has no try/catch.

This code section is entirely disabled until I figure out what I want to do with the tx filters code. It has been so since the rework that reintroduced DB_BARE.

[tomrobbo]

It would help me if you still had the mempool file. I'd would like to watch the tx in the debugger. If you don't or can't provide the file, could you provide a full backtrace from a debug build instead?

Mempool file (from ~/.armory/databases, written by the 5b98eca build): https://we.tl/t-738fezkzd2ZAQeXT

--rescan aborts straight away:

It's likely trying to run the db as DB_BARE, which is current default. Try to give it an explicit --db-type=DB_FULL argument along with --rescan

--rescan was with an explicit DB_FULL:

./ArmoryDB --db-type=DB_FULL --public --ram-usage=2 --rescan
...
dbmode: DB_FULL
Executing: doInitialSyncOnLoad
BDM thread failed: unexpected db

he getDBKeyForHash call at BlockchainScanner.cpp:1062 has no try/catch.

This code section is entirely disabled until I figure out what I want to do with the tx filters code. It has been so since the rework that reintroduced DB_BARE.

Understood

[goatpig]

latest commit tries to address rescan and zc load issues