KarpelesLab / libwallet / 25301339880

24%

push

github

wlttx/bitcoin: fix double txid byte-reversal that broke every send

Root cause of the long-standing bitcoin "bad-txns-inputs-
missingorspent" failures: parseTxoRef was reversing the txid
bytes before storing them in BtcTxInput.TXID, but
outscript.BtcTxInput.Bytes() ALSO reverses TXID at marshal time
(btctx.go:548 — the comment there even calls out that TXID is
stored in displayable / big-endian form internally and the
marshaler converts to wire format). Compounded, the wire bytes
emitted in our broadcast tx were the modchain-reported txid
WITH ITS BYTES REVERSED — i.e. the broadcast referenced a
bogus txid the chain has never seen, so the node rejected it
with "missing or spent".

Verified end-to-end with the user's actual failure: modchain
reported owned UTXO d36a0d…b9b:0 (alive on-chain, gettxout
confirms 0.1 LTC at that displayable id). Our broadcast wrote
wire bytes that, re-reversed, displayed as 9bbf61…6ad3:0
(no such UTXO). Litecoin node returned -25.

Same bug pattern in SignRawBitcoinTx (mpurse / Counterparty
flows): when looking up wire-decoded inputs against modchain's
owned set, we reversed TXID before formatting the lookup key.
TXID is already in displayable form post-UnmarshalBinary
(btctx.go:626 reverses on read), so the extra reverse made
every input report as "not owned by this account".

Both fixes: drop the reverse, let outscript handle the wire
conversion. Regression test pins the byte-order convention.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

1 of 10 new or added lines in 1 file covered. (10.0%)

3192 of 15279 relevant lines covered (20.89%)

124.45 hits per line