I maintain a wallet that uses bitcoin-cli for signing transactions, and its test system uses regtest to create a bunch of UTXOs for the signing tests to sign. These tests use a git-committed wallet.dat that is copied into the newly-created datadir before launching bitcoind.

For years, up until bc886fc (part of #24699), this was stable and worked well. Occasionally I would need to recreate all the transactions when this project changed mining (#24732) or coin selection (#24584) but I could handle that.

What I can't handle is having the generated transactions be different each time I run the exact same sequence of bitcoin-cli commands starting from the same wallet.dat. Starting with bc886fc, that's what happens.

Is this a supported use model, or is it unreasonable to expect this to be stable?

Summarizing:

• v0.20.0 -- stable
• v0.21.0 -- stable
• v22.0 -- stable
• v23.0 -- stable
• 2723560 (which is bc886fc^) -- stable
• bc886fc -- unstable
• v24.0rc2 -- unstable
• Current master (f37bd15) -- unstable

Expected behavior

The generated transactions should be identical given an identical series of bitcoin-cli calls.

Actual behavior

The generated transactions choose different input coins from the regtest blockchain every time I run.

To reproduce

In my Glacier wallet fork, run t/online_regtest_wallet.py recreate-all-tests; git commit the result. Run the same command again -- git status should show no changes.

When it's unstable, dozens of files change.

System information

Running on Whonix 16. Building inside the git repo as such:

(cd depends && make -j4 NO_QT=1) && ./autogen.sh && ./configure --prefix=`pwd`/depends/x86_64-pc-linux-gnu && make -j4