as EraseLastKElements() called in the next line performs the same operation.
Thanks to Martin Zumsande (lightlike) for seeing this while reviewing.
Co-authored-by: Martin Zumsande <mzumsande@gmail.com>
ffff0d0442 refactor: Switch serialize to uint8_t (1/n) (MarcoFalke)
Pull request description:
Replace `char` -> `uint8_t` in serialization where a sign doesn't make sense (char might be signed/unsigned).
ACKs for top commit:
practicalswift:
cr ACK ffff0d0442: patch looks correct and commit hash is ffffresh (was bbbbadass)
kristapsk:
ACK ffff0d0442
Tree-SHA512: cda682280c21d37cc3a6abd62569732079b31d18df3f157aa28bed80bd6f9f29a7db5c133b1f57b3a8f8d5ba181a76e473763c6e26a2df6d9244813f56f893ee
fe3d17df04 net: ignore block-relay-only peers when skipping DNS seed (Anthony Towns)
Pull request description:
Since #17428 bitcoind will attempt to reconnect to two block-relay-only anchors before doing any other outbound connections. When determining whether to use DNS seeds, it will currently see these two peers and decide "we're connected to the p2p network, so no need to lookup DNS" -- but block-relay-only peers don't do address relay, so if your address book is full of invalid addresses (apart from your anchors) this behaviour will prevent you from recovering from that situation.
This patch changes it so that it only skips use of DNS seeds when there are two full-outbound peers, not just block-relay-only peers.
ACKs for top commit:
Sjors:
utACK fe3d17d
amitiuttarwar:
ACK fe3d17df04, this impacts the very common case where we stop/start a node, persisting anchors & have a non-empty addrman (although, to be clear, wouldn't be particularly problematic in the common cases where the addrman has valid addresses)
mzumsande:
ACK fe3d17df04
jonatack:
ACK fe3d17df04
prayank23:
tACK fe3d17df04
Tree-SHA512: 9814b0d84321d7f45b5013eb40c420a0dd93bf9430f5ef12dce50d1912a18d5de2070d890a8c6fe737a3329b31059b823bc660b432d5ba21f02881dc1d951e94
0829516d1f [refactor] Remove unused ForEachNodeThen() template (John Newbery)
09cc66c00e scripted-diff: rename address relay fields (John Newbery)
76568a3351 [net processing] Move addr relay data and logic into net processing (John Newbery)
caba7ae8a5 [net processing] Make RelayAddress() a member function of PeerManagerImpl (John Newbery)
86acc96469 [net processing] Take NodeId instead of CNode* as originator for RelayAddress() (John Newbery)
Pull request description:
This continues the work of moving application layer data into net_processing, by moving all addr data into the new Peer object added in #19607.
For motivation, see #19398.
ACKs for top commit:
laanwj:
Code review ACK 0829516d1f
mzumsande:
ACK 0829516d1f, reviewed the code and ran tests.
sipa:
utACK 0829516d1f
hebasto:
re-ACK 0829516d1f
Tree-SHA512: efe0410fac288637f203eb37d1999910791e345872d37e1bd5cde50e25bb3cb1c369ab86b3a166ffd5e06ee72e4508aa2c46d658be6a54e20b4f220d2f57d0a6
7075f604e8 scripted-diff: update noban documentation in net_processing.cpp (Jon Atack)
a95540cf43 scripted-diff: rename NetPermissionFlags enumerators (Jon Atack)
810d0929c1 p2p, refactor: make NetPermissionFlags a uint32 enum class (Jon Atack)
7b55a94497 p2p: NetPermissions::HasFlag() pass flags param by value (Jon Atack)
91f6e6e6d1 scripted-diff: add NetPermissionFlags scopes where not already present (Jon Atack)
Pull request description:
While reviewing #20196, I noticed the `NetPermissionFlags` enums are frequently called as if they were scoped, yet are still global. This patch upgrades `NetPermissionFlags` to a scoped class enum and updates the enumerator naming, similarly to #19771. See https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#enum-enumerations for more info.
This change would eliminate the class of bugs like https://github.com/bitcoin/bitcoin/pull/20196#discussion_r610770148 and #21644, as only defined operations on the flags would compile.
ACKs for top commit:
laanwj:
Code review ACK 7075f604e8
vasild:
ACK 7075f604e8
Tree-SHA512: 7fcea66ee499f059efc78c934b5f729b3c8573fe304dee2c27c837c2f662b89324790568246d75b2a574cf9f059b42d3551d928996862f4358055eb43521e6f4
105941b726 net: use stronger AddLocal() for our I2P address (Vasil Dimov)
Pull request description:
There are two issues:
### 1. Our I2P address not added to local addresses.
* `externalip=` is used with an IPv4 address (this sets automatically `discover=0`)
* No `discover=1` is used
* `i2psam=` is used
* No `externalip=` is used for our I2P address
* `listenonion=1 torcontrol=` are used
In this case `AddLocal(LOCAL_MANUAL)` [is used](94f83534e4/src/torcontrol.cpp (L354)) for our `.onion` address and `AddLocal(LOCAL_BIND)` [for our](94f83534e4/src/net.cpp (L2247)) `.b32.i2p` address, the latter being [ignored](94f83534e4/src/net.cpp (L232-L233)) due to `discover=0`.
### 2. Our I2P address removed from local addresses even if specified with `externalip=` on I2P proxy restart.
* `externalip=` is used with our I2P address (this sets automatically `discover=0`)
* No `discover=1` is used
* `i2psam=` is used
In this case, initially `externalip=` causes our I2P address to be [added](94f83534e4/src/init.cpp (L1266)) with `AddLocal(LOCAL_MANUAL)` which overrides `discover=0` and works as expected. However, if later the I2P proxy is shut down [we do](94f83534e4/src/net.cpp (L2234)) `RemoveLocal()` in order to stop advertising our I2P address (since we have lost I2P connectivity). When the I2P proxy is started and we reconnect to it, restoring the I2P connectivity, [we do](94f83534e4/src/net.cpp (L2247)) `AddLocal(LOCAL_BIND)` which does nothing due to `discover=0`.
To resolve those two issues, use `AddLocal(LOCAL_MANUAL)` for I2P which is also what we do with Tor.
ACKs for top commit:
laanwj:
Code review ACK 105941b726
Tree-SHA512: 0c9daf6116b8d9c34ad7e6e9bbff6e8106e94e4394a815d7ae19287aea22a8c7c4e093c8dd8c58a4a1b1412b2575a9b42b8a93672c8d17f11c24508c534506c7
- drop redundant PF_ permission flags prefixes
- drop ALL_CAPS naming per https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#Renum-caps
- rename IsImplicit to Implicit
-BEGIN VERIFY SCRIPT-
s() { git grep -l "$1" src | xargs sed -i "s/$1/$2/g"; }
s 'PF_NONE' 'None'
s 'PF_BLOOMFILTER' 'BloomFilter'
s 'PF_RELAY' 'Relay'
s 'PF_FORCERELAY' 'ForceRelay'
s 'PF_DOWNLOAD' 'Download'
s 'PF_NOBAN' 'NoBan'
s 'PF_MEMPOOL' 'Mempool'
s 'PF_ADDR' 'Addr'
s 'PF_ISIMPLICIT' 'Implicit'
s 'PF_ALL' 'All'
-END VERIFY SCRIPT-
-BEGIN VERIFY SCRIPT-
s() { git grep -l "$1" -- 'src' ':!src/net_permissions.h' | xargs sed -i -E "s/([^:])$1/\1NetPermissionFlags::$1/"; }
s 'PF_NONE'
s 'PF_BLOOMFILTER'
s 'PF_RELAY'
s 'PF_FORCERELAY'
s 'PF_DOWNLOAD'
s 'PF_NOBAN'
s 'PF_MEMPOOL'
s 'PF_ADDR'
s 'PF_ISIMPLICIT'
s 'PF_ALL'
-END VERIFY SCRIPT-
Co-authored-by: Hennadii Stepanov <32963518+hebasto@users.noreply.github.com>
792be53d3e refactor: Replace std::bind with lambdas (Hennadii Stepanov)
a508f718f3 refactor: Use appropriate thread constructor (Hennadii Stepanov)
30e4448215 refactor: Make TraceThread a non-template free function (Hennadii Stepanov)
Pull request description:
This PR does not change behavior.
Its goal is to improve readability and maintainability of the code.
ACKs for top commit:
jnewbery:
utACK 792be53d3e
jonatack:
tACK 792be53d3e
MarcoFalke:
cr ACK 792be53d3e
Tree-SHA512: a03142f04f370f6bc02bd3ddfa870819b51740fcd028772241d68c84087f95a2d78207cbd5edb3f7c636fcf2d76192d9c59873f8f0af451d3b05c0cf9cf234df
36fb036d25 p2p: allow NetPermissions::ClearFlag() only with PF_ISIMPLICIT (Jon Atack)
4e0d5788ba test: add net permissions noban/download unit test coverage (Jon Atack)
dde69f20a0 p2p, bugfix: use NetPermissions::HasFlag() in CConnman::Bind() (Jon Atack)
Pull request description:
This is a bugfix follow-up to #16248 and #19191 that was noticed in #21506. Both v0.21 and master are affected.
Since #19191, noban is a multi-flag that implies download, so the conditional in `CConnman::Bind()` using a bitwise AND on noban will return the same result for both the noban status and the download status. This means that download peers are incorrectly not being added to local addresses because they are mistakenly seen as noban peers.
The second commit adds unit test coverage to illustrate and test the noban/download relationship and the `NetPermissions` operations involving them.
The final commit adds documentation and disallows calling `NetPermissions::ClearFlag()` with any second param other than `NetPermissionFlags` "implicit" -- per current usage in the codebase -- because `ClearFlag()` should not be called with any second param that is a subflag of a multiflag, e.g. "relay" or "download," as that would leave the result in an invalid state corresponding to none of the existing NetPermissionFlags. Thanks to Vasil Dimov for noticing this.
ACKs for top commit:
theStack:
re-ACK 36fb036d25☕
vasild:
ACK 36fb036d25
hebasto:
ACK 36fb036d25, I have reviewed the code and it looks OK, I agree it can be merged.
kallewoof:
Code review ACK 36fb036d25
Tree-SHA512: 5fbc7ddbf31d06b35bf238f4d77ef311e6b6ef2e1bb9893f32f889c1a0f65774a3710dcb21d94317fe6166df9334a9f2d42630809e7fe8cbd797dd6f6fc49491
There are two issues:
1. Our I2P address not added to local addresses.
* `externalip=` is used with an IPv4 address (this sets automatically
`discover=0`)
* No `discover=1` is used
* `i2psam=` is used
* No `externalip=` is used for our I2P address
* `listenonion=1 torcontrol=` are used
In this case `AddLocal(LOCAL_MANUAL)` is used for our `.onion` address
and `AddLocal(LOCAL_BIND)` for our `.b32.i2p` address, the latter being
ignored due to `discover=0`.
2. Our I2P address removed from local addresses even if specified
with `externalip=` on I2P proxy restart.
* `externalip=` is used with our I2P address (this sets automatically
`discover=0`)
* No `discover=1` is used
* `i2psam=` is used
In this case, initially `externalip=` causes our I2P address to be added
with `AddLocal(LOCAL_MANUAL)` which overrides `discover=0` and works as
expected. However, if later the I2P proxy is shut down we do
`RemoveLocal()` in order to stop advertising our I2P address (since we
have lost I2P connectivity). When the I2P proxy is started and we
reconnect to it, restoring the I2P connectivity, we do
`AddLocal(LOCAL_BIND)` which does nothing due to `discover=0`.
To resolve those two issues, use `AddLocal(LOCAL_MANUAL)` for I2P which
is also what we do with Tor.
d66f283ac0 scripted-diff: Replace three dots with ellipsis in the UI strings (Hennadii Stepanov)
Pull request description:
This PR is split from #21463.
The change was suggested on [Transifex.com](https://www.transifex.com/bitcoin/bitcoin/), and it does not touch `LogPrint` and `LogPrintf` calls.
The only comment on #21463 [was](9030e4b5a6 (r597220100)):
> Mind that these messages also end up in the log. In principle the log is already UTF-8 (as are all strings and text in bitcoind). But, just noting, that it might make browsing the log a less pleasant experience on systems with misconfigured locale like some BSDs by default.
ACKs for top commit:
laanwj:
ACK d66f283ac0
Tree-SHA512: 5ab1cb3160f3f996f1ad7d7486662da3eb7f06a857f4a1874963ce10caed5b86b0ad6151b1b9ebeb2b8aa5f0c85efad3b768ea9cafe5db86f78f88912b756d1e
- Use `SanitizeString` when logging message errors to make sure that the
message type is sanitized.
- For the `MESSAGESTART` error don't inspect and log header details at
all: receiving invalid start bytes makes it likely that the packet isn't
even formatted as valid P2P message. Logging the four unexpected start
bytes should be enough.
- Update `p2p_invalid_messages.py` test to check this.
Issue reported by gmaxwell.
9096b13a47 net: remove unnecessary check of CNode::cs_vSend (Vasil Dimov)
Pull request description:
It is not possible to have a node in `CConnman::vNodesDisconnected` and
its reference count to be incremented - all `CNode::AddRef()` are done
either before the node is added to `CConnman::vNodes` or while holding
`CConnman::cs_vNodes` and the object being in `CConnman::vNodes`.
So, the object being in `CConnman::vNodesDisconnected` and its reference
count being zero means that it is not and will not start to be used by
other threads.
So, the lock of `CNode::cs_vSend` in `CConnman::DisconnectNodes()` will
always succeed and is not necessary.
Indeed all locks of `CNode::cs_vSend` are done either when the reference
count is >0 or under the protection of `CConnman::cs_vNodes` and the
node being in `CConnman::vNodes`.
ACKs for top commit:
MarcoFalke:
review ACK 9096b13a47🏧
jnewbery:
utACK 9096b13a47
Tree-SHA512: 910899cdcdc8934642eb0c40fcece8c3b01b7e20a0b023966b9d6972db6a885cb3a9a04e9562bae14d5833967e45e2ecb3687b94d495060c3da4b1f2afb0ac8f
8c8237a4a1 net, refactor: Fix style in CConnman::StopNodes (Hennadii Stepanov)
229ac1892d net: Combine two loops into one, and update comments (Hennadii Stepanov)
a3d090d110 net: Restrict period when cs_vNodes mutex is locked (Hennadii Stepanov)
Pull request description:
This PR restricts the period when the `cs_vNodes` mutex is locked, prevents the only case when `cs_vNodes` could be locked before the `::cs_main`.
This change makes the explicit locking of recursive mutexes in the explicit order redundant.
ACKs for top commit:
jnewbery:
utACK 8c8237a4a1
vasild:
ACK 8c8237a4a1
ajtowns:
utACK 8c8237a4a1 - logic seems sound
MarcoFalke:
review ACK 8c8237a4a1👢
Tree-SHA512: a8277924339622b188b12d260a100adf5d82781634cf974320cf6007341f946a7ff40351137c2f5369aed0d318f38aac2d32965c9b619432440d722a4e78bb73
It is not possible to have a node in `CConnman::vNodesDisconnected` and
its reference count to be incremented - all `CNode::AddRef()` are done
either before the node is added to `CConnman::vNodes` or while holding
`CConnman::cs_vNodes` and the object being in `CConnman::vNodes`.
So, the object being in `CConnman::vNodesDisconnected` and its reference
count being zero means that it is not and will not start to be used by
other threads.
So, the lock of `CNode::cs_vSend` in `CConnman::DisconnectNodes()` will
always succeed and is not necessary.
Indeed all locks of `CNode::cs_vSend` are done either when the reference
count is >0 or under the protection of `CConnman::cs_vNodes` and the
node being in `CConnman::vNodes`.
PF_NOBAN is a multi-flag that includes PF_DOWNLOAD, so the conditional
in CConnman::Bind() using a bitwise AND will return the same result
for both the "noban" status and the "download" status.
Example:
`PF_DOWNLOAD` is `0b1000000`
`PF_NOBAN` is `0b1010000`
This makes a check like `flags & PF_NOBAN` return `true` even if `flags`
is equal to `PF_DOWNLOAD`.
If `-whitebind=download@1.1.1.1:8765` is specified, then `1.1.1.1:8765`
should be added to the list of local addresses. We only want to avoid
adding to local addresses (that are advertised) a whitebind that has a
`noban@` flag.
As a result of a mis-check in `CConnman::Bind()` we would not have added
`1.1.1.1:8765` to the local addresses in the example above.
Co-authored-by: Vasil Dimov <vd@FreeBSD.org>
b2ee8b207d net: Deserialize hardcoded seeds from BIP155 blob (W. J. van der Laan)
9b29d5df7f contrib: Add explicit port numbers for testnet seeds (W. J. van der Laan)
2a257de113 contrib: Add a few TorV3 seed nodes (W. J. van der Laan)
06030f7a42 contrib: generate-seeds.py generates output in BIP155 format (W. J. van der Laan)
Pull request description:
Closes#20239 and mitigates my node's problem in #21351.
- Add a few hardcoded seeds for TorV3
- As the [bitcoin-seeder](https://github.com/sipa/bitcoin-seeder) doesn't collect TorV3 addresses yet, I have extracted these from my own node using [a script](https://gist.github.com/laanwj/b3d7b01ef61ce07c2eff0a72a6b90183) and added them manually. This is intended to be a temporary stop gap until 22.0's seeds update.
- Change hardcoded seeds to variable length BIP155 binary format.
- It is stored as a single serialized blob in a byte array, instead of pseudo-IPv6 address slots. This is more flexible and, assuming most of the list is IPv4, more compact.
- Only the (networkID, addr, port) subset (CService). Services and time are construed on the fly as before.
- Change input format for `nodes_*.txt`.
- Drop legacy `0xAABBCCDD` format for IPv4. It is never generated by `makeseeds.py`.
- Stop interpreting lack of port as default port, interpret it as 'no port', to accomodate I2P and other port-less protocols (not handled in this PR). An explicit port is always generated by `makeseeds.py` so in practice this makes no difference right now.
A follow-up to this PR could do the same for I2P.
ACKs for top commit:
jonatack:
ACK b2ee8b207d
Tree-SHA512: 11a6b54f9fb0192560f2bd7b218f798f86c1abe01d1bf37f734cb88b91848124beb2de801ca4e6f856e9946aea5dc3ee16b0dbb9863799e42eec1b239d40d59d
4783115fd4 net: add ifaddrs.h include (fanquake)
879215e665 build: check if -lsocket is required with *ifaddrs (fanquake)
87deac66aa rand: only try and use freeifaddrs if available (fanquake)
Pull request description:
Fixes#21485 by linking against `-lsocket` when it's required for using `*ifaddrs` functions.
ACKs for top commit:
laanwj:
Code review ACK 4783115fd4
hebasto:
ACK 4783115fd4, I have reviewed the code and it looks OK, I agree it can be merged.
Tree-SHA512: 4542e036e9b029de970eff8a9230fe45d9204bb22313d075f474295d49bdaf1f1cbb36c0c6e2fa8dbbcdba518d8d3a68a6116ce304b82414315f333baf9af0e4
40316a37cb test: add I2P test for a runaway SAM proxy (Vasil Dimov)
2d8ac77970 fuzz: add tests for the I2P Session public interface (Vasil Dimov)
9947e44de0 i2p: use pointers to Sock to accommodate mocking (Vasil Dimov)
82d360b5a8 net: change ConnectSocketDirectly() to take a Sock argument (Vasil Dimov)
b5861100f8 net: add connect() and getsockopt() wrappers to Sock (Vasil Dimov)
5a887d49b2 fuzz: avoid FuzzedSock::Recv() repeated errors with EAGAIN (Vasil Dimov)
3088f83d01 fuzz: extend FuzzedSock::Recv() to support MSG_PEEK (Vasil Dimov)
9b05c49ade fuzz: implement unimplemented FuzzedSock methods (Vasil Dimov)
Pull request description:
Change the networking code and the I2P code to be fully mockable and use `FuzzedSocket` to fuzz the I2P methods `Listen()`, `Accept()` and `Connect()`.
Add a mocked `Sock` implementation that returns a predefined data on reads and use it for a regression unit test for the bug fixed in https://github.com/bitcoin/bitcoin/pull/21407.
ACKs for top commit:
practicalswift:
Tested ACK 40316a37cb
MarcoFalke:
Concept ACK 40316a37cb
jonatack:
re-ACK 40316a37cb reviewed `git range-diff 01bb3afb 23c861d 40316a3` and the new unit test commit, debug built, ran unit tests, ran bitcoind with an I2P service and network operation with seven I2P peers (2 in, 5 out) is looking nominal
laanwj:
Code review ACK 40316a37cb
Tree-SHA512: 7fc4f129849e16e0c7e16662d9f4d35dfcc369bb31450ee369a2b97bdca95285533bee7787983e881e5a3d248f912afb42b4a2299d5860ace7129b0b19623cc8
0cca08a8ee Add unit test coverage for our onion peer eviction protection (Jon Atack)
caa21f586f Protect onion+localhost peers in ProtectEvictionCandidatesByRatio() (Jon Atack)
8f1a53eb02 Use EraseLastKElements() throughout SelectNodeToEvict() (Jon Atack)
8b1e156143 Add m_inbound_onion to AttemptToEvictConnection() (Jon Atack)
72e30e8e03 Add unit tests for ProtectEvictionCandidatesByRatio() (Jon Atack)
ca63b53ecd Use std::unordered_set instead of std::vector in IsEvicted() (Jon Atack)
41f84d5ecc Move peer eviction tests to a separate test file (Jon Atack)
f126cbd6de Extract ProtectEvictionCandidatesByRatio from SelectNodeToEvict (Jon Atack)
Pull request description:
Now that #19991 and #20210 have been merged, we can determine inbound onion peers using `CNode::m_inbound_onion` and add it to the localhost peers protection in `AttemptToEvictConnection`, which was added in #19670 to address issue #19500.
Update 28 February 2021: I've updated this to follow gmaxwell's suggestion in https://github.com/bitcoin/bitcoin/pull/20197#issuecomment-713865992.
This branch now protects up to 1/4 onion peers (connected via our tor control service), if any, sorted by longest uptime. If any (or all) onion slots remain after that operation, they are then allocated to protect localhost peers, or a minimum of 2 localhost peers in the case that no onion slots remain and 2 or more onion peers were protected, sorted as before by longest uptime.
This patch also adds test coverage for the longest uptime, localhost, and onion peer eviction protection logic to build on the welcome initial unit testing of #20477.
Suggest reviewing the commits that move code with `colorMoved = dimmed-zebra` and `colorMovedWs = allow-indentation-change`.
Closes#11537.
ACKs for top commit:
laanwj:
Code review ACK 0cca08a8ee
vasild:
ACK 0cca08a8ee
Tree-SHA512: 2f5a63f942acaae7882920fc61f0185dcd51da85e5b736df9d1fc72343726dd17da740e02f30fa5dc5eb3b2d8345707aed96031bec143d48a2497a610aa19abd
52dd40a9fe test: add missing netaddress include headers (Jon Atack)
6f09c0f6b5 util: add missing braces and apply clang format to SplitHostPort() (Jon Atack)
2875a764f7 util: add ParseUInt16(), use it in SplitHostPort() (Jon Atack)
6423c8175f p2p, refactor: pass and use uint16_t CService::port as uint16_t (Jon Atack)
Pull request description:
As noticed during review today in https://github.com/bitcoin/bitcoin/pull/20685#discussion_r584873708 of the upcoming I2P network support, `CService::port` is `uint16_t` but is passed around the codebase and into the ctors as `int`, which causes uneeded conversions and casts. We can avoid these (including in the incoming I2P code without further changes to it) by using ports with the correct type. The remaining conversions are pushed out to the user input boundaries where they can be range-checked and raise with user feedback in the next patch.
ACKs for top commit:
practicalswift:
cr ACK 52dd40a9fe: patch looks correct
MarcoFalke:
cr ACK 52dd40a9fe
vasild:
ACK 52dd40a9fe
Tree-SHA512: 203c1cab3189a206c55ecada77b9548b810281cdc533252b8e3330ae0606b467731c75f730ce9deb07cbaab66facf97e1ffd2051084ff9077cba6750366b0432
Now that we have a reliable way to detect inbound onion peers, this commit
updates our existing eviction protection of 1/4 localhost peers to instead
protect up to 1/4 onion peers (connected via our tor control service), sorted by
longest uptime. Any remaining slots of the 1/4 are then allocated to protect
localhost peers, or 2 localhost peers if no slots remain and 2 or more onion
peers are protected, sorted by longest uptime.
The goal is to avoid penalizing onion peers, due to their higher min ping times
relative to IPv4 and IPv6 peers, and improve our diversity of peer connections.
Thank you to Gregory Maxwell, Suhas Daftuar, Vasil Dimov and Pieter Wuille
for valuable review feedback that shaped the direction.
