wallet: rct specific output selection

Before the normal selection, we attempt to find either one or two
suitable outputs to use as inputs to the rct tx. The intent is that
most rct txes will have one or two inputs, and we want all to look
the same if possible.
When two outputs are needed, we try to find a pair which are not
related (ie, by being from the same or similar block height).

1 files changed, 123 insertions, 3 deletions

diff --git a/src/wallet/wallet2.cpp b/src/wallet/wallet2.cpp
index 737a24fe8..75c912461 100644
--- a/src/wallet/wallet2.cpp
+++ b/src/wallet/wallet2.cpp
@@ -97,14 +97,18 @@ void do_prepare_file_names(const std::string& file_path, std::string& keys_file,
   }
 }
 
-uint64_t calculate_fee(const cryptonote::blobdata &blob, uint64_t fee_multiplier)
+uint64_t calculate_fee(size_t  bytes, uint64_t fee_multiplier)
 {
   THROW_WALLET_EXCEPTION_IF(fee_multiplier <= 0 || fee_multiplier > 3, tools::error::invalid_fee_multiplier);
-  uint64_t bytes = blob.size();
   uint64_t kB = (bytes + 1023) / 1024;
   return kB * FEE_PER_KB * fee_multiplier;
 }
 
+uint64_t calculate_fee(const cryptonote::blobdata &blob, uint64_t fee_multiplier)
+{
+  return calculate_fee(blob.size(), fee_multiplier);
+}
+
 } //namespace
 
 namespace tools
@@ -2098,6 +2102,16 @@ namespace
     size_t idx = crypto::rand<size_t>() % vec.size();
     return pop_index (vec, idx);
   }
+
+  template<typename T>
+  T pop_back(std::vector<T>& vec)
+  {
+    CHECK_AND_ASSERT_MES(!vec.empty(), T(), "Vector must be non-empty");
+
+    T res = vec.back();
+    vec.pop_back();
+    return res;
+  }
 }
 //----------------------------------------------------------------------------------------------------
 // Select random input sources for transaction.
@@ -3012,6 +3026,92 @@ static size_t estimate_rct_tx_size(int n_inputs, int mixin, int n_outputs)
   return size;
 }
 
+// This returns a handwavy estimation of how much two outputs are related
+// If they're from the same tx, then they're fully related. From close block
+// heights, they're kinda related. The actual values don't matter, just
+// their ordering, but it could become more murky if we add scores later.
+float wallet2::get_output_relatedness(const transfer_details &td0, const transfer_details &td1) const
+{
+#if 0
+  // expensive test, and same tx will fall onto the same block height below
+  if (get_transaction_hash(td0.m_tx) == get_transaction_hash(td1.m_tx))
+    return 1.0f;
+#endif
+
+  // same block height -> possibly tx burst, or same tx (since above is disabled)
+  if (td0.m_block_height == td1.m_block_height)
+    return 0.9f;
+
+  // could extract the payment id, and compare them, but this is a bit expensive too
+
+  // similar block heights
+  if ((td0.m_block_height > td1.m_block_height ? td0.m_block_height - td1.m_block_height : td1.m_block_height - td0.m_block_height) < 10)
+    return 0.2f;
+
+  // don't think these are particularly related
+  return 0.0f;
+}
+
+std::vector<size_t> wallet2::pick_prefered_rct_inputs(uint64_t needed_money) const
+{
+  std::vector<size_t> picks;
+  float current_output_relatdness = 1.0f;
+
+  LOG_PRINT_L2("pick_prefered_rct_inputs: needed_money " << print_money(needed_money));
+
+  // try to find a rct input of enough size
+  for (size_t i = 0; i < m_transfers.size(); ++i)
+  {
+    const transfer_details& td = m_transfers[i];
+    if (!td.m_spent && td.is_rct() && td.amount() >= needed_money && is_transfer_unlocked(td))
+    {
+      LOG_PRINT_L2("We can use " << i << " alone: " << print_money(td.amount()));
+      picks.push_back(i);
+      return picks;
+    }
+  }
+
+  // then try to find two outputs
+  // this could be made better by picking one of the outputs to be a small one, since those
+  // are less useful since often below the needed money, so if one can be used in a pair,
+  // it gets rid of it for the future
+  for (size_t i = 0; i < m_transfers.size(); ++i)
+  {
+    const transfer_details& td = m_transfers[i];
+    if (!td.m_spent && td.is_rct() && is_transfer_unlocked(td))
+    {
+      LOG_PRINT_L2("Considering input " << i << ", " << print_money(td.amount()));
+      for (size_t j = i + 1; j < m_transfers.size(); ++j)
+      {
+        const transfer_details& td2 = m_transfers[j];
+        if (!td2.m_spent && td2.is_rct() && td.amount() + td2.amount() >= needed_money && is_transfer_unlocked(td2))
+        {
+          // update our picks if those outputs are less related than any we
+          // already found. If the same, don't update, and oldest suitable outputs
+          // will be used in preference.
+          float relatedness = get_output_relatedness(td, td2);
+          LOG_PRINT_L2("  with input " << j << ", " << print_money(td2.amount()) << ", relatedness " << relatedness);
+          if (relatedness < current_output_relatdness)
+          {
+            // reset the current picks with those, and return them directly
+            // if they're unrelated. If they are related, we'll end up returning
+            // them if we find nothing better
+            picks.clear();
+            picks.push_back(i);
+            picks.push_back(j);
+            LOG_PRINT_L0("we could use " << i << " and " << j);
+            if (relatedness == 0.0f)
+              return picks;
+            current_output_relatdness = relatedness;
+          }
+        }
+      }
+    }
+  }
+
+  return picks;
+}
+
 // Another implementation of transaction creation that is hopefully better
 // While there is anything left to pay, it goes through random outputs and tries
 // to fill the next destination/amount. If it fully fills it, it will use the
@@ -3096,6 +3196,26 @@ std::vector<wallet2::pending_tx> wallet2::create_transactions_2(std::vector<cryp
   adding_fee = false;
   needed_fee = 0;
 
+  // for rct, since we don't see the amounts, we will try to make all transactions
+  // look the same, with 1 or 2 inputs, and 2 outputs. One input is preferable, as
+  // this prevents linking to another by provenance analysis, but two is ok if we
+  // try to pick outputs not from the same block. We will get two outputs, one for
+  // the destination, and one for change.
+  std::vector<size_t> prefered_inputs;
+  if (use_rct)
+  {
+    // this is used to build a tx that's 1 or 2 inputs, and 2 outputs, which
+    // will get us a known fee.
+    uint64_t estimated_fee = calculate_fee(estimate_rct_tx_size(2, fake_outs_count + 1, 2), fee_multiplier);
+    prefered_inputs = pick_prefered_rct_inputs(needed_money + estimated_fee);
+    if (!prefered_inputs.empty())
+    {
+      string s;
+      for (auto i: prefered_inputs) s += print_money(m_transfers[i].amount()) + " ";
+      LOG_PRINT_L1("Found prefered rct inputs for rct tx: " << s);
+    }
+  }
+
   // while we have something to send
   while ((!dsts.empty() && dsts[0].amount > 0) || adding_fee) {
     TX &tx = txes.back();
@@ -3108,7 +3228,7 @@ std::vector<wallet2::pending_tx> wallet2::create_transactions_2(std::vector<cryp
 
     // get a random unspent output and use it to pay part (or all) of the current destination (and maybe next one, etc)
     // This could be more clever, but maybe at the cost of making probabilistic inferences easier
-    size_t idx = !unused_transfers_indices.empty() ? pop_random_value(unused_transfers_indices) : pop_random_value(unused_dust_indices);
+    size_t idx = !prefered_inputs.empty() ? pop_back(prefered_inputs) : !unused_transfers_indices.empty() ? pop_random_value(unused_transfers_indices) : pop_random_value(unused_dust_indices);
 
     const transfer_details &td = m_transfers[idx];
     LOG_PRINT_L2("Picking output " << idx << ", amount " << print_money(td.amount()));