Error Accumulation

Accumulating per-object validation errors and per-region syntax errors.

Rather than stopping at the first problem, a conformant processor accumulates errors and returns them together, alongside whatever data parsed successfully. This gives authors a full picture in one pass.

Per-object validation errors

Each record is validated independently and may contribute zero, one, or many errors. A failing record is marked as an error; the others are unaffected:

~ $schema: { name: string, age: { int, max: 25 } }
---
~ James, 20    # ✓
~ Alex, 30     # ✗ mismatched-max
~ Bob, 22      # ✓

The result contains two valid records and one error entry — not a single fatal failure.

Per-region syntax errors

Syntax errors are accumulated per recovered region (between boundaries). One unparsable record yields one error, and parsing resumes at the next ~ or ---.

Partial output

Because errors are accumulated rather than thrown, the loaded result includes the records that succeeded. Consumers can render valid data and surface the error list side by side (for example, editor markers at each error's position).

Duplicate section names

When two sections share a name, the duplicate is automatically renamed so the rest of the document still loads. A recovering parser MUST NOT drop a section, and MUST NOT let one overwrite another: either would lose data with nothing to show for it.

The renaming rule, stated exactly, because two implementations that disagree here produce differently-named sections from the same document:

On encountering a section whose name is already in use, append _2 to the original name. If that name is also in use, try _3, then _4, and so on, until an unused name is found. The counter is per name, not per document, and it counts names already taken — including names a later section spelled out for itself.

So a document with three users sections yields users, users_2, users_3; and a document with sections named a, b, a, b yields a, b, a_2, b_2 — not a_2, b_3.

Because the rule counts names already taken, an explicit name cannot be silently displaced:

--- users        → users
--- users_2      → users_2   (written that way by the author)
--- users        → users_3   (skips users_2, which is taken)

This applies to sections that carry no name of their own, too: they take the default name data, so three unnamed sections become data, data_2, data_3.

The document is still invalid: section names must be unique, and the error is reported alongside the recovered data. The error is duplicate-section-name — a structural fault, not a lexical one, since every character in the document is valid.

See Also

• Error Codes · Error Model · Parser Behavior & Recovery

• Collection Rules