Pieter Wuille

I've pushed a new update, just replacing the example benchmarks. They're now gathered based on real-world clusters (seen in simulations of historical mempool data) at 3 distinct commits ("track upper...

I'm happy with either approach: * Conditionally stubbing out the body of `secp256k1_context_create` etc. and have them always return failure when compiled without allocation functions. * Splitting `secp256k1.h` into separate...

Concept ACK

@davidgumberg Good points. I've made minimal changes in this PR to make sure no position-effecting operations remain, as we're past the feature freeze. Extending the `AutoFile` interface so that `AutoFile::Get`...

I made a significant change here, making `m_position` an `std::optional`, which is `std::nullopt` when the position is unknown (due to `ftell` failing, and no successful `fseek` afterwards). The position being...

@achow101 I've added a commit that gets rid of `AutoFile::Get` entirely (based on @davidgumberg's d238c46a44292d5ed81d703089b66be333a68083).

I'm aware this is a big chunk of code, but perhaps this can aid review a bit. The biggest commits is [txgraph: (feature) add initial version](https://github.com/bitcoin/bitcoin/pull/31363/commits/afc6a665770004411f1b57f4f43dda916dadd8f7). To get an idea...

I have pushed a substantial change to the `TxGraphImpl::GroupClusters` function (in commit "txgraph: (feature) add initial version"). In a benchmark with 64000 transactions being combined into 1000 clusters, this makes...

@ismaelsadeeq TxGraph (and DepGraph, and FeeFrac) just treat "fee" and "size" as numbers whose ratio is to be maximized. These classes don't care what meaning they correspond to. In practice,...

A few changes: * Added a `TxGraph::DoWork()` functions which performs queued-up computations now, so that future operations are fast. * Make `Ref&` arguments to `TxGraph::AddDependency`, `TxGraph::RemoveTransaction`, and `TxGraph::SetTransactionFee` const. They...