v0.9.0

All 19 published crates are on crates.io at version 0.9.0:

shipstern, shipstern-core, shipstern-proto, shipstern-proc-macro, shipstern-parser, shipstern-mock, shipstern-stream, shipstern-bpf-loader-parser, shipstern-spl-token-parser, shipstern-spl-token-extensions-parser, shipstern-block-meta-parser, shipstern-slot-parser, shipstern-stake-pool-parser, shipstern-solana-rpc-source, shipstern-yellowstone-grpc-source, shipstern-yellowstone-fumarole-source, shipstern-solana-snapshot-source, shipstern-jetstream-source, shipstern-block-coordinator.

See CHANGELOG.md for the full history.

Changed

• Breaking for code that matches on PrefilterError or builds a prefilter with a struct literal. PrefilterError gained variants for the account-filter and data-slice checks, and AccountPrefilter, TransactionPrefilter and SlotPrefilter each gained fields. Neither type is #[non_exhaustive], so an exhaustive match on the error and any AccountPrefilter { accounts, owners } literal stop compiling; add a wildcard arm and ..Default::default() respectively. Builder users are unaffected (#303).
• Bumped yellowstone-grpc-proto from 12.4 to 12.6, picking up the Transaction V1 (SIMD-0385) Message.config field and the cuckoo_account_include transaction filter. The new filter is not exposed through TransactionFilter and is always sent as None (#304 by @ultrasilicon).
• shipstern-jetstream-source: a parser registered on the block pipeline whose prefilter sets block_meta or slot but none of include_transactions / include_accounts / include_entries no longer receives a Block update. Previously any block_meta/slot prefilter forced its filter ID into the block match list, so such a parser received a header-only Block: transactions and accounts were always empty. Use a block_meta pipeline instead.

Deprecated

• shipstern-proc-macro: the cpi_event_discriminator and cpi_event_payload_offset macro arguments are ignored, with a deprecation warning at the call site, when the IDL declares a CPI event envelope. They still apply to IDLs that declare none. Both arguments are removed in 0.10 (#299 by @senzenn).

Added

• The Yellowstone gRPC source replaces the live subscription mid-stream, so a changing filter set no longer costs a reconnect. Runtime::handle hands out a cheap Clone RuntimeHandle before run consumes the runtime; filters reports the last set handed to the source, update_filters patches it under a lock, send_filter_update replaces it wholesale, and reset_filters restores the registered set. The handle exists only for sources implementing FilterUpdateSource, which today is gRPC alone, so taking one from any other source is a compile error rather than a runtime one. Keys are parser IDs, except for instruction parsers, which the runtime bundles behind a single InstructionPipeline::ID entry. A set naming a parser with no registered pipeline is refused with FilterUpdateError::UnknownParser before anything is sent, and Ok(()) means the set reached the source rather than that the server applied it (#302 by @senzenn, #305 and #307 by @kespinola).
• Filters gained get, parser_ids, insert, merge, and remove, keyed by parser ID, so the next subscription can be built without reaching into the map by hand (#305 by @kespinola).
• Prefilter can express the rest of Yellowstone's subscription surface, which impl From<Filters> for SubscribeRequest previously hardcoded to empty. AccountPrefilter gained filters (memcmp, data size, token-account state, and lamport comparisons, via the new AccountFilter, MemcmpData and LamportsCmp types) and nonempty_txn_signature; TransactionPrefilter gained accounts_exclude, vote and signature; SlotPrefilter gained interslot_updates; and the Yellowstone gRPC source config gained accounts-data-slice, which is request-level rather than per-parser. Builder setters check account filters against the rules the server enforces when it builds its own filter (memcmp encoding and size, filter count, at most one data size, no token-account-state = false), so a bad config fails at build time instead of taking the subscription down when it is opened; YellowstoneGrpcConfig::validate checks the data-slice windows for order and overlap, and the source calls it before it dials. Every field defaults to the previous behaviour. transactions_status and entry stay unsupported: they are separate subscription kinds with no pipeline behind them, so nothing could consume what they delivered (#303).
• shipstern-core: InstructionShared gained transaction_config: Option<TransactionConfig>, the inline compute budget that Transaction V1 (SIMD-0385) carries on the message instead of in top-level ComputeBudget instructions. It exposes the priority fee, compute-unit limit, loaded-accounts data-size limit, and heap size. It is None for Legacy and V0 transactions and for jetstream-source transactions, which are decoded from an SDK without a V1 message variant (#304 by @ultrasilicon).
• shipstern-proc-macro: generated account types and the Instructions wrapper now expose DISCRIMINATOR and DISCRIMINATOR_OFFSET constants, so consumers can build memcmp filters and route on discriminators without re-deriving the bytes from the IDL. The pair is a memcmp predicate: the bytes the parser compares at that offset. It is not a payload boundary, because numeric discriminators (for example SPL Governance) are re-read as the first field of the account body, so decoding still goes through try_unpack. No constant is emitted where the parser could not honor it: size-only discriminators, zero-length discriminators, fixed-size fields whose declared width disagrees with the decoded default bytes, and instruction names that collide after case folding (#294).
• shipstern-proc-macro: the CPI event envelope is read from the IDL. An IDL that declares the self-CPI envelope on eventNode.discriminators, the form Carbon and @codama/renderers-vixen-parser emit, previously failed to build with an event discriminator collision, because every event was keyed on the first discriminator in its chain, which for an enveloped IDL is the shared envelope tag. EVENT_PAYLOAD_OFFSET had the mirror problem: it came from the macro arguments and defaulted to 8, so the CPI path always stripped a fixed byte count and only happened to be right for a standard Anchor tag. The macro now infers the envelope from the chain (a constant at offset 0 is the tag, the next discriminator's declared offset is the payload offset), checks that every declaring event agrees on both, rebases the remaining discriminators so the CPI and log paths share one layout, and ANDs a chain of two or more. Padded layouts round-trip because the payload offset is the declared offset rather than the tag length. Chains that cannot be honored fail the build instead of parsing arbitrarily: two constants at offset 0, overlapping ranges, a mix of constant and non-constant nodes behind a tag, an event whose only discriminator is the tag, or discriminator bytes that do not decode, which previously panicked the macro. A build-time guard also rejects an envelope tag that prefixes an instruction discriminator at offset 0 in either direction, because the parser filters any instruction whose data starts with the tag before dispatch; it checks the tag actually in use, whether from the IDL, the macro arguments, or the Anchor default. Instruction parsing is untouched (#299 by @senzenn).

Fixed

• shipstern-jetstream-source: emit SubscribeUpdateBlockMeta and SubscribeUpdateSlot for filters that request them. block_meta and slot prefilters were previously bucketed into the block match list, so their filter IDs rode on UpdateOneof::Block. The runtime dispatches by variant, so those IDs matched no pipeline and were dropped with only a trace! line: block-meta and slot parsers ran to completion having handled zero updates, and shipstern-block-coordinator, which keys on BlockMeta, received none of it on the historical-replay path (delivering it is a prerequisite for driving that crate from replay, not on its own sufficient). Both messages are built from fields already carried on BlockData::Block. Replayed slots report SLOT_FINALIZED, since Old Faithful archives hold only finalized history and cannot reproduce the processed/confirmed transitions that SlotPrefilter::filter_by_commitment = false asks for.

What's Changed

• ci: bust cargo cache keys after the org transfer by @senzenn in #298
• feat(proc-macro): expose discriminator constants on generated types by @senzenn in #295
• feat(proc-macro): read the CPI event envelope from the IDL by @senzenn in #299
• feat(core): expose Transaction V1 inline budget (SIMD-0385) by @ultrasilicon in #304
• docs(changelog): release notes for Transaction V1 inline budget (#304) by @kespinola in #306
• chore(deps): bump tar from 0.4.45 to 0.4.46 by @dependabot[bot] in #290
• chore(deps): bump serde_with from 3.16.1 to 3.22.0 by @dependabot[bot] in #291
• chore(deps): bump rpassword from 7.4.0 to 7.5.4 by @dependabot[bot] in #292
• feat(runtime): update gRPC filters on a live subscription by @senzenn in #302
• fix(jetstreamer-source): deliver block meta and slot updates to their pipelines by @senzenn in #287
• feat(core): implement the remaining SubscribeRequest filters by @senzenn in #309
• chore(release): 0.9.0 by @kespinola in #312

New Contributors

• @ultrasilicon made their first contribution in #304
• @dependabot[bot] made their first contribution in #290

Full Changelog: v0.8.0...v0.9.0