Fix byte-array PDA seeds deriving with a Borsh length prefix

[latent-9]

Problem

A Vec<u8> instruction argument used as a PDA seed keeps its Borsh sizePrefixTypeNode(u32, bytesTypeNode()) type when converted from an Anchor IDL. Anchor derives byte-array seeds from the raw bytes without the length prefix, so the generated client computes a different PDA than the on-chain program.

Minimal reproduction of the derivation difference, for seed data [1, 2, 3, 4, 5]:

• Anchor on-chain (create_program_address(&[seed], program)): CREFgTpumervLBfVvvuLx6w2yvgYCGGiXwcoozu22VHK
• Generated client (length-prefixed): 3mS84rEyJwNSpaYHeJN9umbi4FPZ94pkUg4Wv5hPBJ1s

String seeds already strip the prefix (see the isBorshString check); byte-array seeds were the missing case.

Change

Extend the argument-seed conversion in pdaSeedNodeFromAnchorV01 to also recognize a Borsh size-prefixed bytes argument and emit a raw bytesTypeNode seed, matching the on-chain derivation.

Tests

Adds it removes the bytes prefix from arg Anchor seeds, mirroring the existing string-prefix test. The full @codama/nodes-from-anchor suite passes.

Note

This is complementary to #990, which resolves nested seed paths in the same function. That PR does not change the byte-array seed behavior addressed here.

[changeset-bot]

🦋 Changeset detected

Latest commit: 943692c

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@codama/nodes-from-anchor	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[lorisleiva]

OMG I'm so sorry I was cleaning up some stale branches yesterday and I think I've accidentally deleted your branch. How rude of me. Let me just triage this with the AI reviewer first before I trouble you again. 🙏

[lorisleiva]

@trevor-cortex

[trevor-cortex]

Summary

Fixes PDA derivation for Vec<u8> instruction arguments used as seeds in Anchor v01 IDLs. Anchor derives byte-array seeds from the raw bytes, but the converted argument type kept its Borsh sizePrefixTypeNode(u32, bytesTypeNode()) encoding, so generated clients computed a different PDA than the on-chain program. The fix mirrors the existing isBorshString special case with an isBorshBytes counterpart that emits a raw bytesTypeNode() seed.

This is correct: bytes in an Anchor IDL is the Borsh encoding of Vec<u8> (u32 length prefix + data), while create_program_address receives the unprefixed slice via AsRef<[u8]>. Fixed-size arrays ([u8; N]) are unaffected since they carry no prefix, so this closes out the remaining variable-length case alongside strings.

Notes

• The new test mirrors the existing string-prefix test one-to-one, and the changeset is a patch bump on the independently-versioned @codama/nodes-from-anchor, starting with a verb per CONTRIBUTING — all consistent with repo conventions.
• Behavior note for anyone tracking downstream impact: any client currently generated from an IDL with a Vec<u8> seed will start deriving a different (now correct) PDA after this fix. That's the point of the fix, but worth remembering if a consumer somehow depended on the old, wrong derivation.
• This touches the same arg branch as #990 (nested seed path resolution), so whichever lands second will need a trivial rebase. The two changes are logically independent.
• Optional refactor, not blocking: isBorshString and isBorshBytes share the sizePrefixTypeNode + u32 prefix checks; a small helper like isBorshSizePrefixed(type, innerKind) could collapse the duplication if this pattern grows a third case.

[lorisleiva]

Let me know if you're happy to resubmit (again so sorry) otherwise I'm happy to re-open a similar PR a bit later if you'd rather not.

[latent-9]

Please re-open, happy to let you handle it.
Thank you @lorisleiva @trevor-cortex 🙏