RAM-Reduction & Backup Center Testing (version 0.89.99.16)

[cp1]

When starting it for the second time, after downloading what blocks it had to, it got stuck scanning on 1%. I found this in the log over and over, for a very long time.

-WARN  - 0: (..\BlockUtils.cpp:4865) Marking orphan chain
-WARN  - 0: (..\BlockUtils.cpp:4888) Done marking orphan chain

I just added an empty wallet while it was loading and got this loop.  I closed and restarted and it's scanning transaction history again, says 20 minutes remaining.

[1715296597]

1715296597

[etotheipi]

When starting it for the second time, after downloading what blocks it had to, it got stuck scanning on 1%. I found this in the log over and over, for a very long time.

-WARN  - 0: (..\BlockUtils.cpp:4865) Marking orphan chain
-WARN  - 0: (..\BlockUtils.cpp:4888) Done marking orphan chain

I just added an empty wallet while it was loading and got this loop.  I closed and restarted and it's scanning transaction history again, says 20 minutes remaining.

Ugh, that's been happening too much.  Typically a restart solves it.  Before I officially release this version, I'm going to figure out what's causing that and fix it.  But it will not finish loading in that state.  It thinks that every block is an orphan.  It has to be restarted.

[cp1]

I haven't seen that loop since I started running bitcoin-qt independent of armory.  But I've only open and closed it a few times, so it could just be a coincidence.

I know you're busy, but I have (hopefully) a quick question -- I'm trying to add an address to a wallet following the add 500k address in extras, but only adding 1 key (I took out the key reading loop and just used line = private key).  I'm not doing it right because armoryqt.py gives me a checksum error on keys that were already in the wallet: line 3245, in unserialize '('+hash160_to_addrStr(self.addrStr20)+')' armoryengine.UnserializeError: Checksum mismatch in PrivateKey (1LMsGerE3FmWmJig6e5DpmjctFar6RVR2L)

Code:

import os
import sys
sys.path.append('..')
from armoryengine import *


# Could use sys.argv but this script will be used, like, once.  Hardcode it!
wltID        = 'ydeMUhu' # this was the ID of the wallet I tested with
wltDir       = 'ydeMUhu' # this was the ID of the wallet I tested with
wltfile    = os.path.join(ARMORY_HOME_DIR, 'armory_%s_.wallet' % wltID)
wltfilebak = os.path.join(ARMORY_HOME_DIR, 'armory_%s_backup.wallet' % wltID)


if not os.path.exists(wltfile):
   print 'ERROR: Wallet does not exist:', wltfile
   exit(1)

# Remove the backup if it exists
if os.path.exists(wltfilebak):
   os.remove(wltfilebak)




# If you don't delete the backup, Armory will think the primary wallet 
# is corrupted and restore the backup

exampleEntry = hex_to_binary( \
  '0047b8ad 0b1d6803 260ce428 d9e09e2c d99fd3b3 5947b8ad 0b1d6803 260ce428 '
  'd9e09e2c d99fd3b3 59fb1670 0860fecd 00030000 00000000 00ffffff ffffffff '
  'ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff ff71ca50 49feffff '
  'ffffffff ffffffff ffffffff ff000000 00000000 00000000 00000000 005df6e0 '
  'e2eeeeee eeeeeeee eeeeeeee eeeeeeee eeeeeeee eeeeeeee eeeeeeee eeeeeeee '
  'eec93a79 dd04a706 ad8f7311 5f905002 66f273f7 571df942 9a4cfb4b bfbcd825 '
  '227202da bad1ba3d 35c73aec 698af852 b327ba1c 24e11758 936bb632 2fe93d74 '
  '69b182f6 6631727c 7072ffff ffff0000 00000000 00000000 0000ffff ffff0000 '
  '0000 '.replace(' ',''))

wltOut = open(wltfile, 'ab')

rawAddrEntry = exampleEntry[21:]
addr20 = rawAddrEntry[:24]
fixed1 = rawAddrEntry[ 24:108]
prvkey = rawAddrEntry[    108:144]
pubkey = rawAddrEntry[        144:213]
fixed2 = rawAddrEntry[            213:]

addrDataToWrite = []

line = "5KaPWnaXTLUdv5WdbwbiXbjDqUFXEUT1aAgSGrYHtzuGD5mooWv"

privBin = base58_to_binary(line[1:-4])
pubBin  = CryptoECDSA().ComputePublicKey(SecureBinaryData(privBin)).toBinStr()
addr20  = hash160(pubBin)
# Pre-PyBtcAddr Entry Header
addrDataToWrite.append('\x00')
addrDataToWrite.append(addr20)

# PyBtcAddr itself
addrDataToWrite.append(addr20)
addrDataToWrite.append(computeChecksum(addr20))

addrDataToWrite.append(fixed1)

addrDataToWrite.append(privBin)
addrDataToWrite.append(computeChecksum(privBin))

addrDataToWrite.append(pubBin)
addrDataToWrite.append(computeChecksum(pubBin))

addrDataToWrite.append(fixed2)

wltOut.write(''.join(addrDataToWrite))
wltOut.close()

Any ideas?

[etotheipi]

I haven't seen that loop since I started running bitcoin-qt independent of armory.  But I've only open and closed it a few times, so it could just be a coincidence.

It really does only happen like once every 10 loads.  And then goes away.  Which will make it tough for me to track it down.  So it's more "expected" than "coincidence"...

I know you're busy, but I have (hopefully) a quick question -- I'm trying to add an address to a wallet following the add 500k address in extras, but only adding 1 key ...

Code:

...

Any ideas?

You just made me realize I should probably remove that example script... that script is actually "special" because someone had wanted to try running Armory with 500k addresses, and I said that it would be ridiculously slow to import them "the right way" (shown below).    So I recommended he manually modify the wallet files with the proper wallet format -- you should not import addresses that way!  

Instead, use the built-in function, and leave everything blank except for the privKey argument:

Code:

def importExternalAddressData(self, privKey=None, privChk=None, \
                                    pubKey=None,  pubChk=None, \
                                    addr20=None,  addrChk=None, \
                                    firstTime=UINT32_MAX,  firstBlk=UINT32_MAX, \
                                    lastTime=0,   lastBlk=0):

I haven't tested the following code, but it would look something like this (note that key data needs to be crammed into a "SecureBinaryData" object before being passed to any of these methods -- but still pretty simple)

# If run from extras dir, need to add parent dir to path
import sys
sys.path.append('..')
from armoryengine import *

# Create the SBD object for the private key which is "aaaaaaaa..."
hardcodedPrivKey = SecureBinaryData(hex_to_binary('aa'*32))

# Read the wallet file (it handles all the backup stuff automatically)
wlt = PyBtcWallet().readWalletFile('/path/to/my.wallet')

# Import the address using the built-in method
wlt.importExternalAddressData(privKey=hardcodedPrivKey)

# Nothing else needs to be done -- the wallet is closed after every operation already
print 'Done!'

# But let's verify it worked:
addr160 = convertKeyDataToAddress(hardcodedPrivKey)
addrStr = hash160_to_addrStr(addr160)
print 'The new address is', addrStr, 'and is in the wallet:', wlt.hasAddr(addr160)

Yeah, it's dead simple when you're using the right calls... sorry for the confusion!  

Edit updated the script with formatting and with the hex_to_binary() call which is needed if you are plugging in a hex private key.  Also, you may have damaged your previous wallet file trying the original script.  You might consider removing it and trying the new script on the original to make sure you're not mixing two different things.

[cp1]

Thanks, that worked great for a 32-byte hex key.  Is it possible to import from WIF?  I've tried:

wlt.importExternalAddressData(privKey=SecureBinaryData(base58_to_binary("5KaPWnaXTLUdv5WdbwbiXbjDqUFXEUT1aAgSGrYHtzuGD5mooWv")))

But I get: assert(plainPrivKey.getSize()==32)

Do I have to convert from WIF to 32-byte first?  Is there a function in armory?  I was snooping in bulkImportAddresses() and found snippets to deal with base-58 private key, but couldn't turn it into anything working.


Edit:  It must be in there somewhere because I think you can import WIF from armoryqt.  Now I'm on a hunt!

[etotheipi]

Thanks, that worked great for a 32-byte hex key.  Is it possible to import from WIF?  I've tried:

wlt.importExternalAddressData(privKey=SecureBinaryData(base58_to_binary("5KaPWnaXTLUdv5WdbwbiXbjDqUFXEUT1aAgSGrYHtzuGD5mooWv")))

But I get: assert(plainPrivKey.getSize()==32)

Do I have to convert from WIF to 32-byte first?  Is there a function in armory?  I was snooping in bulkImportAddresses() and found snippets to deal with base-58 private key, but couldn't turn it into anything working.


Edit:  It must be in there somewhere because I think you can import WIF from armoryqt.  Now I'm on a hunt!

The "parsePrivateKeyData()" method autodetects basically every private key format, including Casascius' mini-private-keys.  You could manually read the keys using "base58_to_binary()", but the following will work just as well to convert the data into 32-byte hex that can imported per my previous example:

Code:

>>> encodePrivKeyBase58('\xaa'*32)
'5K7T2qR7K5MUMDmkJvCEPbUeALuPyc6LFEnBiwWLuKCEVdBp8qV'
>>> parsePrivateKeyData(_)
('\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa\xaa', 
'Standard Base58 key with checksum')

Edit:  I just noticed you that your problem was actually that you crammed the whole 37 bytes into the import function.  The first byte is 0x80 which is the "network private key byte", and the last four bytes are a checksum: So you could've just done "base58_to_binary(...)[1:-4]" and that would've worked, too.

[cp1]

Success!  Thanks

parsePrivateKeyData worked great.  I still can't figure out base58_to_binary though (not a big deal).  I did try substringing at first using [1:-4], but it never worked for me, I always get too many bytes (and also the wrong answer).

For example:

line = "5JdeQ39z8NUkNVvB37tt74Cu2WSNVj7qb9PdY651UoQnqyCm937"
hardcodedPrivKey = base58_to_binary(line[1:-4])
print prettyHex(binary_to_hex(hardcodedPrivKey))

0x0000:  0d728779 7539eb09 87a0c18f 9e3fac6c 68dbe7f9 1eb106e5 0793d7b9 6e03697f
0x0020:  f673


But parsePrivateKeyData works perfect:

line = "5JdeQ39z8NUkNVvB37tt74Cu2WSNVj7qb9PdY651UoQnqyCm937"
pk = (parsePrivateKeyData(line)[0])
print prettyHex(binary_to_hex(pk))

0x0000:  6b88c087 247aa2f0 7ee1c595 6b8e1a9f 4c7f892a 70e324f1 bb3d161e 05ca107b

Ha, and funny enough it just loaded the transactions for this key in armoryqt and there was a small transaction in May.  People use the worst brainwallets.

[picobit]

Hi etotheipi,

I have a strange crash when Armory quits on Mac OS.  LevelDB gives a segfault, the stack trace from gdb is below.

I am by the way almost done making a Mac OS X App.  But since this is not a race for a bounty, I prefer to sit on it for a few days to fix the worst issues    (No real testing yet, will it run on another machine?  It probably crashes on old hardware if built on new hardware.  The icon and other stuff is missing.  The app is beyond huge, 250M !!).  I think I know how to address all of them.

Anyway, here is the crash data.  It occurs when I quit the app, and it does not depend on whether I build a real app, or just compile Armory and run it from the command line.

Code:

-INFO  - 1380265090: (BlockUtils.cpp:3988) Saving wallet history for next load

Program received signal EXC_BAD_ACCESS, Could not access memory.
Reason: KERN_INVALID_ADDRESS at address: 0x0000000000000000
[Switching to process 899 thread 0x1903]
0x0000000103e083c0 in InterfaceToLDB::commitBatch ()
(gdb) where
#0  0x0000000103e083c0 in InterfaceToLDB::commitBatch ()
#1  0x0000000103e5d064 in BlockDataManager_LevelDB::shutdownSaveScrAddrHistories ()
#2  0x00000001040250fe in _wrap_BlockDataManager_LevelDB_shutdownSaveScrAddrHistories ()
#3  0x000000010001d5a9 in PyEval_EvalFrameEx ()
#4  0x0000000100021869 in _PyEval_SliceIndex ()
#5  0x000000010001d63a in PyEval_EvalFrameEx ()
#6  0x0000000100021869 in _PyEval_SliceIndex ()
#7  0x000000010001d63a in PyEval_EvalFrameEx ()
#8  0x0000000100021869 in _PyEval_SliceIndex ()
#9  0x000000010001d63a in PyEval_EvalFrameEx ()
#10 0x0000000100021869 in _PyEval_SliceIndex ()
#11 0x000000010001d63a in PyEval_EvalFrameEx ()
#12 0x000000010001b147 in PyEval_EvalCodeEx ()
#13 0x0000000100054d7a in PyFunction_SetClosure ()
#14 0x00000001000136c6 in PyObject_Call ()
#15 0x00000001000309bf in PyMethod_New ()
#16 0x00000001000136c6 in PyObject_Call ()
#17 0x0000000100021018 in PyEval_CallObjectWithKeywords ()
#18 0x000000010007d7f6 in initthread ()
#19 0x00007fff96982772 in _pthread_start ()
#20 0x00007fff9696f1a1 in thread_start ()
(gdb) list
1	// algparam.cpp - written and placed in the public domain by Wei Dai
2	
3	#include "pch.h"
4	
5	#ifndef CRYPTOPP_IMPORTS
6	
7	#include "algparam.h"
8	
9	NAMESPACE_BEGIN(CryptoPP)
10	
(gdb) up
#1  0x0000000103e5d064 in BlockDataManager_LevelDB::shutdownSaveScrAddrHistories () at basic_string.h:279
279	      _M_data() const
Current language:  auto; currently c++
(gdb) list 
274	    private:
275	      // Data Members (private):
276	      mutable _Alloc_hider	_M_dataplus;
277	
278	      _CharT*
279	      _M_data() const
280	      { return  _M_dataplus._M_p; }
281	
282	      _CharT*
283	      _M_data(_CharT* __p)
(gdb) up
#2  0x00000001040250fe in _wrap_BlockDataManager_LevelDB_shutdownSaveScrAddrHistories ()
(gdb) list 
284	      { return (_M_dataplus._M_p = __p); }
285	
286	      _Rep*
287	      _M_rep() const
288	      { return &((reinterpret_cast<_Rep*> (_M_data()))[-1]); }
289	
290	      // For the internal use we have functions similar to `begin'/`end'
291	      // but they do not call _M_leak.
292	      iterator
293	      _M_ibegin() const
(gdb) up
#3  0x000000010001d5a9 in PyEval_EvalFrameEx ()
(gdb) list
294	      { return iterator(_M_data()); }
295	
296	      iterator
297	      _M_iend() const
298	      { return iterator(_M_data() + this->size()); }
299	
300	      void
301	      _M_leak()    // for use in begin() & non-const op[]
302	      {
303		if (!_M_rep()->_M_is_leaked())

[picobit]

I did a bit more debugging, putting a breakpoint in the crashing function and stepping line by line.  It looks like the call to KEY_NOT_IN_MAP triggers an exception in some python code.  It is a testnet wallet, so I could in principle send it to you.

Code:

Breakpoint 1, BlockDataManager_LevelDB::shutdownSaveScrAddrHistories (this=0x104ba23e0) at BlockUtils.cpp:3988
3988	   LOGINFO << "Saving wallet history for next load";
(gdb) n
Current language:  auto; currently c++
-INFO  - 1380282523: (BlockUtils.cpp:3988) Saving wallet history for next load
3990	   iface_->startBatch(BLKDATA);
(gdb) n
3992	   uint32_t i=0;
(gdb) n
3993	   set<BtcWallet*>::iterator wltIter;
(gdb) 
3994	   for(wltIter  = registeredWallets_.begin();
(gdb) 
3995	       wltIter != registeredWallets_.end();
(gdb) 
3998	      for(uint32_t a=0; a<(*wltIter)->getNumScrAddr(); a++)
(gdb) 
4000	         ScrAddrObj & scrAddr = (*wltIter)->getScrAddrObjByIndex(a);
(gdb) 
4001	         BinaryData uniqKey = scrAddr.getScrAddr();
(gdb) 
4003	         if(KEY_NOT_IN_MAP(uniqKey, registeredScrAddrMap_))
(gdb) 
(ERROR) armoryengine.py:12189 - Waiting for BDM output that didn't come after 20s.
(ERROR) armoryengine.py:12190 - BDM state is currently: Uninitialized
(ERROR) armoryengine.py:12191 - Called from: armoryengine.py:12314 (83837112)
(ERROR) armoryengine.py:12192 - BDM currently doing: Shutdown (83837112)
(ERROR) armoryengine.py:12193 - Direct traceback
  File "ArmoryQt.py", line 4981, in <module>
    os._exit(QAPP.exec_())
  File "ArmoryQt.py", line 4835, in closeForReal
    TheBDM.execCleanShutdown(wait=True)
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12314, in execCleanShutdown
    return self.waitForOutputIfNecessary(expectOutput, rndID)
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12194, in waitForOutputIfNecessary
    traceback.print_stack()
(ERROR) armoryengine.py:12195 - Traceback:
Traceback (most recent call last):
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12185, in waitForOutputIfNecessary
    return self.outputQueue.get(True, self.mtWaitSec)
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/Queue.py", line 176, in get
    raise Empty
Empty
4009	         RegisteredScrAddr & rsa = registeredScrAddrMap_[uniqKey];

[TitanBTC]

Just wanted to report that this update is definitely Usable on our windows7 64-bit machine.  Typically we use ubuntu, but I've wanted to try out some watch only copies on windows.  While it does seem to hang regularly when closing the program, its been stable enough to test functionality and everything we typically use works great.  

Current RAM usage on this version is hovering around 315 Mb, which is at least an order of magnitude improvement.  I'll continue troubleshooting to provide more helpful feedback but for now I just wanted to say THANKS for this much needed fix and good luck in your troubleshooting.

[etotheipi]

I did a bit more debugging, putting a breakpoint in the crashing function and stepping line by line.  It looks like the call to KEY_NOT_IN_MAP triggers an exception in some python code.  It is a testnet wallet, so I could in principle send it to you.

Code:

Breakpoint 1, BlockDataManager_LevelDB::shutdownSaveScrAddrHistories (this=0x104ba23e0) at BlockUtils.cpp:3988
3988	   LOGINFO << "Saving wallet history for next load";
(gdb) n
Current language:  auto; currently c++
-INFO  - 1380282523: (BlockUtils.cpp:3988) Saving wallet history for next load
3990	   iface_->startBatch(BLKDATA);
(gdb) n
3992	   uint32_t i=0;
(gdb) n
3993	   set<BtcWallet*>::iterator wltIter;
(gdb) 
3994	   for(wltIter  = registeredWallets_.begin();
(gdb) 
3995	       wltIter != registeredWallets_.end();
(gdb) 
3998	      for(uint32_t a=0; a<(*wltIter)->getNumScrAddr(); a++)
(gdb) 
4000	         ScrAddrObj & scrAddr = (*wltIter)->getScrAddrObjByIndex(a);
(gdb) 
4001	         BinaryData uniqKey = scrAddr.getScrAddr();
(gdb) 
4003	         if(KEY_NOT_IN_MAP(uniqKey, registeredScrAddrMap_))
(gdb) 
(ERROR) armoryengine.py:12189 - Waiting for BDM output that didn't come after 20s.
(ERROR) armoryengine.py:12190 - BDM state is currently: Uninitialized
(ERROR) armoryengine.py:12191 - Called from: armoryengine.py:12314 (83837112)
(ERROR) armoryengine.py:12192 - BDM currently doing: Shutdown (83837112)
(ERROR) armoryengine.py:12193 - Direct traceback
  File "ArmoryQt.py", line 4981, in <module>
    os._exit(QAPP.exec_())
  File "ArmoryQt.py", line 4835, in closeForReal
    TheBDM.execCleanShutdown(wait=True)
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12314, in execCleanShutdown
    return self.waitForOutputIfNecessary(expectOutput, rndID)
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12194, in waitForOutputIfNecessary
    traceback.print_stack()
(ERROR) armoryengine.py:12195 - Traceback:
Traceback (most recent call last):
  File "/Users/schiotz/src/BitcoinArmory/armoryengine.py", line 12185, in waitForOutputIfNecessary
    return self.outputQueue.get(True, self.mtWaitSec)
  File "/System/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/Queue.py", line 176, in get
    raise Empty
Empty
4009	         RegisteredScrAddr & rsa = registeredScrAddrMap_[uniqKey];

Actually, that's a red herring.  "KEY_NOT_IN_MAP" is actually a macro that is pretty safe from seg faults.  The problem was you waiting 10+ seconds between calling "Shutdown" and letting it finish shutting down.  At least that's why you got that error message from python -- It's difficult for the app to determine whether the blockdatamanager is stuck in an infinite loop or not, and starts throwing errors if any wait=True operation takes more than 10s.   Shutting down should not take long.  

The registeredScrAddrMap_[uniqKey] is much more likely to throw the seg fault, but I call "KEY_NOT_IN_MAP()" to make sure the key actually exists before attempting to access it like that.  So again, that should be fairly safe.  Unless the uniqKey object had destroyed/cleared and it was attempting to access the data behind it...

I'm going to go on a hunt for this bug this weekend.  It seems frequent enough I should be able to catch it if I restart the app enough times.  Unfortunately, I can't get the project compiling in debug mode in Windows, where debugging would be quite a bit easier...  I'll figure it out, either way.

By the way, goatpig came through and got me a full Windows port of the latest LevelDB for MSVS.  He's pretty badass with this obscure windows stuff!  I have a bit of my own testing to do on the new code, but it looks like it's building the DB about the same speed as Linux now -- in under an hour.  If that's true, then it will rescan in 10-15 min, too.  

Glad to see that the app is mostly working for people.  My wedding and then honeymoon starts in exactly a week, so I will be getting these bugs cleaned up and do a release towards the end of the week.  So far everything seems to be smooth, and any lingering usability issues are far preferable to the old usability issues (as long as the integrity is there, which it is).

[etotheipi]

P.S -- Has no one played with the new backup center?  Come on!  It's so badass!

[SimonBelmond]

P.S -- Has no one played with the new backup center?  Come on!  It's so badass!

One back from the conference I'll spend some time on it. Have some other things to to do also but I am also excited about the new features.

[picobit]

OK, so tracing the bug is difficult if one cannot single-step through the function unless it is done in less than 10 sec

It seems that it sometimes cause corruption of the database, so that a rescan is necessary on the next restart.  Othewise transactions may be missing.

[etotheipi]

OK, so tracing the bug is difficult if one cannot single-step through the function unless it is done in less than 10 sec

It seems that it sometimes cause corruption of the database, so that a rescan is necessary on the next restart.  Othewise transactions may be missing.

Modify the variable MT_WAIT_TIMEOUT_SEC in armoryengine.py.  Set it to UINT32_MAX or something.   That will help with the debugging.

Also, be aware that I explicitly try to avoid corrupting the wallet history by only saving it if there's a clean shutdown.  If you kill Armory, or you close it while it's scanning, offline, or dirty, it will not save any history and will require rescanning. 

[Kyune]

P.S -- Has no one played with the new backup center?  Come on!  It's so badass!

Great to see a M of N paper backup implemented so smoothly.  I've really been looking forward to that as a GUI feature.
Been poking around 0.89.99.3 in a Ubuntu 12.04 VM.  No issues so far.

[cp1]

I look around in parsePrivateKeyData and figured out what I needed to add to my base58 conversion, thanks.

[etotheipi]

I look around in parsePrivateKeyData and figured out what I needed to add to my base58 conversion, thanks.

So what was it?  I just did the same thing as you did at first and got a wacky answer.  I haven't looked at that in forever, so I'm curious what I forgot.

[P_Shep]

It appeared to be working... started up, took forever to scan then showed my balance... but then I hit exit and my machine immediately BSOD'ed :/

[etotheipi]

It appeared to be working... started up, took forever to scan then showed my balance... but then I hit exit and my machine immediately BSOD'ed :/

Wtf?  Certainly haven't heard that one yet...

System specs?