0.10.0 (2026-09-21)

0.10.0 (2026-09-21)

Grouped release for main.

Breaking changes

Raise the Dart and Flutter baseline

Packages: solana_kit, solana_kit_accounts, solana_kit_address, solana_kit_anchor, solana_kit_address_constants, solana_kit_addresses, solana_kit_codecs, solana_kit_codecs_core, solana_kit_codecs_data_structures, solana_kit_codecs_numbers, solana_kit_codecs_strings, solana_kit_fixed_points, solana_kit_errors, solana_kit_fast_stable_stringify, solana_kit_jupiter, solana_kit_instruction_plans, solana_kit_instructions, solana_kit_keys, solana_kit_lints, solana_kit_mpl_core, solana_kit_mpl_token_metadata, solana_kit_options, solana_kit_pyth, solana_kit_program_client_core, solana_kit_programs, solana_kit_rpc, solana_kit_rpc_api, solana_kit_rpc_parsed_types, solana_kit_rpc_spec, solana_kit_rpc_spec_types, solana_kit_rpc_subscriptions, solana_kit_rpc_subscriptions_api, solana_kit_rpc_subscriptions_channel_websocket, solana_kit_rpc_transformers, solana_kit_rpc_transport_http, solana_kit_rpc_types, solana_kit_signers, solana_kit_sns, solana_kit_squads, solana_kit_subscribable, solana_kit_transaction_confirmation, solana_kit_transaction_introspection, solana_kit_transaction_messages, solana_kit_transactions

The workspace now builds against Dart 3.13.3 and Flutter 3.47.4, and every package declares that floor instead of the previous Dart 3.12 range. Consumers on older SDKs can no longer resolve these packages, so this release is breaking even though no Dart API changed.

The Flutter floor rises from 3.44 to 3.47 for solana_kit_mobile_wallet_adapter, solana_kit_mobile_wallet_adapter_protocol, and solana_kit_wallet_adapter, matching the floor solana_kit_wallet_ui already required. solana_kit_lints ships the raised floor to consumers, so it carries the same breaking bump. Every other package raises only the Dart SDK floor.

Raising the language version also switches dart format to the tall style, so 83 files across library, test, script, and Codama-generated trees are reflowed. The renderer pipes generated output through dart format, so regenerating stays consistent.

Align your own SDK constraint with the workspace:

environment:
  sdk: ^3.13.0
  # Omit for pure Dart packages; required for the Flutter packages above.
  flutter: ">=3.47.0"

Owner: Ifiok Jr. · Introduced in: 5f5fe01 · Last updated in: 19932db

Align error code numbers with upstream and report malformed UTF-8 as a code

Packages: solana_kit_errors

SolanaErrorCode numbers now match upstream @solana/kit exactly. Two port-only codes were occupying numbers upstream uses for its UTF-8 codes, which made any cross-SDK comparison of those numbers wrong.

The UTF-8 codec also stops raising a bare FormatException and reports through the error codes upstream defines:

• codecsInvalidUtf8Bytes (8078026) for a malformed byte sequence, carrying the offset where decoding failed.
• codecsInvalidUtf8String (8078027) for a lone surrogate, carrying its index. This is thrown when encoding with fatal: true and, with fatal: false, both directions keep replacing the offending unit with U+FFFD.

Two port-only codes moved or went away:

• codecsInvalidBoolean moved from 8078027 to 8078999. The number had to change because 8078027 is upstream's CODECS__INVALID_UTF8_STRING. This port validates that booleans are encoded as 0 or 1 and upstream does not, so the code has no upstream counterpart and now sits at the end of the codec block, where upstream cannot collide with it. If you match on SolanaErrorCode.codecsInvalidBoolean.value, update the number; matching on the enum member is unaffected.
• codecsStringContainsNullCharacters was removed. Nothing in the workspace threw it, and upstream has no equivalent. Use Utf8CodecConfig.removeNullCharacters to control null handling instead.

try {
  getUtf8Decoder().decode(bytes);
} on SolanaError catch (error) {
  if (error.code == SolanaErrorCode.codecsInvalidUtf8Bytes) {
    print('bad bytes at ${error.context['offset']}');
  }
}

solana_kit_memo, solana_kit_offchain_messages, and solana_kit_attestation_service only had tests asserting the old FormatException for malformed UTF-8; those assertions now check the error code, and solana_kit_memo declares the solana_kit_errors dev dependency that needs.

upstream:error-codes, which also runs as part of docs:check, compares this enum against .repos/kit/packages/errors/src/codes.ts and fails on a number mismatch or an occupied number, so this cannot drift again without CI saying so.

Owner: Ifiok Jr. · Introduced in: a8f643f

Features

Replace stubbed functions with real implementations

Packages: solana_kit, solana_kit_addresses, solana_kit_rpc, solana_kit_transaction_messages

Several public functions promised behavior they did not deliver. Each is now implemented, with the missing API surface added alongside it.

estimateResourceLimitsFactory now simulates

The previous implementation returned its argument unchanged, so it performed no simulation: no compute unit measurement, no loaded accounts data size, no failure reporting. It also could not have worked where it lived, because it needs an RPC client and the transaction compiler, and solana_kit_transaction_messages depends on neither. Upstream defines this function in the umbrella @solana/kit package for the same reason, so it now lives in package:solana_kit with that dependency available.

It takes an EstimateResourceLimitsFactoryConfig holding the RPC client and returns a function that:

• Sets the compute unit limit to the maximum (1400000) and, for version 1 messages, the loaded accounts data size limit to the maximum (67108864) before simulating, so the simulation is not cut short by a resource ceiling.
• Asks the node to replace the blockhash for blockhash-lifetime transactions, and uses the real nonce for durable nonce transactions.
• Returns the unitsConsumed the node reported, capped at the u32 ceiling, plus loadedAccountsDataSize.
• Throws transactionFailedToEstimateComputeLimit when the node reports no compute units, transactionFailedToEstimateLoadedAccountsDataSizeLimit when a version 1 simulation omits the loaded accounts size, and transactionFailedWhenSimulatingToEstimateResourceLimits with the decoded transaction error as cause when the transaction itself fails. All three codes already existed and were never thrown.

Three supporting pieces land with it:

• simulateTransaction and its simulateTransactionValue result are now available on the RPC client. The method was reachable only by hand-assembling a request before this.
• maxLoadedAccountsDataSizeLimit (67108864) is exported.
• estimateAndSetResourceLimitsFactory no longer computes a loaded accounts data size for legacy and version 0 messages. It previously did, which spent an extra simulation and could attach a SetLoadedAccountsDataSizeLimit instruction the runtime ignores. The loaded accounts limit is now only ever set on version 1 messages, matching upstream.

final estimate = estimateResourceLimitsFactory(
  EstimateResourceLimitsFactoryConfig(rpc: rpc),
);
final withLimits = await estimateAndSetResourceLimitsFactory(estimate)(message);

solana_kit_functional removed

The package is gone. Its only utility, the pipe extension, has lived in solana_kit_transaction_messages since the previous breaking release and is re-exported by solana_kit, so the package duplicated what the SDK already provided and existed only as an empty placeholder pending retirement. Anyone still importing it should switch to solana_kit_transaction_messages (or the solana_kit umbrella), which is a one-line import change.

solana_kit_addresses gains the PDA guards

isProgramDerivedAddress and assertIsProgramDerivedAddress were absent, leaving addressesMalformedPda and addressesPdaBumpSeedOutOfRange defined but unreachable. Both are now implemented: they validate that a value is an (Address, int) record, that the bump seed is in [0, 255], and that the address is well formed.

solana_kit_helius builds real smart transactions

createSmartTransaction returned a bare blockhash while documenting that it would estimate compute units and priority fees. It now performs the full sequence: validate, estimate compute units through simulateTransaction, sample the priority fee by account key, resolve the fee in both microLamports-per-unit and total lamports, and refresh the blockhash. It returns a SmartTransaction carrying the limits, fee, lifetime, instructions, fee payer, and account keys; signing stays with the caller because the client holds signer addresses rather than keys.

Two related silent defaults were removed:

• getComputeUnits returned 200000 when the node omitted unitsConsumed. It now throws, because inventing a number sizes the transaction for work the simulation never confirmed. It also reports a failed simulation instead of returning the units of one that did not succeed, and it serializes real Instruction objects, which previously failed at JSON encoding.
• broadcastTransaction, sendTransactionWithSender, and sendSmartTransaction accepted a senderUrl parameter they never used. The parameter is gone; the REST client already targets the sender base URL.

Version 1 durable nonce transactions are recognized

getTransactionLifetimeConstraintFromCompiledTransactionMessage only inspected the legacy instruction list, which a version 1 compiled message leaves empty in favour of separate instruction headers and payloads. Every version 1 durable nonce transaction therefore decompiled as a blockhash transaction, so a caller could not tell that its lifetime depended on a nonce. The version 1 branch now reads the headers and payloads, throws transactionInvalidNonceAccountIndex for an out-of-range nonce account index, and returns the blockhash lifetime only when the first instruction is genuinely not an advance-nonce instruction.

Priority fee lamports API

getTransactionMessagePriorityFeeLamports and setTransactionMessagePriorityFeeLamports add the missing read/write surface for the total-lamport priority fee that only version 1 messages carry. The setter removes the fee on null, drops an emptied config, and is a no-op when the value already matches.

Wallets can now express and check version 1 support

SolanaTransactionVersion gains version1 plus wireValue and fromWireValue, so the values a wallet advertises (legacy, 0, 1) round-trip instead of being collapsed. A supportsVersion1 extension makes the check usable. Two related corrections:

• The browser registry used to map any advertised entry other than legacy onto version 0, so a wallet advertising 1 was reported as version 0 and a caller could build a transaction the wallet cannot sign. Unrecognized entries are now dropped rather than mislabelled.
• The MWA-backed mobile wallet advertises an explicit legacy-and-version-0 list instead of SolanaTransactionVersion.values, which would have silently started claiming version 1 support as the enum grew. This matches upstream's wallet-standard-mobile.

Error codes that were defined but unreachable

Three codes had no throw site. signerWalletAccountCannotSignTransaction is now thrown when a WalletAccountSigner is created for an account advertising neither transaction feature, matching upstream's createSignerFromWalletAccount. heliusApiKeyRequired is thrown by HeliusConfig for a blank key, which previously produced a request that could only fail with a 401. heliusTransactionSimulationFailed replaces a bare StateError when a compute-unit simulation reports a transaction failure.

The remaining defined-but-unthrown codes were checked against upstream and are parity-faithful: upstream defines them without throwing them anywhere either (addressesInvalidBase58EncodedAddress, the four wallet* codes, subscribableRetryNotSupported, transactionInvalidNonceTransactionFirstInstructionMustBeAdvanceNonce), or they belong to abstractions this port intentionally does not have (the React hook path behind signerWalletMultisignUnimplemented, the fs-impl package behind fsUnsupportedEnvironment, the named-channel pubsub plan behind invariantViolationDataPublisherChannelUnimplemented).

solana_kit_dapp_publisher_cli reports unreadable balances

parseLamportsValue returned 0 for a balance response it could not parse. A malformed response therefore looked like an empty wallet. It now throws a FormatException, so a transport or schema change is reported as itself rather than as insufficient funds.

Owner: Ifiok Jr. · Introduced in: 19932db

Track @solana/kit v8.3.0

Packages: solana_kit, solana_kit_codecs_core, solana_kit_codecs_numbers

The workspace now tracks upstream @solana/kit v8.3.0 (previously v8.2.0), and upstream:parity passes against it. This entry maps every change in that upstream release to its Dart counterpart.

Ported in this release:

• u256 and i256 number codecs (getU256Codec, getI256Codec and their encoder/decoder pairs) serialize 32 bytes, honour the endian option, validate the full range on encode, and decode to BigInt, mirroring the existing 64-bit and 128-bit codecs.
• Tap codec helpers observe values or bytes without modifying them: tapEncoder, tapDecoder, and tapCodec observe values, while tapEncoderBytes, tapDecoderBytes, and tapCodecBytes observe raw bytes and offsets. Each wrapper preserves the size characteristics of what it wraps, and any tap may throw, which makes them validation guards that need no identity transformEncoder.

Already present before this release, and now covered by the v8.3.0 claim:

• The getAgGenesisCert RPC method and its allowed numeric keypaths.
• isSolanaRequest recognising getAgGenesisCert and getTransactionsForAddress, which is also what fixed upstream's bigint parsing for getTransactionsForAddress responses.

No Dart change needed:

• HasAddress and the InstructionAccountInput / InstructionSignerInput widening are TypeScript type-level changes. Dart has no structural typing, and this port's ResolvedInstructionAccount already wraps an Object value, so it accepts addresses, address-bearing objects, ProgramDerivedAddress values, and AccountMeta role overrides at runtime without a cast.
• Marking role as readonly on the writable and signer account types is already true here: AccountMeta.role is a final field.
• createLazyKeyPairSignerFromBytes exists upstream to defer an asynchronous WebCrypto key import. Signer creation in this port is synchronous, so there is nothing to defer and the type is not needed.

Verification:

• upstream:parity passes against @solana/kit@8.3.0
• upstream:check reports the metadata is internally consistent for tracked version 8.3.0
• The @solana/kit reference pin moved to tag v8.3.0

Owner: Ifiok Jr. · Introduced in: 220066e

Track the memo v4 program and add memo extraction helpers

Packages: solana_kit_address_constants

The workspace now tracks solana-program/memo at js@v0.14.1 (previously js@v0.13.1). The mpl-token-metadata reference pin also moves to 353d01be4af3; its IDL is byte-identical, so that package is unaffected.

solana_kit_address_constants moves memoProgramAddress to the v4 memo program (Memo4c2pN8afCj432Lb7RMVKi9PbQnnW7ewFFaV3oAH), matching the upstream IDL publicKey as of js@v0.14.0. The previous v3 address stays available as the new memoLegacyProgramAddressV3 constant, and memoLegacyProgramAddress (v1) is unchanged. Code that builds new memo instructions picks up the v4 program automatically; code that must keep targeting v3 names the legacy constant explicitly.

solana_kit_memo ports the upstream extraction helpers from js@v0.14.1:

• getMemosFromInstructions scans a list of instructions, matches every deployed Memo program address, and returns the UTF-8 decoded memo text with the raw bytes, source program address, and instruction index.
• ExtractedMemo carries one extracted memo.
• supportedMemoProgramAddresses lists every deployed Memo program address ordered from oldest (v1) to newest (v4).

The generated layer is regenerated against js@v0.14.1 with the current renderer: the program page is byte-identical because the program address flows through the well-known constants, and the AddMemo data decoder now validates byte length strictly, throwing SolanaError with codecsInvalidByteLength on trailing bytes. solana_kit_errors moves from a dev dependency to a dependency because the generated decoder references it.

codama-renderers-dart maps the v4 address to memoProgramAddress and the v3 address to memoLegacyProgramAddressV3 in its well-known address registry, so regenerated clients re-export the canonical constants instead of hardcoding address strings.

Build new memo instructions against memoProgramAddress (v4). Where a memo must be executed by a program the runtime provides, check what is deployed: SurfPool, used by this repository's on-chain integration tests, ships the v1 and v3 programs as executable bytecode but resolves v4 to a placeholder, so those tests invoke memoLegacyProgramAddressV3. All three programs share the same instruction format, so only the program the transaction targets differs.

Owner: Ifiok Jr. · Introduced in: 9dcb59a

Add 256-bit number codecs and tap codec helpers

Packages: solana_kit_codecs_core, solana_kit_codecs_numbers

Ports the two additive codec surfaces from upstream @solana/kit v8.3.0.

solana_kit_codecs_numbers gains 256-bit integer codecs: getU256Codec, getU256Encoder, and getU256Decoder for unsigned values in [0, 2^256 - 1], plus getI256Codec, getI256Encoder, and getI256Decoder for signed values in [-(2^255), 2^255 - 1]. Both serialize as 32 bytes, honour the endian option, and always decode to BigInt, matching the existing 64-bit and 128-bit codecs.

solana_kit_codecs_core gains tap helpers that observe values or bytes without modifying them. Because any callback may throw, they double as validation guards that need no identity transformEncoder:

final guarded = tapDecoderBytes(getU8Decoder(), (bytes, offset) {
  if (bytes[offset] > 1) throw StateError('Expected a 0 or a 1');
});

final spanned = tapEncoderBytes(getU8Encoder(), (bytes, pre, post) {
  print('wrote ${post - pre} bytes at $pre');
});

tapEncoder, tapDecoder, and tapCodec observe values; tapEncoderBytes, tapDecoderBytes, and tapCodecBytes observe bytes and offsets. Each wrapper preserves the size characteristics of the codec it wraps, so FixedSizeEncoder stays fixed-size and VariableSizeEncoder keeps its maxSize.

Note that Dart's Codec is not an Encoder or a Decoder, so the value-level wrappers are typed against the encoder or decoder they observe; use tapCodec and tapCodecBytes to wrap a codec.

Owner: Ifiok Jr. · Introduced in: 220066e

Add the upstream size-prefix and UTF-8 codec options

Packages: solana_kit_codecs_data_structures, solana_kit_codecs_strings

Ports the last two surfaces from upstream @solana/kit v8.3.0. Both are additive: no existing behavior changes.

requireSizePrefix is a named parameter on getArrayDecoder, getArrayCodec, getSetDecoder, getSetCodec, getMapDecoder, and getMapCodec. Upstream defaults it to false, where an exhausted byte array decodes as an empty collection so a collection can be appended to an existing layout. This port defaults it to true and throws, because a silently empty collection hides truncated input. Pass requireSizePrefix: false to opt in:

final lenient = getArrayCodec(getU8Codec(), requireSizePrefix: false);
lenient.decode(Uint8List(0)); // []
getArrayCodec(getU8Codec()).decode(Uint8List(0)); // throws

Utf8CodecConfig is accepted by getUtf8Encoder, getUtf8Decoder, and getUtf8Codec, and carries upstream's three options:

• fatal rejects malformed input instead of replacing it. Upstream defaults to false, decoding bad bytes as U+FFFD; this port defaults to true and raises a FormatException from decoding and, for lone surrogates, from encoding.
• ignoreBOM preserves a leading byte order mark. Both default to false, which strips it, matching Dart's Utf8Decoder.
• removeNullCharacters strips U+0000 from decoded strings. Upstream defaults to true; this port defaults to false so the decoded value reflects the bytes exactly.

The two divergent defaults exist so that malformed or null-padded account and instruction data cannot decode silently. To take upstream's behavior explicitly:

final codec = getUtf8Codec(
  const Utf8CodecConfig(fatal: false, removeNullCharacters: true),
);

Owner: Ifiok Jr. · Introduced in: cf4ab87

Align error code numbers with upstream and report malformed UTF-8 as a code

Packages: solana_kit_codecs_strings

SolanaErrorCode numbers now match upstream @solana/kit exactly. Two port-only codes were occupying numbers upstream uses for its UTF-8 codes, which made any cross-SDK comparison of those numbers wrong.

The UTF-8 codec also stops raising a bare FormatException and reports through the error codes upstream defines:

• codecsInvalidUtf8Bytes (8078026) for a malformed byte sequence, carrying the offset where decoding failed.
• codecsInvalidUtf8String (8078027) for a lone surrogate, carrying its index. This is thrown when encoding with fatal: true and, with fatal: false, both directions keep replacing the offending unit with U+FFFD.

Two port-only codes moved or went away:

• codecsInvalidBoolean moved from 8078027 to 8078999. The number had to change because 8078027 is upstream's CODECS__INVALID_UTF8_STRING. This port validates that booleans are encoded as 0 or 1 and upstream does not, so the code has no upstream counterpart and now sits at the end of the codec block, where upstream cannot collide with it. If you match on SolanaErrorCode.codecsInvalidBoolean.value, update the number; matching on the enum member is unaffected.
• codecsStringContainsNullCharacters was removed. Nothing in the workspace threw it, and upstream has no equivalent. Use Utf8CodecConfig.removeNullCharacters to control null handling instead.

try {
  getUtf8Decoder().decode(bytes);
} on SolanaError catch (error) {
  if (error.code == SolanaErrorCode.codecsInvalidUtf8Bytes) {
    print('bad bytes at ${error.context['offset']}');
  }
}

solana_kit_memo, solana_kit_offchain_messages, and solana_kit_attestation_service only had tests asserting the old FormatException for malformed UTF-8; those assertions now check the error code, and solana_kit_memo declares the solana_kit_errors dev dependency that needs.

upstream:error-codes, which also runs as part of docs:check, compares this enum against .repos/kit/packages/errors/src/codes.ts and fails on a number mismatch or an occupied number, so this cannot drift again without CI saying so.

Owner: Ifiok Jr. · Introduced in: a8f643f

Fixes

Add scoped mutation testing tooling

Packages: solana_kit

Adds a mutation testing setup built on mutation_test, so the suite can be checked for tests that assert current behavior rather than the intended contract. Coverage cannot detect that failure mode: a test asserting the wrong contract still executes every line.

Four devenv tasks wrap a new driver at scripts/run_mutation_testing.dart:

• mutation:list shows the configured scopes and their sizes.
• mutation:check reports how completely each scope's test list covers its transitive dependents.
• mutation:changed runs only the scopes touched since origin/main.
• mutation:run runs a scope directly, with --scope, --full, and --coverage options.

Scopes live in config/mutation/scopes.json and pair source files with the test directories that can detect a change in them, covering the numeric codecs, transaction messages, transaction envelopes, string codecs, codec core, keys, PDAs, hashing, and the untrusted-input parsers.

A default run executes only a scope's listed tests, which costs about two seconds per mutant. Because shared packages have many dependents (solana_kit_codecs_numbers has 58, solana_kit_codecs_core has 59), a survivor in code other packages exercise can be a false positive; --full expands the run to every dependent so no survivor is. mutation:check reports the size of that gap per scope.

config/mutation/rules.xml replaces the builtin rule set, which is unsuitable for Dart: the builtin < rule also matches the first character of << and the builtin argument rules reorder call arguments, so both produce mutations that fail to compile and are then reported as survivors. The rules here use lookarounds to match genuine comparisons, and exclude comments and string literals.

This is advisory tooling only. It is not wired into CI: equivalent mutants survive legitimately and would make a required check permanently red. See docs/agents/mutation-testing.md.

Owner: Ifiok Jr. · Introduced in: 186ba3d

Restore complete version inventories in package READMEs

Packages: solana_kit_anchor, solana_kit_jupiter, solana_kit_mpl_core, solana_kit_mpl_token_metadata, solana_kit_pyth, solana_kit_sns, solana_kit_squads

The versions.json data source that renders every package README's installation section had drifted: packages first released after the legacy knope era were never added, so their READMEs told consumers to depend on a bare ^ with no version. The retired solana_kit_functional entry and a stale solana_kit_mobile_wallet_adapter_example version were also lingering.

The inventory now matches every package's pubspec.yaml, the solana_kit_functional key is gone, and the affected installation sections render the real published version again:

dependencies:
  "solana_kit_jupiter": ^0.9.3

Owner: Ifiok Jr. · Introduced in: 4c8855b

Verify both hand-written SHA-256 implementations against an independent hash

Packages: solana_kit_anchor, solana_kit_sns

solana_kit_anchor and solana_kit_sns each carry their own pure-Dart SHA-256. Anchor uses it to derive program discriminators and SNS to derive name account addresses, so a defect in either changes which instruction or account a program resolves to — a wrong result rather than an exception. Both were covered only by a handful of fixed NIST vectors, which pin the round function but rarely land on the padding and block boundaries where a length bug hides.

Each package now has a sha256_differential_test.dart comparing its implementation against package:crypto (the same dependency solana_kit_addresses already uses). Coverage includes every input length from 0 to 200, which crosses the 55/56-byte padding transitions and the 64-byte block boundary, lengths from 255 through 10000 for multi-block behavior, randomized inputs, and all-zero and all-0xff buffers. Anchor additionally checks instructionDiscriminator against the truncated namespaced digest, and SNS checks getHashedName against sha256("SPL Name Service" + name).

Both implementations agree with the reference on every case. No library behavior changes.

Owner: Ifiok Jr. · Introduced in: f21cd21

Speed up base-X codecs, BigInt JSON parsing, and fixed-point parsing

Packages: solana_kit_addresses, solana_kit_codecs_strings, solana_kit_fixed_points, solana_kit_rpc_spec_types, solana_kit_transaction_messages

Same results, less work per call. No public API, output byte, or error behavior changed; upstream:parity passes against @solana/kit@8.3.0 and the full workspace suite is green.

The base-X codecs converted through BigInt, scaled the alphabet with String.indexOf and a fresh one-character string per input character, and built output with repeated insert(0, …) calls that reallocate and shift the whole list each time. Encoding a typical base58 address cost roughly 600 µs, so compiling a 20-account transaction spent measurable milliseconds on address encoding alone. The conversion now folds digits into a byte buffer with word-sized carry arithmetic, indexes the alphabet through a cached code-unit lookup, and writes output in order:

// Before: BigInt division per character plus O(n²) list inserts.
// After: one carry pass per character over a preallocated buffer.
final converted = _convertToBytes(value, alphabet);
bytes
  ..fillRange(offset, offset + converted.leadingZeroes, 0)
  ..setAll(offset + converted.leadingZeroes, converted.bytes);

Measured on an interleaved best-of-N benchmark, with the previous implementation running in the same process to cancel machine noise:

Operation	Before	After	Improvement
base58 encode, typical address	22.97 µs	2.88 µs	8.0x
base58 decode, 32 bytes	16.79 µs	3.24 µs	5.2x
base58 encode, 64-byte signature	52.89 µs	8.44 µs	6.3x

parseJsonWithBigInts allocated a one-character string and ran up to two regular expressions per character of the payload, then rebuilt the document character by character. It now scans by code unit, copies literal runs as substrings, and hoists the exponent pattern:

Operation	Before	After	Improvement
parse a 28 KB RPC payload	2.198 ms	1.089 ms	2.0x

Also removed in the same pass: a redundant full copy in the UTF-8 encoder and the base64 decoder (both utf8.encode and base64.decode already return Uint8List), a double copy of every version-1 instruction payload, and two regular expressions that were compiled per parsed value in the fixed-point codecs.

Behavior is pinned by tests rather than assumed. base_x_property_test.dart round-trips random byte strings across every length from 0 to 512 bytes, covers leading-zero-only input, alphabets wider than a byte, and non-power-of-two alphabets, and asserts arbitrary-precision digit strings beyond 64 bits. A degenerate one-character alphabet is now rejected with an ArgumentError when the codec is created; previously it produced silent zero output on encode and looped forever on decode.

bench:all also gains coverage where the regression was invisible: the address benchmark previously measured only the all-ones System Program address, which is base58's leading-zero fast path and exercises no base conversion at all.

Owner: Ifiok Jr. · Introduced in: b22257e

Reject truncated numeric reads with SolanaError instead of RangeError

Packages: solana_kit_codecs_numbers, solana_kit_transaction_messages, solana_kit_transactions

A decoder reading a fixed-width field out of malformed wire data could raise a raw RangeError or IndexError out of the SDK instead of the documented SolanaError, so a caller catching SolanaError saw an escaped exception rather than a rejection it could handle. A truncated account, transaction, or RPC payload reaching the decoders was enough to trigger it; no signature or cluster access was required.

The guard that upstream @solana/kit applies in its number decoder factory was missing from three ports of it. numberDecoderFactory and floatDecoderFactory read through a ByteData view without first checking that the requested width was available, and the six BigInt decoders (u64, i64, u128, i128, u256, i256) indexed bytes directly with the same gap. All eight now call assertByteArrayIsNotEmptyForCodec and assertByteArrayHasEnoughBytesForCodec before reading, raising codecsCannotDecodeEmptyByteArray or codecsInvalidByteLength as upstream does. A new bigIntDecoderFactory carries the guard for the multi-word widths.

Two further escape sites in solana_kit_transaction_messages are fixed. The version 1 instruction payload slice computed pos + numInstructionDataBytes and sliced without confirming the buffer held that many bytes; it now asserts the length the way upstream's fixDecoderSize wrapper does. The transaction version decoder read bytes[offset] unguarded — upstream tolerates this because JavaScript yields undefined, which then silently takes the legacy branch and misreports a truncated buffer as an unversioned message, so this port rejects the empty buffer rather than reproducing that fallback.

Encoders are unchanged and still raise RangeError when a destination buffer is too small. Upstream writes into a scratch buffer and then bytes.sets it, which throws in JavaScript too, so that behavior is deliberate parity rather than a defect.

Callers that caught RangeError around a decode should catch SolanaError. Callers that already caught SolanaError now see malformed input rejected where it previously surfaced as a crash.

Owner: Ifiok Jr. · Introduced in: f21cd21

Track @solana/kit v8.3.0

Packages: solana_kit_errors

The workspace now tracks upstream @solana/kit v8.3.0 (previously v8.2.0), and upstream:parity passes against it. This entry maps every change in that upstream release to its Dart counterpart.

Ported in this release:

• u256 and i256 number codecs (getU256Codec, getI256Codec and their encoder/decoder pairs) serialize 32 bytes, honour the endian option, validate the full range on encode, and decode to BigInt, mirroring the existing 64-bit and 128-bit codecs.
• Tap codec helpers observe values or bytes without modifying them: tapEncoder, tapDecoder, and tapCodec observe values, while tapEncoderBytes, tapDecoderBytes, and tapCodecBytes observe raw bytes and offsets. Each wrapper preserves the size characteristics of what it wraps, and any tap may throw, which makes them validation guards that need no identity transformEncoder.

Already present before this release, and now covered by the v8.3.0 claim:

• The getAgGenesisCert RPC method and its allowed numeric keypaths.
• isSolanaRequest recognising getAgGenesisCert and getTransactionsForAddress, which is also what fixed upstream's bigint parsing for getTransactionsForAddress responses.

No Dart change needed:

• HasAddress and the InstructionAccountInput / InstructionSignerInput widening are TypeScript type-level changes. Dart has no structural typing, and this port's ResolvedInstructionAccount already wraps an Object value, so it accepts addresses, address-bearing objects, ProgramDerivedAddress values, and AccountMeta role overrides at runtime without a cast.
• Marking role as readonly on the writable and signer account types is already true here: AccountMeta.role is a final field.
• createLazyKeyPairSignerFromBytes exists upstream to defer an asynchronous WebCrypto key import. Signer creation in this port is synchronous, so there is nothing to defer and the type is not needed.

Verification:

• upstream:parity passes against @solana/kit@8.3.0
• upstream:check reports the metadata is internally consistent for tracked version 8.3.0
• The @solana/kit reference pin moved to tag v8.3.0

Owner: Ifiok Jr. · Introduced in: 220066e

Raise the Dart and Flutter baseline

Packages: solana_kit_test_matchers

The workspace now builds against Dart 3.13.3 and Flutter 3.47.4, and every package declares that floor instead of the previous Dart 3.12 range. Consumers on older SDKs can no longer resolve these packages, so this release is breaking even though no Dart API changed.

The Flutter floor rises from 3.44 to 3.47 for solana_kit_mobile_wallet_adapter, solana_kit_mobile_wallet_adapter_protocol, and solana_kit_wallet_adapter, matching the floor solana_kit_wallet_ui already required. solana_kit_lints ships the raised floor to consumers, so it carries the same breaking bump. Every other package raises only the Dart SDK floor.

Raising the language version also switches dart format to the tall style, so 83 files across library, test, script, and Codama-generated trees are reflowed. The renderer pipes generated output through dart format, so regenerating stays consistent.

Align your own SDK constraint with the workspace:

environment:
  sdk: ^3.13.0
  # Omit for pure Dart packages; required for the Flutter packages above.
  flutter: ">=3.47.0"

Owner: Ifiok Jr. · Introduced in: 5f5fe01 · Last updated in: 19932db

Add version 1 transaction coverage with oversized payloads

Packages: solana_kit_transactions

Adds regression tests for the version 1 transaction wire format introduced by SIMD-0296 and SIMD-0385, which raise the transaction size ceiling from 1232 to 4096 bytes.

packages/solana_kit_transactions/test/v1_transaction_test.dart pins the exact bytes against vectors generated by upstream @solana/kit 8.3.0, so a regression in the v1 envelope, config mask, or config value encoding fails loudly instead of silently producing transactions the cluster rejects. The vectors live in test/fixtures/v1_wire_vectors.json, alongside a README documenting how to regenerate them; the largest asserts a 3216-byte transaction byte-for-byte, which no legacy transaction can represent.

On-chain coverage lands in packages/solana_kit_integration_tests/test/integration/v1_transaction_test.dart. It submits a v1 transaction with a 1600-byte memo — over the legacy ceiling — and asserts the node accepts it, recovers the full memo, reports the inline transactionConfig with the requested compute unit and loaded accounts limits, and moves lamports. The same suite asserts the node rejects a payload above 4096 bytes.

A second on-chain suite, resource_limit_estimation_test.dart, submits a v1 transaction whose limits were produced by estimateResourceLimitsFactory and asserts the node executes it.

IntegrationTestEnv gains sendV1Instructions and buildV1WireTransaction helpers, and fetchTransaction now requests maxSupportedTransactionVersion: 1 so v1 responses can be read back rather than failing with RPC error -32015. No public library behavior changes.

Owner: Ifiok Jr. · Introduced in: 19932db