v0.5.0

Changed

• Breaking: a new editing API that can no longer corrupt a document. Every operation that can fail returns Result<_, EditError>, and a refused edit changes nothing. Editor::new(source, &parse_options, &edit_options)? replaces new, with_parse_options, with_edit_options and with_options. Editor::section(name) is a lookup that returns Option and never creates a section; section_or_insert(name)? creates one, and preamble() edits the keys before the first header (0.4's section("") appended a [] section instead). Positions count body lines, so the header is never one: insert_entry, append_body_lines, insert_body_lines, remove_body_lines and body_line_count replace insert_entry_at_line, append_raw_lines, insert_raw_lines_at, remove_lines and the header-at-0 indexing, and out-of-range positions and reversed ranges are errors instead of being clamped. Handles borrow the editor exclusively, so a stale or removed handle no longer compiles; entry(key) and entry_at(index) replace entries_mut(), and EntryEditor::index() and the new body_line() are always current. remove_entry and rename_key return Result<bool, EditError>, and a rename onto an existing key is Err(DuplicateKey). SeparatorSpacing::exact(before, after)? validates its whitespace, and the Exact fields are the new Whitespace type. The editor module docs have a migration guide whose examples are compiled (#90)
• Breaking: move to rowan 0.17. SyntaxNode, SyntaxToken, SyntaxElement, Parse::green() and IniLang's rowan::Language impl now use rowan 0.17's types, so code that names rowan types needs rowan = "0.17". rowan 0.17 removes its mutable tree API (clone_for_update, detach, SyntaxNode::splice_children, new_root_mut, the *_by_kind helpers), and SyntaxNode::green() returns &GreenNodeData instead of a Cow: replace green().into_owned() with green().to_owned(), and edit through Editor or by rebuilding green nodes with replace_with (#99)
• Diagnostics describe the INI text instead of internal token names, and each position gets at most one diagnostic. =bad reports expected a key before '=' (was unexpected token: LEX_ERROR), :x reports expected a key before ':', key junk reports only expected '=' or ':' (or unexpected text after key when allow_no_value is on), and a lone [ reports expected section name and ']' instead of two diagnostics. Code that matches on ParseError::message or counts diagnostics will see the new text (#101)

Added

• Editor is Send + Sync + UnwindSafe + RefUnwindSafe, and SectionEditor and EntryEditor are Send + Sync. The crate docs' new "Threads" section lists which types can cross threads; syntax nodes, tokens and the ast types still cannot (#99, #102)
• ini_edit::MAX_INPUT_LEN, the largest document a tree can describe (u32::MAX bytes). The editor refuses documents and edits past its limit with EditError::DocumentTooLarge (#90, #93)

Fixed

• The editor no longer writes text that a reader would read back differently. Keys, values, section names, separator spacing and raw lines that would not round-trip are refused instead of written: for example, setting a value to "a\n[evil]\nx=1" used to move every following entry into a new [evil] section, a key like a=b read back as the key a, and padded section names created a duplicate section every time the file was reopened. Values ending in a backslash, which absorbed the next line, are refused too (#90)
• The editor's view always equals a reparse of its text. Raw lines containing line breaks, lines inserted above a header and removing a header used to leave the editor's tree disagreeing with the saved file, so later edits went to different lines depending on whether the file was reopened in between. Raw lines are now parsed, so a raw k=2 is an entry that set and rename_key find (#90)
• ParseError::display puts the caret under the reported column. It was two columns to the left, three from line 1000 on, and further with tabs (#101)
• parse and parse_with panic with a clear message on input longer than MAX_INPUT_LEN instead of returning a tree whose length wrapped around (a 4 GiB document gave an empty tree with no diagnostics in release builds) (#93)
• The published crate's own tests pass: 0.4.0's package left out the snapshot files, so 31 tests failed on its crates.io sources (#84)

Performance

Measured against 0.4.0 from crates.io on the same machine; parsing speed is unchanged.

• Editing is faster: Editor::finish is about three times faster (gitea's app.example.ini: 364 to 115 µs), 200 set calls in one section of a 10,000-entry file are 29% faster and 200 append_entry calls 55% faster, because each edit rebuilds one immutable section instead of splicing a mutable tree line by line (#99)
• The typed AST reads keys, values, names and comments directly from the tree's green nodes: walking every entry of an already-parsed app.example.ini takes 7 µs instead of 22, and finding one key 0.8 µs instead of 2.4 (#100)

Documentation

• Add SECURITY.md (private vulnerability reporting), CONTRIBUTING.md, and docs/standards.md, which records how ini-edit applies TigerStyle and SQLite's testing practices and the deliberate deviations (#86)

v0.4.0

Changed

• Breaking: every malformed line is now a single node in the syntax tree. A line that starts with an unexpected token is wrapped in the new SyntaxKind::ERROR_LINE, and text after a key without a separator (key junk) stays inside its ENTRY. Code that matches exhaustively on SyntaxKind, or walks the children of malformed lines, needs updating. Diagnostics are unchanged (#60)
• Lines added by the editor use the document's line ending instead of always \n. This is the first line break in the current text, or \n if there is none, so editing CRLF or CR files no longer mixes line endings (#63)
• Report unexpected text after a section header ([s] extra) as a parse error. It was previously accepted silently (#52)
• With both allow_no_value and inline_comments enabled, a bare key can carry an inline comment (quick # note). This was previously a parse error, and assigning a value glued the comment onto it (#62)
• Editor::file() returns an immutable snapshot, as documented, instead of the editor's live tree, so a saved view no longer changes when the editor is edited (#54)

Fixed

• Editing a key followed by junk (key junk, MySQL's !includedir /path) no longer glues the new value onto the junk, and removing that key no longer leaves the junk behind as a new key (#60)
• remove_lines and insert_raw_lines_at count each malformed line as one line, so later indices are no longer shifted and removing a line can no longer join two lines (#60)
• Keep a bare \r and a following \n as two line breaks when inserting lines, including between lines inserted together, so line indices match a reopened file (#61)
• Keep a bare \r and a following \n as two line breaks when deleting entries, sections, or lines in files with mixed line endings (#58)
• insert_entry_at_line with a line past the last content line inserts exactly where append_entry does, including after trailing malformed lines (#66)
• Don't count a leading UTF-8 BOM as a column in ParseError::line_col and display (#59)
• Parse a first line that is indented after a UTF-8 BOM, and keep the BOM when removing the first section or entry (#45)
• Fix a panic when adding to a retained section handle after removing all its lines, and keep insertions from joining lines after deletions (#47)
• Give the same results whether or not a file is saved and reopened between edits. Inserted line breaks belong to their line, so remove_lines can no longer join a header to a setting, and a new section after an empty, unterminated value no longer ends up inside that value (#49)
• Clearing a value no longer turns its inline comment into the value when the file is reopened (#50)
• Assigning a continued value that ends on a blank line no longer absorbs the inline comment, and clearing a value with Compact or Exact spacing no longer keeps the old surrounding spaces (#53)

Performance

• Parse input with many malformed lines in linear time instead of quadratic time (#46)
• Repair line boundaries between inserted lines before attaching them to the document, about twice as fast for large raw-line batches that need repairs (#64)

Documentation

• Explain that a value ending in a backslash continues on the next line when the file is reopened (#67)

v0.3.2

Added

• Add configurable separator spacing for edited entries (#43)

Added

• Add configurable separator whitespace for entries created or value-updated through the editor (#43)

v0.3.1

Fixed

• Add a canonical separator when assigning a value to a bare key, preventing invalid output such as flagon (#39)
• Preserve bare-CR line endings during verbatim raw-line insertion (#39)
• Report accurate parse-error locations and source excerpts for LF, CRLF, and bare-CR input (#39)
• Make empty and reversed remove_lines ranges safe no-ops instead of panicking (#39)

Changed

• Require exhaustive source, branch, and MC/DC coverage alongside mutation testing, fuzzing, Miri, dependency auditing, and cross-platform checks (#40)
• Update vulnerable and yanked transitive development dependencies (#40)

v0.3.0

Added

• Add entry handles and guard rename_key against duplicates (#35)
• Separate auto-created sections with a blank line and never glue (#37)
• Add Editor::file() read view (#36)

v0.2.1

Added

• Snapshot tests for error diagnostics (#29)
• Line/column display for ParseError (#28)
• Benchmark comparison against rust-ini (#27)
• allow_no_value parse option for bare keys (#26)
• Real-world INI fixture tests (#25)
• Editor fuzz target and edge case tests (#24)

Fixed

• Out-of-bounds panic in editor surfaced by fuzzing (#24)

v0.2.0

Added

• Rework raw lines API — verbatim insertion, positional support (#22)
• Add criterion benchmarks and coverage workflow (#18)

Changed

• Use #[expect] at statement level for panic lints (#21)
• Remove stale lint allows, scope remaining ones (#19)