[2.x] Adapt nodes-from-anchor to Codama v2

[lorisleiva]

This PR adapts @codama/nodes-from-anchor to the Codama v2 node model, for both legacy (v00) and current (v01) Anchor IDLs.

Behaviour

• Identifiers keep the casing of the IDL. Instructions, accounts, events, PDAs, fields, arguments, types, errors and constants are no longer camelCased, so every cross-reference (e.g. definedTypeLinkNodes and PDA seeds) matches the IDL exactly. Accounts flattened from nested groups are prefixed as ${group}_${account}, e.g. token_program_mint.
• Instruction arguments become instruction data. Arguments are fields of instructionNode.data, a structTypeNode whose first field is the discriminator (with a fieldDiscriminatorNode). PDA seeds referencing arguments use dataValueNodes.
• Wrapper nodes become transforms. Borsh strings and bytes use a u32 sizePrefixTransformNode, discriminators use fixedSizeTransformNodes and event data uses a hiddenPrefixTransformNode.
• Numbers map to integerTypeNode or floatTypeNode (f32/f64). Integer constants are kept as lossless strings, so 64- and 128-bit values no longer lose precision, and float constants are canonicalised textually (e.g. 007.50 → 7.5) without losing precision.
• Enums use enumVariantTypeNodes whose data is absent, a struct or a tuple.
• Docs are joined into a single string and omitted when empty. Program docs and the docs of account and event types are now carried over.

Removed exports

• instructionArgumentNodeFromAnchorV00 and instructionArgumentNodeFromAnchorV01: use structFieldTypeNodeFromAnchorV00 and structFieldTypeNodeFromAnchorV01 instead.
• The per-kind enum variant helpers (enumEmptyVariantTypeNodeFromAnchorV0x, enumStructVariantTypeNodeFromAnchorV0x, enumTupleVariantTypeNodeFromAnchorV0x), replaced by enumVariantTypeNodeFromAnchorV0x.

Bug fixes

• extractPdasVisitor no longer throws when fingerprinting PDAs.
• v01 argument seeds are matched by their exact identifier, so snake_case arguments no longer throw ARGUMENT_TYPE_MISSING.
• Shank instruction discriminators below index 16 are no longer odd-length hex strings.
• A missing generic type argument throws GENERIC_TYPE_MISSING instead of crashing.
• v01 errors without a message no longer get "name: " as docs, and empty v00 seed descriptions no longer produce empty docs.

Tests

All tests are ported to v2 nodes and raw casing, with new cases for joined docs, lossless and normalised constants, Shank discriminators, raw PDA collision renames and missing argument seeds.

[changeset-bot]

⚠️ No Changeset found

Latest commit: 563892d

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

This PR includes no changesets

When changesets are added to this PR, you'll see the packages that this PR includes changesets for and the associated semver types

Click here to learn what changesets are, and how to add one.

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

[lorisleiva]

@trevor-cortex

[lorisleiva]

• [2.x] Adapt the dynamic-instructions builder to Codama v2 #1194
• [2.x] Reject values of the wrong type when encoding dynamic codecs #1193
• [2.x] Adapt dynamic-address-resolution to Codama v2 #1192
• [2.x] Accept bytes, enum shorthands and struct defaults when encoding dynamic codecs #1191
• [2.x] Add resolveTypePath helper to visitors-core #1190
• [2.x] Adapt dynamic-parsers to Codama v2 #1187
• [2.x] Add withPath and visitPath helpers to NodeStack #1186
• [2.x] Adapt dynamic-codecs to Codama v2 #1185
• [2.x] Add relations, nested argument seeds and legacy PDAs to nodes-from-anchor #1181
• [2.x] Adapt nodes-from-anchor to Codama v2 #1180 👈 (View in Graphite)
• [2.x] Adapt instruction visitors to Codama v2 #1179
• [2.x] Adapt update visitors to Codama v2 #1178
• [2.x] Adapt structure visitors to Codama v2 #1175
• [2.x] Move casing helpers to @codama/fragments #1174
• [2.x] Adapt validators to Codama v2 #1173
• [2.x] Adapt visitors-core to Codama v2 #1172
• [2.x] Adapt nodes to Codama v2 #1171
• [2.x] Adapt errors to Codama v2 #1167
• [2.x] Rename node name attribute to identifier #1162
• [2.x] Regenerate the node types from the v2 spec #1158
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[trevor-cortex]

Summary

Ports @codama/nodes-from-anchor (v00 + v01) to the v2 node model: raw-cased identifiers, instructionNode.data as a structTypeNode with a leading discriminator field, wrapper nodes replaced by transforms (sizePrefixTransformNode, fixedSizeTransformNode, hiddenPrefixTransformNode), numberTypeNode split into integerTypeNode/floatTypeNode with string-backed value nodes, unified enumVariantTypeNode, and docs joined into a single string. Along the way it fixes a handful of latent bugs (extractPdas fingerprint throwing on '', snake_case arg seeds, odd-length Shank discriminators, missing generic crash, empty-docs artifacts).

The refactor is clean and consistent across both IDL versions — the shared helpers in utils.ts (docsFromAnchor, fixedSizeBytesTypeNode, borshSizePrefixedTypeNode, removeBorshSizePrefix) remove a lot of duplication, and the test coverage for the new behaviours (lossless integers, raw-casing collisions, Shank padding, missing arg seeds, joined docs) is solid.

Things worth a look

• canonicalFloat canonicalises through Number() — see inline. The DecimalString docs say canonicalisation is about spelling, not the f64 the string rounds to, and the current approach both silently mangles high-precision / large-magnitude decimals and downgrades small ones (0.0000001 → 1e-7 → string constant). A purely textual canonicaliser is a few lines and avoids both. Not blocking, but I'd fix it before merge since it's a new helper.
• camelCase is still imported in both InstructionAccountNode.ts files. That's intentional — hasDuplicateAccountNames uses it to detect collisions under the spec's casing-collision rule (the new token_mint vs tokenMint test covers it). Flagging so nobody treats it as a leftover from the camelCasing removal.
• origin is dropped from programNode. Correct for v2 (ProgramNode has no origin), but it means the Shank/Anchor distinction is no longer carried anywhere on the output tree. If a renderer ever needs it, a pluginNode is presumably the home — just noting it as a deliberate loss of information.
• Changeset: none in the PR. Per CONTRIBUTING.md the seeded major changeset on the 2.x line covers all public packages, so this may be intentional — worth confirming that's the convention being followed for this branch given the removed exports (instructionArgumentNodeFromAnchorV0x, per-kind enum variant helpers).

For subsequent reviewers

• removeBorshSizePrefix requires exactly one transform (transforms?.length === 1). That's fine for what this parser emits, but it's stricter than the v1 logic it replaces conceptually; anything decorating fields with extra transforms before PDA extraction would keep the prefix.
• Nested account seeds are prefixed with the current group prefix (pdaSeedNodeFromAnchorV01 → ${prefix}_${accountName}), so a nested PDA referencing a top-level account still gets the wrong name. Pre-existing, unchanged here.
• IdlV01Metadata.description exists and could serve as program-docs fallback — optional nit inline.

[trevor-cortex]

Re-review

Only utils.ts and the two ConstantNode test files changed since my last pass.

canonicalFloat is now textual — the regex-based canonicaliser replaces the String(Number(x)) round-trip, so high-precision, large-magnitude and small-fraction decimals all keep their exact spelling (3.14159265358979323846, 123456789012345678901.5, 0.0000001 are now covered by test.each cases in both v00 and v01). Output still matches DECIMAL_REGEX: leading zeros are stripped down to a single 0, trailing fraction zeros are dropped, and -0.00 → -0 is allowed by the spec since floats have signed zero. Non-decimal spellings like 1e3 still fall back to a string constant, which is the right call given the spec forbids exponent notation.

The optional metadata.description fallback for program docs wasn't taken, which is fine — that was a nit.

Everything else from the previous review stands; nothing new to flag. Good to merge.