Conversation
…the docs
Works through the solutions the closed issue templates describe, and fixes what
verifying them surfaced. The unit suite went from 57 failures across 15 suites to
449 passing across 35; lint, typecheck, build, size budget and the package smoke
test are all green.
Correctness
- useSendPayment and useAddTrustline built their Horizon client from the network
*name* rather than the resolved config, so a custom horizonUrl was ignored on
exactly the two paths that move value. Every other hook already passed the
config (bug-04).
- StellarProvider rebuilt its context value on every render, re-rendering every
consumer in the tree regardless of what changed. Memoized, along with the
autoConnect options it derives (bug-02).
- toSubmissionError re-tested Horizon result codes with a stale copy of the table
fromResultCodes already owns, so tx_bad_seq and tx_insufficient_fee were named
on the rejection path but flattened to TRANSACTION_FAILED when Horizon answered
200-with-failure. It now delegates to the one table.
- useSorobanContract re-wrapped its error on every render, handing consumers a new
object each time and re-firing any useEffect(..., [error]) indefinitely. Its
cache key now also distinguishes spec-aware from raw calls, which decode through
different paths and must not share an entry.
- A Soroban argument outside Number.MAX_SAFE_INTEGER is refused with a message
naming the precision loss rather than the generic width question.
- freighterAdapter imports the package's default export. @stellar/freighter-api is
a minified CommonJS bundle whose named exports Node's ESM loader cannot detect,
so `import { WatchWalletChanges }` threw "Named export not found" at load time
for every ESM consumer. The bundled build had hidden this.
Test infrastructure (test-03, test-04)
- The SDK double now re-exports the real, deterministic SDK — Asset, Operation,
Memo, Networks, BASE_FEE, StrKey, Keypair, TransactionBuilder — and fakes only
Horizon.Server and SorobanRpc.Server, through per-test factories rather than a
shared mutable singleton. A full payment now builds, signs and round-trips its
XDR through the real encoder, which was impossible before.
- Six fixture addresses were not valid strkey; the real SDK rejects them. Replaced
with freshly generated testnet keys throughout src and docs.
- jest.config.js dropped resetMocks/restoreMocks, which stripped the module-scope
mock implementations each file defines once.
- useBalance.test.tsx compiled to the same output name as useBalance.test.ts, so
TypeScript skipped emitting one of them; renamed to useBalance.watch.test.tsx.
- useAnchor's SSR test deleted global.window before renderHook, which react-dom
cannot survive, and the leaked global failed the next three tests. It mocks
isBrowser instead, as ssr-guard.test.tsx already did.
- useSendPayment.504's jest.mock sat inside describe(), where it ran after the
file's imports had already resolved and therefore never applied.
- Refetch tests asserted the pre-state-02 behaviour of blanking data on a failed
refresh; they now assert stale-while-revalidate. A failed *first* fetch still
yields null, and that distinction is asserted separately.
Packaging (pkg-01, pkg-02, pkg-04)
- exports declares separate types for the import and require conditions, so ESM
consumers resolve index.d.mts instead of a CommonJS declaration file, and
./package.json is exported.
- The wallet SDKs are external instead of inlined; they were shipped twice, once
bundled and once installed, which stopped consumers deduping or overriding them.
- Dropped @lobstrco/signer-extension-api: imported nowhere, backing an explicitly
unsupported adapter, downloaded by everyone. Dropped the root's duplicate
@albedo-link/intent.
- src is published so the shipped source maps resolve; target and engines.node
state one support floor. treeshake stays off deliberately — its rollup pass
strips the "use client" banner the package cannot ship without.
- The smoke test now typechecks under bundler and node16 as well as legacy node;
only the modern algorithms consult the exports map at all.
CI (ci-07, test-02)
- Restored --frozen-lockfile in the test job and the release workflow.
- The integration workflow filtered on a package name that does not exist, so pnpm
matched nothing and exited 0 — a green check certifying that no test ran. It
also ordered setup-node before pnpm/action-setup, so its cache could not work.
Repo and docs (repo-01)
- Untracked the committed junk: a file named "et --hard HEAD@{5}", dev-server
logs, a misspelled CLUADE.MD, one-off notes, personal scripts, a tsup temp
artifact, and the raw Figma dumps. Rewrote .gitignore to cover them plus the
smoke test's leftovers.
- Both READMEs list all 20 exported hooks (10 were undocumented), the full
provider prop set, the cache and its stale-while-revalidate contract, and the
fee strategy. CHANGELOG's Unreleased section moved to the top and its
LEDGER_OUT_OF_RELENTION typo corrected to match the exported code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
A Windows checkout with core.autocrlf=true rewrote the working tree to CRLF, so git reported ~90 files as modified when only a handful had real changes. `* text=auto eol=lf` normalizes the repository so the next contributor's diff shows what they actually changed. Also marks binaries so git never rewrites them, and pnpm-lock.yaml as generated so it stays out of diffs and language stats. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
The merge wave that landed on `dev` resolved conflicts by concatenating both sides. Nine files ended up holding two or three implementations at once and no longer parsed at all, so `tsc`, `prettier` and `jest` each failed before they could report anything useful. Repaired, keeping the newest architecture in each case: - types/index.ts — two interfaces never closed before the next declaration, the Asset type and a doc comment were duplicated (orphaning a `*/`), plus a stray brace and a leftover `refetch` fragment. - errors/codes.ts — a stray `o` prefixed to SEP10_VALIDATION_FAILED and a duplicate closing brace. - useOffers.ts — an entire dead useState implementation, importing from a `../providers/StellarProvider` path that does not exist, sat above the current one. Restored to b0fe56f, which is byte-identical to the surviving half. - useAccount.ts / useClaimableBalance.ts — the old useState half was spliced through the middle of the useQuery one. Rebuilt from the live halves. Both declare `isStale` in their return type but the surviving code had dropped it, so it is reinstated from the documented stale-while-revalidate contract. - usePayments.ts / useTransactionHistory.ts — three parallel branches (reducer pagination a38f351, the limit+1 hasNext heuristic 8339135, and 429 backoff 581b65b) were never actually integrated; no single revision ever contained all three. Took the reducer architecture as the base and ported the limit+1 heuristic and `maxRetries` onto it. limit+1 matters because `records.length >= limit` reports hasNext incorrectly whenever the total is an exact multiple of the page size. - usePayments.test.tsx — restored to the revision matching that architecture. - __mocks__/@stellar/stellar-sdk.ts — two independent solutions to the same test-04 issue had been concatenated. Kept 25b8f66, the superset: it exports everything the other did except TESTNET_ISSUER/resetMockServers, which nothing references. This commit only restores parseability. 223 type errors of the same origin remain across 53 files and are not addressed here. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
StellarNetwork had reverted to a two-network union while NETWORK_PASSPHRASES, NETWORK_CONFIGS and getNetworkPassphrase all still assumed futurenet and custom. StellarContextValue had lost queryStore and autoConnect, which alone accounted for 66 'Property does not exist' errors across the hooks. AutoConnectOptions had gone entirely, and fourteen exported types (Anchor*, ContractEvents*, Payment paths, path payment, WalletNetworkId) were missing while src/index.ts still re-exported them. Also deduplicated UseOffersReturn/UseOffersOptions and UseManageOfferReturn, which two branches had each declared with incompatible shapes; kept whichever matches the surviving implementation in each case. Corrected two import paths in the barrel (useManageOffer -> useManagerOffer, useOrderbook -> useOrderBook, the latter a casing collision that breaks case-sensitive filesystems). useBalance.ts had the old useState implementation spliced through the middle of the useQuery one, including two 'refetch' keys in its returned object. 223 -> 122 type errors. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
…r existed
useLiquidityPool, useLiquidityPoolActions and useManagerOffer all imported
`useStellar` from `../providers/StellarProvider` — a module that does not exist
in this repository — and destructured `{ server, adapter, publicKey }` from it.
They were merged in without ever compiling.
Rewired to the real `useStellarContext`, deriving the Horizon client from
`networkConfig` and the signer from `getWalletAdapter(wallet.wallet)`. The two
writing hooks now follow the same build/sign/submit path as useAddTrustline:
the adapter signs an XDR string rather than an operation object, and the fee is
bid from the network's current base fee.
Also fixed along the way:
- useLiquidityPool had the literal token `ts` pasted into its finally block.
- manageBuyOffer takes `buyAmount`, not `amount`; the two operations are not
interchangeable and the shared param object silently produced the wrong one.
- Added the missing `isLiquidityPoolShares` guard to utils. It was imported but
never existed; pool shares cannot be either side of an offer, so the paths
that build operations need to reject them rather than fall through to the
issued-asset branch.
- Restored networkPassphrase on NetworkConfig/CustomNetworkConfig and
walletNetworkPassphrase on WalletState, and retyped walletNetwork as
WalletNetworkId — the extension can report a network this library ships no
config for.
122 -> 85 type errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
- SendPaymentOptions and AddTrustlineOptions lost `extends FeeOptions`, so the `fee`/`feeMultiplier` they document could not be passed to resolveFee. - ContractCallOptions lost `spec` and `sourceAccount`, both of which useSorobanContract reads and its tests exercise. - UsePaymentHistoryOptions/Return lost maxAccumulationPages and accumulationBoundHit, which the hook sets and its tests assert on. - LOW_RESERVE had a default message but was never added to STELLAR_ERROR_CODES. - getWalletAdapter was imported from ../utils in six files; it lives in ../wallets. - useFriendBot.test.ts and useOrderBook.test.tsx imported their subjects with the wrong filename casing, which resolves on Windows and fails on Linux. - Asset narrowing: several sites tested `!== "XLM"` and then read `.code`, but Asset also includes the pool-share pseudo-asset, so that never narrowed to IssuedAsset. They now use the isIssuedAsset guard, which also rejects pool shares from order books and payment filters, where they are not valid. - The paginated hooks passed `toStellarError(err)` straight into a reducer action typed StellarError, but it returns null for an abort. An abort is a deliberate cancellation, so those paths now leave page state untouched instead of coercing null into an error. 85 -> 54 type errors. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
WalletType had been narrowed to a closed union, but registerWalletAdapter() exists precisely so an application or wallet vendor can add its own; restored the open form so custom adapters typecheck again. SDK alignment (these hooks had never compiled against the real SDK): - WalletAdapter.signTransaction takes (xdr, options); useSorobanWrite, useSep10Auth and useCreateAccount each called it with three positional args. - SendTransactionResponse exposes errorResult, not errorResultXdr. - GetTransactionStatus has no PENDING member. NOT_FOUND is how the RPC reports "not yet in a ledger", which is what the polling loop should test. - submitTransaction's response carries no created_at. - base_reserve_in_stroops is already a number. useSendPayment only ever built text memos, so the MemoInput union added by issue-202 could not actually be used. It now builds the right memo per type, which matters on the wire: an exchange expecting an id memo will not credit a text memo holding the same digits. usePayments.test.tsx did not contain tests at all. The merge had written a copy of useTransactionHistory.ts source to that path. Restored the real suite from 8339135, whose hasNext expectations match the limit+1 heuristic. Removed src/hooks/__tests__/useOffers.test.tsx, a vitest-style duplicate that used "vi" and imported a provider path that does not exist, of the working jest test beside the hook. This project runs jest, so it could never have executed. 54 -> 0 type errors. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
…t strkeys
The merged mock called jest.requireActual("@stellar/stellar-sdk"). requireActual
bypasses a manual mock but NOT moduleNameMapper, and the mapper points that
specifier straight back at this file — so `actual` was the mock itself and every
re-export from it was undefined ("actual.Account is not a constructor", which
failed 36 suites before a single test ran). It now requires the real SDK by
relative file path, the only form that escapes both the mapper and the package's
`exports` map.
With the real SDK actually loaded, its strkey validation rejected seven fixture
addresses that were never valid Ed25519 public keys despite being named
TESTNET_ACCOUNT, TESTNET_SOURCE and TARGET. Replaced with freshly generated
testnet keys across 16 files. The deliberate GXXXX... invalid-address fixture is
left alone.
399 -> 543 tests discovered; 384 -> 501 passing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
…bypassed it useFriendbot, useCreateAccount and useSep10Auth all threw plain Errors with `err.name = "SOME_CODE"`. Their `error` state is typed StellarError, and the README promises a stable `error.code`, but `name` on a StellarError is always "StellarError" — so every one of those codes was unreachable to a consumer branching on it. Converted 21 throws to createStellarError, and moved the matching assertions from `.name` to `.code`. WALLET_NOT_FOUND was not a code this library defines; the situation it named is "this wallet has no registered adapter", which is WALLET_UNSUPPORTED. Test doubles: getWalletAdapter moved to ../wallets, so the suites that drive it now automock that module rather than ../utils; useTrades' utils mock replaced the whole module and lost the pure asset guards the hook uses, so it spreads requireActual; and the SDK double now re-exports SorobanDataBuilder and the other pure builders a Soroban fixture needs. 518 -> 531 passing, typecheck still clean. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
…page on an empty result useBalance and useAccount both key on accountKey but cached different shapes under it, so whichever fetched first decided what the other read — the cross-hook dedup the integration suite asserts could never have worked. useBalance now caches the whole account and selects its balances, so one Horizon request genuinely serves both. useQuery left `loading: true` forever when a query became disabled mid-flight (the usual cause being the wallet disconnecting and the address going null): the effect returned early, so nothing was ever going to arrive and clear it. It now projects the current key's snapshot on the disabled path. The paginated hooks regained two behaviours the reducer rewrite had dropped: - Landing on an empty page keeps the page already on screen and updates only the navigation state, rather than blanking the list. - A superseded fetchNext can no longer install its results over a newer page. The reducer's queryKey check cannot catch this because a refetch does not change the key, so the hooks carry a monotonic request id and discard stale responses. lastUpdated is memoized on its timestamp; a fresh Date each render looked like a change to any consumer comparing it or using it as an effect dependency. The three dedup integration tests each rendered their hooks in *separate* renderHook calls, so every hook got its own provider and therefore its own store — they could not have observed deduplication under any implementation. They now share one provider, and the remount test toggles a child component rather than tearing down the tree (which took the cache with it). Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DBQmbmwJJUgb5MMtdsiyfW
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.