There is a good reason why I use old versions of bitcoind (and GUI client). I just let you all test it for half a year or so... than it seems like a good and well tested software for me.  

So, MtGox finally GOXXXED himself? I am not surprised at all.

The descriptions of the opcodes also describe this as a push-zero (alias OP_FALSE) rather than push-nothing (there are already 11 NOP opcodes, why do we need another?)

This script appears to me to be potentially solvable if OP_0 is a NOP, unsolvable otherwise.  However it does feel to me like interpreting it as push-zero makes a lot more sense.

I do recognize that this opcode is unlikely to have been used anywhere, but any use of it (or any successful claim of these coins) would make it impossible to make some kind of change here without causing a netsplit.

I recognize that I've probably gone badly off-topic with respect to this topic title now...

Thanks for the contribution (destroy ).

You two are talking about two different things. Miners are allowed to exclude any transactions they want, but they all have to agree on which transactions are valid.

Currently, most miners reject most turing-complete strange scripts, but still accept the full set of op-codes as valid. This is important because each miner must verify the validity or every transaction in the block-chain. For example a rogue miner generating a block with a 100BTC coin creation transaction will be ignored by the honest nodes.

TL:DR: Even if we wanted to prohibit coin destruction (I did not know it was unambiguously possible), it would create a split in the block-chain.

PS: I am still of the opinion that the protocol is essentially untouchable at least until the block-reward drops. What part of "experimental currency" don't people understand? We have to expect a few "flaws". Those flaws will be easier to interpret if we don't monkey with the experiment.

Edit: All Turing-complete scripts must be invalid. Else all miners would stall on the first endless loop included in the block-chain. Currently, "strange" transactions are transactions other than: tranferring a coin from one (set of) address to another, and coin creation.

Bitcoin scripts do not form a turing complete language (there's no loop or "goto"), so the question doesn't arise.

[edit: this was a deliberate design decision by Satoshi]

There's a Solver in the code that matches transactions to only 2 allowed templates/formats. Without passing this template, transactions cannot make it into the memory pool and don't get propagated.

If you wish to make a new tx type you have to lobby for it to be included in the list of templates.

It's highly improbably that 50% would be able to evaluate the matter at hand and make that decision within the timeframe that allows to "undo" this without too much damage (blocks mined after questionable block).

So people might come up with the idea of proxying this decision-making. This might put the decision-making into too few hands, of course, and all the decentralization is down the drain.

I don't think this is a good idea, therefore.

Maybe MagicalTux should try to get some insurance against fuckups like this. Might be expensive and put a lot of hassle on him by ways of auditor wanting to see all code pre-production

In a way, this has already happened.  Surely the three (or so) largest pools already constitute exactly what you have described.

500k BTC are uninsurable, because it is practically impossible to replace them once lost.