fix(evolution): keep untagged sets untagged on re-encode (#574)

[Kshot3000]

Summary

Fixes #574: a transaction whose sets are plain or indefinite arrays, with no tag 258, came back from Transaction.fromCBORHex → Transaction.toCBORHex with d90102 added. The body bytes changed, so the transaction id changed and every witness already collected no longer verified.

I followed the fix the issue prescribes: let the captured format decide whether the tag is written.

The bug, reproduced first

New regression tests failed on main with exactly the reported symptom — every untagged set position gained d90102:

• body inputs (definite and indefinite forms), certificates, collateral inputs, required signers and reference inputs
• an untagged vkey witness set
• the nested choice the issue calls out: an input index written as 1800 sits inside the array node the decoder captured, and the old encodeTagSync dropped that whole subtree by encoding the inner array with no format

At the CBOR level the vector is minimal: a Tag 258 value encoded with an array-node format produced d90102820102 instead of 820102.

Fix

• CBOR.ts — encodeTagSync: when the tag is 258 and the captured format at that position is an array node, encode the inner array with that node and omit the tag. When the format is a tag node the tag is written (with its captured width), and with no captured format the tag is written exactly as before, so freshly built transactions do not change.
• TransactionWitnessSet.ts — withAddedFormats: its array branch used to convert an untagged set's format to the fresh (tagged) format when items were added, on the assumption that a decoded untagged set is written tagged. It now keeps the decoded array node and takes only the added items' formats from the fresh tree's inner array, so an edit — a datum appended to an untagged datum set — keeps the original form of everything around it, as the issue requires. One existing test asserted the old tagged outcome for that case; its expectation now asserts the untagged form and says why.

Nothing else changes: no encoder change was needed in TransactionBody.ts or the witness-set transform — both already build the Tag node, and the format tree is where the tag/no-tag fact lives.

Tests

• CBOR level (5 new): definite array node omits the tag, indefinite array node omits it and keeps the indefinite form, nested widths inside the array node survive (1800 index), tag node keeps the tag, no format keeps the tag.
• Transaction level (9 new): byte-identical round trips for definite and indefinite untagged inputs via both the explicit WithFormat API and the cached default fromCBORHex/toCBORHex path; body hash equality against CML; the combined certificates/collateral/signers/reference-inputs body; an untagged vkey witness set; addVKeyWitnessesHex leaving the body bytes unchanged; a witness appended to an untagged witness set keeping the untagged form; and the tagged/fresh controls.
• Full suite: 1692 passed / 75 skipped / 0 failed (test files 101 passed / 1 skipped of 102; upstream-main baseline 1678 + the 14 new tests). tsc --noEmit and eslint clean; the touched files' prettier diff-line counts exactly equal main's; a changeset is included.

—

Independent builder, not affiliated with the Cardano Foundation, IOG or Intersect.

@Kshot9000 (https://x.com/kshot9000) · Tips: addr1q8hnl6vl5a6k3rw3n5g3jtte696zcl76kfatzv7gpswa9r0dj7fma6klq55y4ffm7tf0em09udnyhuk4ah92pl5x9jpqjae44v

cc @cardano-foundation @IntersectMBO

[Kshot3000]

Correction to the test count in the description above (now edited there too): the full suite result is 1692 passed / 75 skipped / 0 failed, not 1704. The 1704 figure added the 14 new tests to the wrong baseline (1690 was a previous branch's total; upstream main's baseline is 1678, and 1678 + 14 = 1692). The commit message still says 1704 — I am not amending it (no force-push); this comment and the edited description carry the correct count. Nothing else in the description changes.