[asp24]

Advance and AdvanceIter skipped one word too many after a NOP. #9 found this error and did not change it. This PR corrects it.

The error

DeleteElems, SetNull and Deserialize write NOPs. They count the skip of a NOP from the NOP itself. AdvanceInto, PeekNext, NextElementBytes and MarshalJSON read it in the same way. Advance and AdvanceIter counted it from the word after the NOP.

Because of this, ForEach, Interface, FindKey and the helpers of Array lost or misread elements after a NOP:

document	edit	read with Array.ForEach before	after
[1,2,3,4]	DeleteElems removes 2	[1]	[1,3,4]
[true,false,null,true]	DeleteElems removes false	[true,true]	[true,null,true]
[1.5,2,3]	SetNull on 1.5	the value word of 2 read as a tag	[null,2,3]

Upstream has the same code.

The change

• Advance and advanceIter, the slow path of AdvanceIter, start the skip at the NOP.
• A skip of 0 now comes back to the same NOP. Advance cannot return an error, and it goes to the end, as AdvanceInto does. AdvanceIter already returned an error for it.
• The common path of each function does not change, because it does not read NOPs.
• The README adds the correction to the list of differences from upstream.

Testing

nop_skip_test.go edits documents with DeleteElems and SetNull, also after a round trip through Serialize and Deserialize. It reads each document through Advance, AdvanceIter, ForEach, Interface, FindKey and Array.MarshalJSON. It compares each result with encoding/json. MarshalJSON of the document and Object.Map were correct before, and they are the control.

• Before this change, the test failed on all readers except the control.
• Four intentional errors in the correction each made a test fail. Without the check for a skip of 0, the test stops at its timeout.
• The references in advance_diff_test.go had the same error, and they get the same correction.
• FuzzParse, FuzzCorrect and FuzzSerialize ran for one minute each, with no failure.

🤖 Generated with Claude Code