Documentation
¶
Overview ¶
Package hashline applies edits guarded by content hashes instead of trusting stale offsets alone. It is intentionally exact by default: a hunk only applies when its old bytes hash to the recorded anchor hash and the same span can be found unambiguously in the current source.
Index ¶
Constants ¶
const ( // HashHexLength keeps anchors compact while preserving enough entropy for // local edit validation. The full SHA-256 is still used before truncation. HashHexLength = 16 // RelocationWindow bounds stale-offset recovery so the matcher stays cheap and // does not wander through unrelated repeated code looking for a plausible fit. RelocationWindow = 64 * 1024 )
const MaxEditFileBytes = 64 * 1024 * 1024 // 64 MiB
MaxEditFileBytes bounds how large a file this package will read fully into memory to apply a hash-anchored edit. This is not the same budget as the read tools' receipt-window caps: those bound what one read call returns to the model, while an edit target can be a much larger file than any single read window (offset can start anywhere in it). The bound here exists to reject an implausible target — a stray multi-GB log or data file someone points the tool at — before it is loaded into memory, not to constrain ordinary source files.
Variables ¶
This section is empty.
Functions ¶
func Apply ¶
Apply validates every hunk against src and returns a new byte slice with all edits applied. The input slice is never modified.
func ApplyFile ¶
ApplyFile reads path, applies hunks, and replaces the file with a same-directory temp file rename. The same directory keeps the rename on the same filesystem, which gives POSIX platforms atomic replacement semantics for readers.
func ReadFileForEdit ¶
ReadFileForEdit reads path as the source of a hash-anchored edit (for Apply, ApplyFile, or a manual Apply+ReplaceFileIfUnchanged sequence), refusing a file over MaxEditFileBytes before it is loaded into memory.
func ReplaceFileIfUnchanged ¶
ReplaceFileIfUnchanged atomically replaces path only when it still contains expected.
Types ¶
type Anchor ¶
Anchor records the exact byte span observed at read time.
type ApplyError ¶
type ApplyError struct {
Kind ErrorKind
Message string
ExpectedHash string
FoundHash string
RereadStart int
RereadEnd int
}
ApplyError carries structured context while still formatting as a plain error for tools that only expose textual failures.
func (*ApplyError) Error ¶
func (e *ApplyError) Error() string
type ErrorKind ¶
type ErrorKind string
ErrorKind identifies structured edit failures for model-facing recovery loops.
const ( ErrStaleAnchor ErrorKind = "stale_anchor" ErrAmbiguousAnchor ErrorKind = "ambiguous_anchor" ErrOverlap ErrorKind = "overlapping_hunks" ErrInvalidText ErrorKind = "invalid_text" ErrInvalidRange ErrorKind = "invalid_range" // ErrEmptyAnchor means a hunk has no old text. An empty span hashes to a // constant, so it verifies nothing about the file; anchor on adjacent text. ErrEmptyAnchor ErrorKind = "empty_anchor" // ErrHashMismatch means the hunk's Old text does not hash to Anchor.Hash. // It is distinct from ErrStaleAnchor, which means the input is internally // consistent but no longer matches the live source file. ErrHashMismatch ErrorKind = "hash_mismatch" ErrInvalidHash ErrorKind = "invalid_hash" ErrConcurrentWrite ErrorKind = "concurrent_write" )
const ErrFileTooLarge ErrorKind = "file_too_large"
ErrFileTooLarge means the target file exceeds MaxEditFileBytes.
Source Files