Skip to content

Releases: lightningdevkit/ldk-garbagecollected

v0.0.104.0

21 Dec 17:54
d2925e2
Compare
Choose a tag to compare
  • Updates to LDK version 0.0.104. See
    https://github.com/lightningdevkit/rust-lightning/releases/tag/v0.0.104
    for the full changelog.
  • Use Reference.reachabilityFence to work around Java's somewhat
    poorly documented memory model. This seems to be the intended use so
    its likely this will fix the early finalization bugs we've seen, but
    formal documentation on the Java memory model appears to be lacking
    so its possible (albeit unlikely) that more may be required later.
    This bumps our required Java version from 1.8 to 9.
  • Exposes Hash and Eq via Java's hashCode and equals.
  • Thows an exception instead of incorrect operation if multiple
    ChannelMonitors for the same channel are provided to a
    ChannelManagerConstructor.

v0.0.103.1

03 Dec 16:34
6d094e7
Compare
Choose a tag to compare

Based on LDK v0.0.103 and LDK-C-Bindings v0.0.103.1, this fixes a number
of minor (and a few major) issues in the Java bindings.

  • Exposes the full Logger interface to Java, with information on
    severity and module which originated the log call.
  • Expose read on MaybeReadable objects (like Event) which may or
    may not read an object which was written (some Events are not
    written to disk).
  • Include documentation from Rust on enum variants.
  • Allow specifying a UserConfig object in both
    ChannelManagerConstructor constructors.
  • Check array lengths for constant-length arrays (like keys and block
    headers) prior to calling into Rust. This avoids undefined behavior
    (often a crash) when a byte array of incorrect length is passed.
  • Provide a new is_ok method on Result types instead of requiring
    matching of the object's type.
  • Include missing functions on 3-tuples.
  • Resolve a rare use-after-free when returning an object from a Java
    interface method that was called from Rust.
  • Increase pointer validity checking in leaktracking builds.
  • Resolve a NullPointerException when calling trait methods on a Rust
    instance of a trait which was wrapped in an Option.

v0.0.103.0

04 Nov 06:10
99d1a3b
Compare
Choose a tag to compare

Upstream release notes at https://github.com/rust-bitcoin/rust-lightning/releases/tag/v0.0.103 changes in Java bindings are related to new upstream APIs though not insignificant.

v0.0.102.0

19 Oct 01:52
db55d4e
Compare
Choose a tag to compare

Java-specific changes in this release are minimal. For upstream release notes see https://github.com/rust-bitcoin/rust-lightning/releases/tag/v0.0.102

v0.0.101.3

10 Oct 04:53
e6173a7
Compare
Choose a tag to compare

A few minor fixes:

  • Slightly faster networking (with less GC overhead), relying on some minor reflection to work around Java's overly naive package system,
  • Android 11 fixes
  • Bumps Android API to 24 which was required but not correctly marked

v0.0.101.2

28 Sep 06:55
8c0814b
Compare
Choose a tag to compare

This release fixes a number of under-the-hood bugs, but has no user-visible API changes.

  • Fixes a bug where strings passed from Rust by reference to Java may cause double-frees. This is visible occasionally as Events which contain strings passed to Java handle_event methods may cause crashes.
  • Tweak library paths in Android AARs to match the NDK documentation. This may fix shared library load failures on arm64 devices, though its a bit of a shot in the dark.
  • Avoid introducing reference loops for non-trait objects passed to Rust when the object is cloned before being passed. Historically, the bindings generation attempted to aggressively avoid clones in order to avoid significant performance penalties. While we should move our way back there, clones ended up being added over time to fix specific edge cases. As long as the clones are present, there is no reason to add the explicit references via the ptrs_to list, which may cause effective memory leaks.

v0.0.101.1

27 Sep 06:32
7aa095f
Compare
Choose a tag to compare

This release rewrites our tuple handling to avoid Java generics as suggested in #27. By using more common codepaths, this cuts known memory leaks as much as half, depending on the workload. That said, because its relatively fresh mapping logic, errors may occur.

It also fixes an issue in v0.0.101.0 where constructing a NetGraphMsgHandler from serialized NetworkGraph data was not possible.

v0.0.101.0

25 Sep 22:50
Compare
Choose a tag to compare

Upstream release notes:

API Updates

  • Custom message types are now supported directly in the PeerManager,
    allowing you to send and receive messages of any type that is not natively
    understood by LDK. This requires a new type bound on PeerManager, a
    CustomMessageHandler. IgnoringMessageHandler provides a simple default
    for this new bound for ignoring unknown messages (#1031, #1074).
  • Route graph updates as a result of failed payments are no longer provided as
    MessageSendEvent::PaymentFailureNetworkUpdate but instead included in a
    new field in the Event::PaymentFailed events. Generally, this means route
    graph updates are no longer handled as a part of the PeerManager but
    instead through the new EventHandler implementation for
    NetGraphMsgHandler. To make this easy, a new parameter to
    lightning-background-processor::BackgroundProcessor::start is added, which
    contains an Optional NetGraphmsgHandler. If provided as Some, relevant
    events will be processed by the NetGraphMsgHandler prior to normal event
    handling (#1043).
  • NetworkGraph is now, itself, thread-safe. Accordingly, most functions now
    take &self instead of &mut self and the graph data can be accessed
    through NetworkGraph.read_only (#1043).
  • The balances available on-chain to claim after a channel has been closed are
    now exposed via ChannelMonitor::get_claimable_balances and
    ChainMonitor::get_claimable_balances. The second can be used to get
    information about all closed channels which still have on-chain balances
    associated with them. See enum variants of ln::channelmonitor::Balance and
    method documentation for the above methods for more information on the types
    of balances exposed (#1034).
  • When one HTLC of a multi-path payment fails, the new field all_paths_failed
    in Event::PaymentFailed is set to false. This implies that the payment
    has not failed, but only one part. Payment resolution is only indicated by an
    Event::PaymentSent event or an Event::PaymentFailed with
    all_paths_failed set to true, which is also set for the last remaining
    part of a multi-path payment (#1053).
  • To better capture the context described above, Event::PaymentFailed has
    been renamed to Event::PaymentPathFailed (#1084).
  • A new event, ChannelClosed, is provided by ChannelManager when a channel
    is closed, including a reason and error message (if relevant, #997).
  • lightning-invoice now considers invoices with sub-millisatoshi precision
    to be invalid, and requires millisatoshi values during construction (thus
    you must call amount_milli_satoshis instead of amount_pico_btc, #1057).
  • The BaseSign interface now includes two new hooks which provide additional
    information about commitment transaction signatures and revocation secrets
    provided by our counterparty, allowing additional verification (#1039).
  • The BaseSign interface now includes additional information for cooperative
    close transactions, making it easier for a signer to verify requests (#1064).
  • Route has two additional helper methods to get fees and amounts (#1063).
  • Txid and Transaction objects can now be deserialized from responses when
    using the HTTP client in the lightning-block-sync crate (#1037, #1061).

Bug Fixes

  • Fix a panic when reading a lightning invoice with a non-recoverable
    signature. Further, restrict lightning invoice parsing to require payment
    secrets and better handle a few edge cases as required by BOLT 11 (#1057).
  • Fix a panic when receiving multiple messages (such as HTLC fulfill messages)
    after a call to chain::Watch::update_channel returned
    Err(ChannelMonitorUpdateErr::TemporaryFailure) with no
    ChannelManager::channel_monitor_updated call in between (#1066).
  • For multi-path payments, Event::PaymentSent is no longer generated
    multiple times, once for each independent part (#1053).
  • Multi-hop route hints in invoices are now considered in the default router
    provided via get_route (#1040).
  • The time peers have to respond to pings has been increased when building
    with debug assertions enabled. This avoids peer disconnections on slow hosts
    when running in debug mode (#1051).
  • The timeout for the first byte of a response for requests from the
    lightning-block-sync crate has been increased to 300 seconds to better
    handle the long hangs in Bitcoin Core when it syncs to disk (#1090).

Serialization Compatibility

  • Due to a bug in 0.0.100, Events written by 0.0.101 which are of a type not
    understood by 0.0.100 may lead to Err(DecodeError::InvalidValue) or corrupt
    deserialized objects in 0.100. Such Events will lead to an
    Err(DecodeError::InvalidValue) in versions prior to 0.0.100. The only such
    new event written by 0.0.101 is Event::ChannelClosed (#1087).
  • Payments that were initiated in versions prior to 0.0.101 may still
    generate duplicate PaymentSent Events or may have spurious values for
    Event::PaymentPathFailed::all_paths_failed (#1053).
  • The return values of ChannelMonitor::get_claimable_balances (and, thus,
    ChainMonitor::get_claimable_balances) may be spurious for channels where
    the spend of the funding transaction appeared on chain while running a
    version prior to 0.0.101. Balance information should only be relied upon
    for channels that were closed while running 0.0.101+ (#1034).
  • Payments failed while running versions prior to 0.0.101 will never have a
    Some for the network_update field (#1043).

In total, this release features 67 files changed, 4980 insertions, 1888
deletions in 89 commits from 12 authors, in alphabetical order:

  • Antoine Riard
  • Devrandom
  • Galder Zamarreño
  • Giles Cope
  • Jeffrey Czyz
  • Joseph Goulden
  • Matt Corallo
  • Sergi Delgado Segura
  • Tibo-lg
  • Valentine Wallace
  • abhik-99
  • vss96

v0.0.100.1

22 Aug 04:54
Compare
Choose a tag to compare

Only change from v0.0.100.0 is a fix for Rust Zero-Sized-Types always being null when created from Java. This specifically appeared for the IgnoringMessageHandler.

v0.0.100.0

18 Aug 04:53
Compare
Choose a tag to compare
v0.0.100.0