rust-bitcoin

Update nightly toolchain to nightly-2026-02-21

Same as done by bot in #5717 but with a manual update to the API text
files because `UnsafeUnpin` impls now show up.

ref: https://doc.rust-lang.org/nightly/core/marker/trait.UnsafeUnpin.html

consensus_encoding, primitives, units: follow rust doc title convention

Follow the standard H1 title convention from the rust standard library.

consensus_encoding: beef up module docs

More conventional C-CRATE-DOC and C-EXAMPLE. The crate doc is essentially
a stripped down version of the original ADR doc. C-EXAMPLE calls for an
example on every public item within reason. It would be very repetative
to apply examples on every item in this crate, a single encoder and
decoder example can probably be used by all items.

units: Introduce coverage for Display on error types

The error types in units are largely untested by existing tests.
Since it's important to ensure that error display some content, we
should include assertions that error display messages are non-empty.

Introduce tests to cover Display impls for all error types in units

Fix Error source for ParseDenominationError

The ParseDenominationError return None for the error source in the
std::error::Error trait implementation. Since the type is a wrapper
enum for other errors, it should return the inner error in source().

Return the inner error types for ParseDenominationError in source().

Rearrange types in hex_codec

The hex_codec module was created over various patches and thus doesn't
follow best practice for ordering. Errors should be near the bottom,
with important types near the top.

Rearrange hex_codec to put HexPrimitive at the top, and
ParsePrimitiveError at the bottom.

primitives: Move hex_codec to independent module

The hex_codec module in lib.rs of primitives now makes up 2/3rds of
the file. At this scale, it is better suited to be moved into its
own file.

Move the hex_codec module in lib.rs to a new module file, and import
it as a pub(crate) mod.

Clean up function comments and visibility in hex_codec

The hex_codec module should be entirely hidden to outside the crate.
While all types are pub(crate), the inner field of HexPrimitive was
pub.

Remove alloc requirement for HexPrimitive

The Display/Debug/UpperHex/LowerHex implementations on HexPrimitive
all relied on alloc due to the implementation of hex_write_with_case.
This locks LowerHex and UpperHex on Header behind alloc, and led to
an unnecessary custom implementation of Display.

Change hex_write_with_case to remove alloc requirement, and remove
corresponding feature gates. Move hex_write_with_case to inherent
function on HexPrimitive, and rename to fmt_hex. Replace Display impl
on Header with call through to HexPrimitive. Remove feature gates on
Header Upper/LowerHex.

Adjust from_str in HexPrimitive to be no-alloc

The from_str implementation for parsing types from hex strings currently
relies on alloc due to the use of hex::decode_to_vec. This locks Header
FromStr decoding behind alloc.

Implement from_str without alloc, and adjust feature gates on
HeaderDecoder to match new relaxed requirements.

Fix to_secret_key docs on Keypair

The to_secret_key function docs state that it returns a PrivateKey
type, but the function actually returns a secp256k1::SecretKey.

Fix to_secret_key docs on Keypair to correctly reference
secp256k1::SecretKey return type.

Replace old encoding logic with new encoders in sighash

The sighash module contains a lot of encoding logic for types that make
use of the old consensus::Encodable trait's consensus_encode function.
With the introduction of the new consensus_encoding crate and
associated encoders, this logic can all be replaced with call-throughs
to the new encoders.

Replace all uses of consensus::Encodable logic with equivalent encoding
logic from consensus_encoding::Encodable/Encoder impls.

consensus_encoding: expose vis fragment on exposed macros

Following the C-MACRO-VIS API convention, expose a vis fragment to allow
the caller to choose the output's visibility. This change is backwards
compatible.

io: Add flush_to_writer function

In some cases, we may have an Encoder instance, without a corresponding
Encodable. In these cases, having a way to flush the encoder to an io
writer is extremely useful to avoid having to replicate the logic of
encode_to_writer inline.

Add flush_to_writer function to match flush_to_writer from
consensus_encoding.

bitcoin: reject 65 bytes signature with sighash 0x00

According to BIP341, if taproot signature is 65 bytes long, the last byte (sighash)
must be different from 0x00, otherwise, it is invalid.
currently, we are accepting it as a valid signature.
this might also break the roundtrip of from_slice -> serialize

ref: https://github.com/bitcoin/bips/blob/master/bip-0341.mediawiki#taproot-key-path-spending-signature-validation

consensus_encoding: add track_caller to panic-able sites

Defensively added to the Decoder trait itself since the performance
impact is negligible.

Re-release without rc suffix

Turns out that the `-rc.0` suffix causes way more problems than it
solves because of how `cargo` resolves the version numbers and what we
intended on using the RC releases for.

In brief

- We wanted to be able to do breaking changes if required
- We wanted to signal that these releases were almost there (TM)
- We wanted to be able to do downstream testing including releasing
downstream crates with the RC releases as part of their public API.

In hindsite we messed up and should have just kept iterating as normal
until we were ready.

Re-release the whole stack without any rc suffix's. However keep
`bitcoin 0.33.0-beta` because we want 0.32.0 to be the latest stable
release and its important that it shows as such on docs.rs

Also, for pre-1.0 crates that had an rc release just jump to the next
version i.e., `io 0.4.0-rc.0` goes to `io 0.5.0`. Just for good
measure.


crate | latest stable | latest RC | with this applied
-----------------------------------------------------------
consensus_encoding 0.0.0 1.0.0-rc.3 0.1.0
units 0.2.0 1.0.0-rc.4 0.3.0
primitives 0.101.0 1.0.0-rc.2 0.102.0
hashes 0.19.0 - 0.20.0
io 0.3.0 0.4.0-rc.0 0.5.0
base58ck 0.3.0 - 0.4.0


bitcoin - 0.33.0-beta.0 to be yanked. Release as 0.33.0-beta

p2p - updated deps, unrelased so no other changes.
internals - not touched (currently 0.5.0)
chacha20_poly1305 - not touched

bitcoin: preserve parity for XOnlyPublicKey

both From impls for XOnlyPublicKey lose parity because they call `from_secp(`,
which defaults to Parity::Even. we now preserve parity by extracting it from
x_only_public_key() instead

consensus_encoding: add conventinal manifest metadata

Follows C-METADATA from the API guidelines.

consensus_encoding: fix up doc link

Confrom to C-LINK API conventions.

Update api

consensus_encoding: remove redundant trait bounds

Fix Unknown NetworkMessage encoding

The NetworkMessageEncoder and old Encodable implementation for
NetworkMessage make use of the Vec<u8> encoding for the raw payload
data. This introduces a vector length prefix to the start of the
encoding which causes the roundtrip for Unknown NetworkMessage types to
fail. In practice, NetworkMessage will only ever be decoded through the
RawNetworkMessage, where a payload length will be known. As such, this
length prefix is not necessary in the encoding and can be trivially
removed.

Change Unknown NetworkMessage encoding to directly write the vector of
bytes instead of using Vec<u8> encoding logic.

Encapsulate PublicKey and PrivateKey

With the previous changes to the PublicKey and PrivateKey types, both
can now be encapsulated to control the scope of access to the inner
fields.

Move PublicKey and PrivateKey to encapsulated module and add them to
the pub use.

Remove public access to all fields of PublicKey and PrivateKey

As with the inner fields, the compressed and network fields of PublicKey
and PrivateKey allows public access for read/write. To allow for
encapsulation, read access should be limited to an accessor function.

Introduce compressed() accessor function for PublicKey and compressed()
and network() accessor functions for PrivateKey. Replace all direct
access to inner fields with use of the accessor functions.