Re: [bitcoin-dev] Testnet3 Reest

[Gregory Maxwell <greg@xiph.org>]

On Thu, Aug 30, 2018 at 11:21 PM Johnson Lau via bitcoin-dev
<bitcoin-dev@lists.linuxfoundation.org> wrote:
> A public testnet is still useful so in articles people could make references to these transactions.
> Maybe we could have 2 testnets at the same time, with one having a smaller block size?

I would much rather have a signed blocks testnet, with a predictable
structured reorg pattern* (and a config flag so you can make your node
ignore all blocks that are going to get reorged out in a reorg of nth
or larger).  There are many applications where the mined testnet just
doesn't give you anything useful... it's too stable when you want it
to be a bit unstable and too wildly unstable when you want a bit of
stability-- e.g. there are very few test cases where a 20,000 block
reorg does anything useful for you; yet they happen on testnet.

We looked at doing this previously in Bitcoin core and jtimon had some
patches,  but the existing approach increased the size of the
blockindex objects in memory  while not in signed testnet mode.   This
could probably have been fixed by turning one of the fields like the
merkel root into a union of it's normal value and a pointer a
look-aside block index that is used only in signed block testnet mode.

Obviously such a mode wouldn't be a replacement for an ordinary
testnet, but it would be a useful middle ground between regtest (that
never sees anything remotely surprising and can't easily be used for
collaborative testing) and full on testnet where your attempts to test
against ordinary noise require you cope your entirely universe being
removed from existence and replaced by something almost but not quite
entirely different at the whim of some cthulhuian blind idiot god.