Improved performance on very large files that are compared by text.
Fixed some cases where changing list delimiters would lead to incorrect diffs.
Fixed an issue where lines were not aligned correctly after correcting sliders.
Fixed an issue the outermost delimiter in lists was sometimes incorrectly marked as unchanged, producing non-optimal diffs.
Display now prefers to align blank lines in the display, producing significantly better results in many cases.
Fixed an issue where some lines in a hunk were not displayed.
Difftastic now requires Rust 1.56 to build.
Added support for PHP.
Fixed handling of < > delimiters in C++ and Rust.
Difftastic will now split files that contain obviously unchanged regions, substantially improving performance when a file has multiple changes that have many unchanged items between them.
Improved diff results when choosing between syntax nodes at different nesting levels. This is restoring a heuristic that was removed in 0.20.
Improved diff results when lists have unequal sizes.
Improved diff results when the language parser thinks that names occur in different syntactic positions.
Adjusted the heuristics for 'so much has changed in this expression that it is confusing to highlight the unchanged parts'. The heuristic is now less aggressive, which helps performance and seems to produce slightly better results.
Difftastic now understands -*- file headers (as used by Emacs) when
performing language detection.
Improved alignment logic. This fixes a bug where the last line of a file wasn't displayed, and generally improves how difftastic chooses to align content.
Fixed a crash when line wrapping produced an entirely blank line.
Improved word diffing (in both comments and textual diffs) when source contains Unicode characters. Word splitting now uses the Unicode alphabetic property.
Fixed a crash when comments contained multibyte Unicode characters.
Diffing now correctly handles nodes being moved to parent lists. Previously this would be ignored, leading to difftastic incorrectly claiming things were unchanged. This also leads to better diffing results in general, although is somewhat slower (2x in testing).
Improved slider logic in larger expressions.
Increased the default value DFT_NODE_LIMIT to 100,000 (from 50,000). This increases the likelihood that files get a syntactic diff whilst still having acceptable performance.
Fixed an issue where whole file additions/removals were printed twice.
Fixed an issue where difftastic didn't show context on hunks where the unchanged content was on different lines.
Hunks are now merged if the lines are immediately adjacent (e.g. hunk 1 ends on line 11, hunk 2 starts on line 12), not just if they're overlapping.
Difftastic will now use a text dif for large files that are too big to
parse in a reasonable amount of time. This threshold is
configurable with --byte-limit and DFT_BYTE_LIMIT.
Fixed a crash when called with zero arguments.
Fixed an issue with changes being ignored in OCaml's {||} string
literals.
Fixed an issue where larger additions were not lined up with removals.
Improved syntax highlighting for Clojure, Common Lisp and TypeScript.
Comments are now highlighted with italics, making it easier to see syntax even when text is red.
Built-in constants are now highlighted consistently with other constants.
Improved minor display issues when one file is longer than the other.
If given binary files, difftastic will now report if the file contents are identical.
Difftastic will now use a text diff for large files, rather than
trying to use more memory than is available. This threshold is
configurable with --node-limit and DFT_NODE_LIMIT.
Fixed a bug in the text diff logic where lines weren't shown if they did not have both word additions and word removals.
Difftastic will now error if either argument does not exist, unless
--missing-as-empty (new argument) is passed. This is a better
default, but requires Mercurial uses to specify this
flag in their
configuration.
Fixed a compilation issue on Rust 1.54 (0.18 only built on newer versions of Rust).
Fixed an issue with missing positions in OCaml attribute syntax.
Fixed parsing issues in Common Lisp: character literals, loop macro
usage with maximizing.
Improved performance when diffing a single large expression.
Fixed display issues where lines were printed more than once.
Subword changes in comments are now shown in bold, to make them more visible.
Improved colours on terminals with light coloured backgrounds.
Added a --width option which allows overriding DFT_WIDTH, and is
more discoverable.
Added a --color option which allows explicitly enabling/disabling
colour output.
Added a --background option which controls whether difftastic uses
bright or dark colours. This can also be controlled by
DFT_BACKGROUND.
Added a --skip-unchanged option which suppresses printing for files
that have no changes.
Improved performance when all file changes are close together.
Fixed a bug where syntax after the last changed item was incorrectly marked as added.
Added syntax highlighting for unchanged comments, strings and types.
Fixed a bug (introduced in 0.15) where identical text files were reported as binary files.
Whitespace in JSX is parsed more closely to React's whitespace trimming rules.
Fixed parsing of heredocs in shell scripts. They are now treated as string literals.
Fixed parsing of type variables and tags in OCaml.
Improved language detection for files with bash/sh syntax.
Fixed a crash when on Mercurial diffs when a whole file has been removed.
Improved display performance when there are a large number of hunks.
Fixed several issues where lines were displayed more than once in a hunk.
Fixed an issue where the first changed line was not displayed.
Improved diffing performance (both time and memory usage).
Sliders are now fixed up after diffing. This produces better looking results in more cases, and makes the primary diffing faster.
Fixed some corner cases in the line parser where it would match up isolated newline character as unchanged, leading to weird alignment.
Moved to the official Elixir parser.
Updated the Bash, C, C++, C#, Haskell, Java, OCaml, Python, Ruby and TypeScript parsers to the latest upstream version.
Fixed a parsing performance regression introduced in 0.13.
Text diffing now has a standalone implementation rather than reusing structural diff logic. This is signficantly faster and highlighted better.
Improved performance when diffing two identical files. This is common when diffing directorires.
Improved highlighting heuristics for added/removed blank lines.
Fixed an alignment bug where the last line being novel would lead to poor alignment of unchanged lines.
Fixed minor formatting issues when reporting that a file is binary.
Improved display performance on large files.
Improved language detection if a file has a recognised filename
(e.g. Rakefile) or a shebang (e.g. #!/usr/bin/env node).
Display width can now be overridden by setting the environment variable DFT_WIDTH.
Fixed terminal width calculations on Windows.
Fixed crash when only one side has changes, but the other side has additional blank lines.
Fixed crash on displaying unicode characters on line boundaries.
Fixed some build issues on Windows.
Added Bash, Common Lisp and Ruby support.
Updated the C, CSS and JSON parsers to the latest upstream versions.
Expanded filename associations, so difftastic recognises more files.
Improved parsing for regex and template string literals in JavaScript and TypeScript.
Improved parsing for float values in CSS.
Improved word diffing on punctuation in comments.
When logging is enabled (e.g. RUST_LOG=warn), difftastic now warns
on syntax errors. Difftastic is intended to be robust against syntax
errors, so this is primarily intended for parser debugging.
Difftastic now requires fewer C compiler flags, so it should build in more environments (e.g. compiling with MSVC).
Every hunk is now shown with the file name and a hunk number. This makes it easier to see which file you're looking at when there are many changes.
Keywords in added/removed regions are now shown in bold, to give modified regions basic syntax highlighting. Previously, all added/removed regions were bold.
Lines with changes are now shown in a different colour in side-by-side display.
The display logic has been written in terms of a Hunk type. This
produces more accurate context, with better alignment, especially when
the context contains blank lines.
If only a single side has changes (e.g. additions but no removals), only one column is shown, to maximise display usage.
Difftastic now wraps rather than truncating lines that are too long for the terminal width.
If a file has no syntactic changes, difftastic now shows the file name consistently with changed files.
The difftastic binary is now named difft, to reduce typing during
usage.
Updated to latest upstream Haskell parser (commit d72f2e4).
Fixed a bug when diffing multiline comments where unchanged parts were not highlighted correctly.
Improved handling of paired delimiters, particularly in C, C++ and C#.
Improved word splitting in when diffing similar comments (it's now more granular).
Fixed a rare issue where single-item lists were flattened.
Diff calculations are now greedier when syntax nodes are identical, making diffing significantly faster when most syntax nodes are the same.
Added support for Mercurial, see this section in the manual for instructions.
Added basic syntax highlighting for comments (dimmed) and keywords (bold) in unchanged source code.
Characters that don't have a position in the parsed syntax tree are now displayed in purple, to make bugs more obvious. Previously they were dimmed.
Fix compilation on macOS where the C++ compiler defaulted to a version of C++ older than C++14.
Added a C parser.
Added a C++ parser. Difftastic prefers the C++ parser for .h
files. Please file a bug if you see issues.
Added a C# parser.
Added a Haskell parser.
Removed legacy regex-based parsing backend.
Some additional runtime optimisations.
Added a chapter on difficult cases for tree diff algorithms.
Added TypeScript parser and TSX parser. Added Elixir parser.
The following extensions are now associated with Clojure: .bb,
.boot, .clj, .cljc, .clje, .cljs, .cljx, .edn, .joke
and .joker.
Fixed an issue with parsing integer values in CSS with units,
e.g. 123px.
Improved parsing of Rust punctuation like & and :: inside macro
invocations. Improved handling of |closure_param| and [ ]
delimiters in Rust.
The line-based parser for text files now uses word-level diffs.
Optimised Dijkstra implementation, improving runtime performance.
Side-by-side displays now uses the same width for the left and right columns, regardless of the content.
Difftastic is now a library with a main binary. No APIs are considered stable for external usage. This is intended to make benchmarking easier.
Fixed a crash on removing whole files.
Tree-sitter parsing is now the default, unless the environment variable DFT_RX is set.
Tree-sitter parser: Improved handling of string literals. Improved matching of delimiters.
Added Python parser.
Added Java parser.
JSON (legacy parser): fixed parsing string literals (broken in 0.7).
Removed Scheme support, as there's no tree-sitter parser available.
Fixed crashes on files with non-ASCII characters on long lines.
Fixed an issue where multiline comments were not highlighted correctly.
Improved display to better use the whole width when whole files are added or removed.
Removed the unused --lang argument.
Difftastic now handles writing to a closed pipe (SIGPIPE) gracefully rather than crashing.
Difftastic now has some debugging logs available. RUST_LOG=trace
will show information on the route found during graph solving.
Fixed issues when adding/removing a whole file meant that difftastic didn't display anything.
Fixed a crash on renaming a file.
Colour is now enabled when using git with a pager.
Side-by-side display now uses "..." for column numbers when aligning lines. This makes hunks more obvious, but hunks now also have two blank lines between them to make it clearer.
Fixed an issue where screen width was not shared evenly by LHS and RHS.
Side-by-side display will now use the full width of the screen when using a pager (i.e. if stdout is a not a TTY).
Side-by-side display now handles whole file additions better, preferring a single column display.
Display width calculations are now based on the longest line visible in the diff, not the longest line in the file.
Added tree-sitter parsers. These have known bugs, but you can try
them by setting the environment variable DFT_TS=y.
Fixed handling of -> in Rust.
Difftastic will now prefer matching up comments that are similar (according to levenshtein distance).
Contiguous syntax logic now considers close delimiter positions, so
[ \n ]; now treats the ; atom as contiguous.
Fixed an issue where diffs would prefer prefer a low depth change on a delimiter over a delimiter that gave contiguous changes.
Removed the --width argument.
Added debug options --dump-syntax and --dump-ts for viewing parse
trees. The output of these options may change without notice.
Fixed handling of @, < and > in elisp.
Fixed crash on binary files. Difftastic now simply shows "binary" for files that don't look like text.
Added a basic Go parser.
Fixed an issue where comment replacements were not detected.
Changed words in comments are now only highlighted when comments are relatively similar (according to their Levenshtein distance).
Multiline comments are now considered unchanged if only their indentation changes.
Improved alignment for lines at the beginning of a changed group of lines.
Improved horizontal spacing between before and after code shown.
Fixed an issue where source code containing tab characters was not correctly aligned.
Removed unused --inline and --context arguments.
Fixed crash when called with no arguments.
Fixed a crash on parsing non-ASCII source files. Fixed a crash on files without an extension. Fixed crashes on empty files.
Input files that aren't valid UTF-8 are now replaced with � rather than giving up.
Improved parsing for Rust punctuation.
Improved parsing for OCaml punctuation, including := and method!.
Improved parsing for Emacs Lisp symbols containing + and =, and
punctuation of #, . and &.
Improved parsing for Scheme symbols containing =, and punctuation of
# and ..
Improved parsing of = and & in Clojure.
Improved parsing of :, ,, and constants in JSON.
Improved parsing of string literals in all languages, supporting
escaped delimiters such as "\"" and removing incorrect support for
single-quoted strings in JSON.
Reduced memory usage when diffing.
Difftastic now highlights word-level changes between comments.
Diffing now prefers contiguous nodes even when entering a list, so
(foo is considered contiguous.
Large AST trees with very few common nodes are now considered wholly
novel, rather than trying to match up the few common nodes. This
avoids nonsensical diffs when toplevel function A is completely
replaced with function B and they only have something trivial in
common (e.g. the function keyword).
Improved --help.
It's now possible to use difftastic with git diff and git show!
Improved parsing for Rust macro definitions and punctuation.
Improved parsing for OCaml punctuation, and added .mli as an OCaml
file extension.
Diff calculation is now significantly faster.
Difftastic now considers nesting depth when comparing AST nodes, and tries to match nodes with similar nesting levels.
Difftastic now prefers marking multiple items on the same line as novel, rather than adjacent items on different lines. This helps avoid sliders, where the diff chooses a keyword on the 'wrong' side.
Fixed an issue where complex diffs would not display some unchanged lines.
Fixed a crash when diff context included the first line.
Fixed a crash when plain text content contained certain non-ASCII characters.
Diffs are now displayed with unchanged lines aligned to the other side.
Improved Rust parsing to recognise lifetime syntax 'foo, character
literals 'x' and punctuation.
Improved punctuation parsing for OCaml and JS.
Fixed an issue where the diff calculated may not be minimal.
Fixed a crash on files with no changes.
First version using Dijkstra's algorithm for calculating diffs.
Experimenting with different implementation ideas.