Class HeliusRpcClient

An account's decoded bytes, or null when no such account exists.

The base64 decode lives here rather than at each call site, and null means absent and nothing else — see HeliusRpcClient.call.

Every account owned by programId, walked one cursor page at a time.

There is deliberately no single-shot getProgramAccounts here. V1 materialises every match before it replies, so it falls over on a large owner set — Helius refuses the call once its account-index service is loaded (-32603, "please use getProgramAccountsV2 with pagination"), and an owner as large as the upgradeable loader, which owns a loader account for every program on the cluster, is only readable through the cursor. A small owner set costs one page here, so V1 bought nothing worth the footgun of a method that works until the data grows.

filters narrows what comes back; dataSlice narrows how much of each one does. They are separate because listing loader buffers needs every match and none of the payload — 28 of them on devnet today, ~5 MB if fetched whole — so a caller can filter on bytes it never downloads.

The node pages over the scan, not over the results, so a page can come back empty and still carry a cursor. Only a null cursor ends the walk.

Signatures for address, newest first.

The RPC only ever walks backwards: before picks the page's starting point and until is the older signature it stops at. There is no key that means "newer than X" — an unknown config key is silently ignored by the node, so a misnamed watermark reads as no watermark and returns the full history.