Add a test parsing transaction with a huge witness

[Kixunil]

This transaction broke past versions of rust-bitcoin and LND so this adds a test to avoid reintroducing the problem in the future.

See also romanz/electrs#783

I'm publishing this immediately for research purposes. I can clean it up later if required (low on time rn) or we may even not merge it if there is a better test.

[romanz]

IIUC, the failure is due to byte_size being 12000072 = 500003 (len) * 24 (mem::size_of::<$type>):

                let byte_size = (len as usize)
                                    .checked_mul(mem::size_of::<$type>())
                                    .ok_or(self::Error::ParseFailed("Invalid length"))?;
                if byte_size > MAX_VEC_SIZE {
                    return Err(self::Error::OversizedVectorAllocation { requested: byte_size, max: MAX_VEC_SIZE })
                }

and indeed the witness definition has changed between 0.27.1 and 0.28.0:

diff --git a/src/blockdata/transaction.rs b/src/blockdata/transaction.rs
index 4ecac18..445da7d 100644
--- a/src/blockdata/transaction.rs
+++ b/src/blockdata/transaction.rs
@@ -196,7 +198,7 @@ pub struct TxIn {
     /// Encodable/Decodable, as it is (de)serialized at the end of the full
     /// Transaction. It *is* (de)serialized with the rest of the TxIn in other
     /// (de)serialization routines.
-    pub witness: Vec<Vec<u8>>
+    pub witness: Witness
 }

The fix was introduced in 2fd0125.

[TheBlueMatt]

Damn, ideally this kind of thing would have been addressed through #1023.

[TheBlueMatt]

Ah, this is about rust-bitcoin 0.28, the above landed in 0.29 (according to github). So hopefully we have a substantially more robust answer to these kinds of bugs in 0.29 already.

[Kixunil]

AFAIU even 0.28 is unaffected. This PR just adds test out of caution.

[elichai]

ACK, this is an important test, as this tx did break production systems

[sanket1729]

I could not foresee the impact when I opened #997. It also lists other potential ways in which block decoding could fail, but all of them required the miner to specifically craft a non-standard tx. Perhaps, this is a lesson for us for back-porting all bug fixes related to decoding/encoding even if they "require miner assistance to deploy".

[sanket1729]

Upon more investigation, this issue is not completely fixed in 0.28. The PR that changes Witness only accidentally fixes it for 0.28 while decoding things with a large Witness. #997 lists other things that are still possible to crash with 0.28. This is cleanly fixed in 0.29 with #1023.

[sanket1729]

I have created #1360 as a backport for 0.28 users.

[tcharding]

ACK d6ca7e4