In descriptor wallets, we consider all outputs to be spendable as we no
longer have mixed mine and watchonly in a wallet. As such,
COutput::spendable is meaningless and can be removed.
Furthermore, CoinFilterParams::only_spendable can be removed as that was
essentially checking for COutput::spendable.
Lastly, AvailableCoinsListUnspent can also be removed as the wrapper is
now only setting the feerate to std::nullopt which is trivial enough that
a dedicated wrapper is not needed.
01960c53c7 fuzz: make FuzzedDataProvider usage deterministic (Martin Leitner-Ankerl)
Pull request description:
There exist many usages of `fuzzed_data_provider` where it is evaluated directly in the function call.
Unfortunately, [the order of evaluation of function arguments is unspecified](https://en.cppreference.com/w/cpp/language/eval_order), and a simple example shows that it can differ e.g. between clang++ and g++: https://godbolt.org/z/jooMezWWY
When the evaluation order is not consistent, the same fuzzing/random input will produce different output, which is bad for coverage/reproducibility. This PR fixes all these cases I have found where unspecified evaluation order could be a problem.
Finding these has been manual work; I grepped the sourcecode for these patterns, and looked at each usage individually. So there is a chance I missed some.
* `fuzzed_data_provider`
* `.Consume`
* `>Consume`
* `.rand`
I first discovered this in https://github.com/bitcoin/bitcoin/pull/29013#discussion_r1420236394. Note that there is a possibility that due to this fix the evaluation order is now different in many cases than when the fuzzing corpus has been created. If that is the case, the fuzzing corpus will have worse coverage than before.
Update: In list-initialization the order of evaluation is well defined, so e.g. usages in `initializer_list` or constructors that use `{...}` is ok.
ACKs for top commit:
achow101:
ACK 01960c53c7
vasild:
ACK 01960c53c7
ismaelsadeeq:
ACK 01960c53c7
Tree-SHA512: e56d087f6f4bf79c90b972a5f0c6908d1784b3cfbb8130b6b450d5ca7d116c5a791df506b869a23bce930b2a6977558e1fb5115bb4e061969cc40f568077a1ad
This allows a transaction's weight to be bound under a certain
weight if possible and desired. This can be beneficial for future
RBF attempts, or whenever a more restricted spend topology is
desired.
Co-authored-by: Greg Sanders <gsanders87@gmail.com>
- This commit renames the coin selection algorithms input parameter `max_weight`
to `max_selection_weight` for clarity.
The parameter represent the maximum weight of the UTXOs the coin selection algorithm
should select, not the transaction maximum weight.
- The commit updates the parameter docstring to provide correct description.
- Also updates coin selection unit and fuzzing test variables to match the new name.
bd34dd85e7 Use `exact_target` shorthand in coinselector_tests (Murch)
7aa7e30441 Fold GetSelectionWaste() into ComputeAndSetWaste() (Murch)
Pull request description:
PR #26152 moved waste calculation into SelectionResult to be able to correct the waste score on basis of the bump_fee_group_discount for overlapping ancestries. This left two functions with largely overlapping purpose, where one was simply a wrapper of the other. This PR cleans up the overlap, and fixes the double-meaning of `change_cost` where the `GetChange()` function assumed that no change was created when `change_cost` was set to 0. This behavior was exploited in a bunch of tests, but is problematic, because a `change_cost` of 0 is permitted with custom settings for feerate and discard_feerate (i.e. when they’re both 0).
ACKs for top commit:
achow101:
ACK bd34dd85e7
furszy:
Code ACK bd34dd85e7
ismaelsadeeq:
Code Review ACK bd34dd85e7
Tree-SHA512: 83a2688d45d719dc61a64b5180fe136107faccf401a59df65245c05d701748a03e85ed56fde8c9b7ef39a3ab54374dd3718c559bda5b3f55dafedfd7fed25161
Both `GetSelectionWaste()` and `ComputeAndSetWaste()` now are part of
`SelectionResult`. Instead of `ComputeAndSetWaste()` being a wrapper for
`GetSelectionWaste()`, we combine them to a new function
`RecalculateWaste()`.
As I was combining the logic of the two functions, I noticed that
`GetSelectionWaste()` was making the odd assumption that the
`change_cost` being set to zero means that no change is created.
However, if we build transactions at a feerate of zero with the
`discard_feerate` also set to zero, we'd organically have a
`change_cost` of zero, even when we create change on a transaction.
This commit cleans up this duplicate meaning of `change_cost` and relies
on `GetChange()` to figure out whether there is change on basis of the
`min_viable_change` and whatever is left after deducting fees.
Since this broke a bunch of tests that relied on the double-meaning of
`change_cost` a bunch of tests had to be fixed.
Copying util::Result values is less efficient than moving them because they
allocate memory and contain strings. Also this is needed to avoid compile
errors in https://github.com/bitcoin/bitcoin/pull/25722 which adds a
std::unique_ptr member to util::Result which implicity disables copying.
There exist many usages of `fuzzed_data_provider` where it is evaluated directly in the function call.
Unfortunately, the order of evaluation of function arguments is unspecified. This means it can differ
between compilers/version/optimization levels etc. But when the evaluation order changes, the same
fuzzing input will produce different output, which is bad for coverage/reproducibility.
This PR fixes all these cases where by moving multiple calls to `fuzzed_data_provider` out of the
function arguments.
The valid results should have a target below the sum of
the selected inputs amounts. Also, it increases the
minimum value for target to make it more realistic.
Instead of using `cost_of_change` for `min_viable_change`
and `change_cost`, and 0 for `change_fee`, use values from
`coin_params`. The previous values don't generate any effects
that is relevant for that context.
I discovered via fuzzing of another coin selection approach that at
extremely high feerates SRD may find input sets that lead to
transactions without change outputs. This is an unintended outcome since
SRD is meant to always produce a transaction with a change output—we use
other algorithms to specifically search for changeless solutions.
The issue occures when the flat allowance of 50,000 ṩ for change is
insufficient to pay for the creation of a change output with a non-dust
amount, at and above 1,613 ṩ/vB. Increasing the change budget by
change_fees makes SRD behave as expected at any feerates.
Uses a min-effective-value heap, so we can remove the least valuable input/s
while the selected weight exceeds the maximum allowed weight.
Co-authored-by: Murch <murch@murch.one>
The simplest scenario where this is useful is on the 'check_max_weight' unit test
already:
We create 1515 UTXOs with 0.033 BTC each, and 1 UTXO with 50 BTC. Then perform
Coin Selection.
As the selection of the 1515 small UTXOs exceeds the max allowed tx size, the
expectation here is to receive a selection result that only contain the big
UTXO (which is not happening for the reasons stated below).
As knapsack returns a result that exceeds the max allowed transaction size, we
fallback to SRD, which selects coins randomly up until the target is met. So
we end up with a selection result with lot more coins than what is needed.
And not hide it inside the `OutputGroup::Insert` method.
This method does not return anything if insertion fails.
We can know before calling `Insert` whether the coin
will be accepted or not.
and also fix spelling in test/lint/lint-locale-dependence.py not caught by the
spelling linter and fix up a paragraph we are touching here in test/README.md.
