Thread byte offset information through all error creation sites,
eliminating the need for SubsliceOffset to recover position from
pointer comparison.
Changes:
- Add Offset field to ParserError struct
- Add offset parameter to NewParserError
- Add Parser.offsetOf helper for suffix-length arithmetic
- Thread base offset through scanner functions (scanComment,
scanBasicString, scanMultilineBasicString, scanLiteralString,
scanMultilineLiteralString, scanWindowsNewline)
- Thread base offset through standalone functions (expect, hexToRune)
- Thread base offset through all decode functions (parseInteger,
parseFloat, parseLocalDate, parseLocalTime, parseLocalDateTime,
parseDateTime, checkAndRemoveUnderscores*)
- Update all unmarshaler call sites to pass value.Raw.Offset
- Update localtime.go UnmarshalText methods with base=0
- Update strict.go to populate Offset from key ranges
- Change wrapDecodeError to read de.Offset directly
- Change Utf8TomlValidAlreadyEscaped to return int index (-1 if valid)
instead of a byte subslice
- Unexport SubsliceOffset (now only used internally by Range())
This makes error positions self-describing: each ParserError carries its
own byte offset, so callers no longer need the original document slice
and address arithmetic to determine where an error occurred.
Co-authored-by: Thomas Pelletier <thomas@pelletier.dev>
Replace cap(parent) - cap(subslice) with a straightforward scan
that compares element addresses: &data[i] == &subslice[0]. This is
well-defined Go pointer comparison on elements of the same backing
array, with no dependency on capacity semantics, reflect, or unsafe.
The scan is O(n) but only runs on error paths, and TOML documents
are small per the project's design constraints.
Also remove the Offset field from ParserError and the setErrOffset
machinery — the offset is computed at the point of consumption
(wrapDecodeError, Parser.Range) rather than cached on the error.
Co-authored-by: Thomas Pelletier <thomas@pelletier.dev>
ParserError.Offset is now a plain exported int field, always set:
- The parser sets it via setErrOffset() when capturing parse errors
- strict.go sets it from the key's Raw range at construction
- wrapDecodeError computes it inline from cap(document) - cap(highlight)
This eliminates:
- The SetOffset/Offset() accessor methods and offsetValid flag
- The subsliceOffset fallback function in errors.go
- Any conditional logic around whether the offset is present
The offset is guaranteed by construction at every path that creates
or consumes a ParserError.
Co-authored-by: Thomas Pelletier <thomas@pelletier.dev>
Instead of requiring downstream consumers to re-derive the byte offset
from pointer arithmetic on the Highlight slice, compute and cache the
offset inside the parser at error-capture time via setErrOffset().
This is safer because:
- The parser is the one place where the backing-array guarantee is known
to hold (Highlight is always a subslice of the parse buffer)
- Downstream consumers (wrapDecodeError) can use the cached offset
directly, avoiding the need for pointer comparison
- Errors created outside the parser (strict.go) set the offset from
existing Raw ranges, which are already correct by construction
Add ParserError.SetOffset/Offset methods for setting and retrieving the
cached offset. Update wrapDecodeError to prefer the cached offset when
available, falling back to subsliceOffset for backward compatibility.
Co-authored-by: Thomas Pelletier <thomas@pelletier.dev>
Removes all unsafe operations from go-toml, making the codebase
fully safe Go code. The internal/danger package that contained
unsafe operations has been deleted.
Changes:
- Replace pointer-based node navigation with index-based navigation
- Node.next and Node.child now store absolute indices into the
backing nodes slice instead of relative offsets
- Add nodes pointer to Node and Iterator for safe navigation
- Replace danger.TypeID with reflect.Type for cache keys
- Delete internal/danger package entirely
Performance overhead is under 10% compared to the unsafe version,
which is acceptable for the safety and maintainability benefits.
[Cursor][claude-sonnet-4-20250514]
* Use pointers instead of copying around ast.Node
Node is a 56B struct that is constantly in the hot path. Passing nodes
around by copy had a cost that started to add up. This change replaces
them by pointers. Using unsafe pointer arithmetic and converting
sibling/child indexes to relative offsets, it removes the need to carry
around a pointer to the root of the tree. This saves 8B per Node. This
space will be used to store an extra []byte slice to provide contextual
error handling on all nodes, including the ones whose data is different
than the raw input (for example: strings with escaped characters), while
staying under the size of a cache line.
* Remove conditional
* Add Raw to track range in data for parsed values
* Simplify reference tracking
* Benchmark script
* Rewrite unmarshaler using the stack
Instead of tracking the build chain using `target`s, use the stack
instead.
Working and most benchmarks look good, but regression on structs unmarshalling.
~60% slower on ReferenceFile/struct.
* Shortcut to check if last node of iterator
* Remove unecessary pointer allocation
* Skip over unused keys without marking them as seen
* Add some tests
* Fix mktemp on macos
Cursor Agent