Changelog

Support 3.5 as a version option.
Many, many compatibility fixes for the parser translation layer.
Handle escapes in named capture names.
The freeze option is added to the various Prism:: APIs to deeply freeze the AST.
Properly support it for the parser and ruby_parser translation layers.
Track the then keyword on rescue nodes.
Add a multiple_statements? flag to parentheses nodes to support desired defined? behavior.

Changed

The strings used in the AST are now frozen.
Fixed handling escaped characters after control sequences in character literals.
Fix reading off the end of an unterminated global variable.
Raise a syntax error for defining []= with endless method syntax.
Increase value of PRISM_DEPTH_MAXIMUM to 10000.
Freeze Prism::VERSION.
Fix up rescue modifier precedence.

1.3.0 - 2024-12-21

Added

Introduce Prism::StringQuery.
Introduce Prism::Relocation.
Track do keyword for WhileNode and UntilNode.
Change the way the gem is built to rely on mkmf instead of make.
Lots more documentation on node fields.

Changed

Properly add an error for def @foo; end.
Properly add an error for foo(**, *).
Fix up regression in string parsing in RubyParser translation.
Reject invalid dot method call after match expression.
Reject invalid operator after match expression.
Fix up %-literals delimited by newlines.
Properly add an error for -> { _1; -> { _1 } }.
Reject blocks and keywords in index writes.

1.2.0 - 2024-10-10

Added

Introduce Prism::CodeUnitsCache.

Changed

Properly handle lexing global variables that begin with $-.
Properly reject invalid multi writes within parentheses.
Fix unary * binding power.
Set contains_keywords flag for implicit gets calls when -p is used.
Properly reject invalid non-associative operator patterns.
Do not warn about unused variables declared on negative lines.

1.1.0 - 2024-10-02

Added

Explicitly type each child node field in the Ruby API.
Add the main_script option to the parse APIs, which controls whether or not shebangs are considered.
Add the partial_script options to the parse APIs, which controls whether or not jumps that would otherwise be considered invalid are allowed. This is useful for parsing things like ERB sources, where you know it will be evaluated in a different context. Note that this functionality is replacing the previous idiom of passing in a list of scopes to indicate an eval context, because that behavior has changed upstream in ruby/ruby.
Add ArgumentsNode#contains_multiple_splats?.
Add ArgumentsNode#contains_forwarding?.
Accept all valid Ruby versions for the version option on parse APIs.
Accept version shorthands like "3.3" and "3.4" for the version option on parse APIs.
Support a max depth to protect against malicious payloads without hitting the stack limit.

Changed

Fix some token incompatibilities in the parser translation.
Fix up parsing tempfiles on Windows.
Fix up handling UTF-8 characters in file paths on Windows.
Do not warn for a \r at the end of a shebang on Windows.
Properly handle erroring for parsing a directory on Windows.
When a numbered reference is out of range, warn instead of raise.
Allow returns in default parameter values.
Reject many more invalid syntax patterns.

1.0.0 - 2024-08-28

Added

Add Node#breadth_first_search.
Add Node#node_id.
Add ArgumentsNode#contains_splat?.
Passing the special value false for the encoding option tells Prism to ignore magic encoding comments.
Expose flags on every node type (allows checking static literal and newline).
Implement mismatched indentation warning.
Add C API for receiving a callback when parsing shebangs with additional flags.

Changed

BREAKING: Some fields are renamed that had illogical names. The previous names all now emit deprecation warnings.
CaseMatchNode#consequent was renamed to CaseMatchNode#else_clause
CaseNode#consequent was renamed to CaseNode#else_clause
IfNode#consequent was renamed to IfNode#subsequent
RescueNode#consequent was renamed to RescueNode#subsequent
UnlessNode#consequent was renamed to UnlessNode#else_clause
Block exits are now allowed in loop predicates (e.g., while _ && break do end).
Multi-writes are now disallowed when not at the statement level.
Ensure that range operators are non-associative.
(JavaScript) Correctly deserialize encoded strings.
Properly support parsing regular expressions in extended mode.
Use gmake on FreeBSD.
Parsing streams now handles NUL bytes in the middle of the stream.
Properly detect invalid returns.

0.30.0 - 2024-06-07

Added

More correctly raise mixed encoding errors.
Implement ambiguous binary operator warning.
Fix up regexp escapes with control and meta characters.
Fix up support for the it implicit local variable.
Heredoc identifiers now properly disallow CLRF.
Errors added for void value expressions in begin clauses.
Many updates to more closely match the parser gem in parser translation.
Many errors added for invalid regular expressions.

Changed

Handle parser translation missing the parser gem.
Handle ruby_parser translation missing the ruby_parser gem.
Various error messages have been updated to more closely match CRuby.
RationalNode now has a numerator and denominator field instead of a numeric field. For the Ruby API we provide a RationalNode#numeric method for backwards-compatibility.

0.29.0 - 2024-05-10

Added

Added Prism::CallNode#full_message_loc, which gives the location including the = if there is one.
A warning for when # shareable_constant_value is not used on its own line.
An error for invalid implicit local variable writes.
Implicit hash patterns in array patterns are disallowed.
We now validate that Unicode escape sequences are not surrogates.

Changed

All fields named operator have been renamed to binary_operator for *OperatorWriteNode nodes. This is to make it easier to provide C++ support. In the Ruby API, the old fields are aliased to the new fields with a deprecation warning.
Many updated error messages to more closely match CRuby.
We ensure keyword parameters do not end in ! or ?.
Fixed some escaping in string literals with control sequences and hex escapes.
Fix a bug with RBS types when used outside the ruby/prism codebase.

0.28.0 - 2024-05-03

Added

Nested hashes will now warn for duplicated keys, as in: { foo: 1, **{ foo: 2 } }.
Prism::ReturnNode now has a flag on it to indicate if it is redundant.
Prism::Location#slice_lines and Prism::Node#slice_lines are now provided to slice the source code of a node including the content before the node on the same line that it starts on and the content after the node on the same line that it ends on.
Symbols with invalid byte sequences now give errors.
You can now pass "3.3.1" to the version: parameter on all Prism.* APIs.
Prism::Source#lines, Prism::Location#source_lines, Prism::Node#source_lines, and Prism::Node#script_lines are now provided, which will all return the source code of the source as an array of strings.
Prism::ASCIISource is now provided, which is a subclass of Prism::Source but specialized to increase performance when the source is entirely ASCII.
Prism now provides errors when parsing Ruby 3.4+ syntax for index expressions with keywords or blocks.
Prism now provides an error when **nil is used after other keyword parameters.
Prism now provides errors when safe navigation is used in call target expressions, e.g., foo&.bar, = 1.
Prism::Node#tunnel is now provided, which returns an array of nodes starting at the current node that contain a given line and column.

Changed

All translation layers now assume an eval context, which means they will not return errors for invalid jumps like yield.
Prism::Node#inspect now uses a queue instead of recursion to avoid stack overflows.
Prism now more closely mirrors CRuby interpolation semantics, which means you could potentially have a static literal string that directly interpolates another static literal string.
The shipped RBI sorbet types no longer use generics.
Prism::ConstantPathNode#child and Prism::ConstantTargetNode#child are now deprecated, replaced by two new fields on these nodes: name and name_loc.

0.27.0 - 2024-04-23

Added

Implemented === for each of the nodes, which will check if equality but ignore the specific ranges of locations.

Changed

Fix translation of ItParametersNode for parser translation.
Fix translation of dstr for ruby_parser translation.
Do not allow omitted hash values whose keys end with ! or ?.
Split up Prism::ParseResult into Prism::Result with subclasses Prism::ParseResult, Prism::LexResult, Prism::ParseLexResult, and Prism::LexCompat::Result.
Change reflection classes to have only a single IntegerField class and rename DoubleField to FloatField.
Fall back to default AR and CC in Makefile.
Use GC-able symbols for the syntax tree to avoid adding to the global symbol table.
Fix a bug with karatsuba_multiply that would result in a stack overflow.
Fix parser translation when looking for tokens with srange_find.

0.26.0 - 2024-04-18

Added

Add Prism::Node::fields, which returns a list of Prism::Reflection::Field objects representing the fields of the node class. This is useful in metaprogramming contexts.
Prism::Location#chop, for removing the last byte from a location.
The void statement warning is now implemented.
The unreachable statement warning is now implemented.
A syntax error has been added for block arguments on yields, e.g., yield(&foo).

Changed

Better fidelity to parser when translating heredocs with interpolation.
Fixed RBI and RBS types for Prism::parse_* signatures.
Remove some incorrect warnings about unused local variables.
More closely match CRuby error messages for global variables.
Fix an issue with parser translation when line continuations are found in string literals.

0.25.0 - 2024-04-05

Added

Prism::Translation::Ripper is now able to mirror all of the Ripper APIs.
Prism::Location#leading_comments and Prism::Location#trailing_comments is added.
Prism::Comment#slice is added.
Warn for writing literal values in conditional predicates.
Check for _POSIX_MAPPED_FILES before using mmap.
Prism::ItParametersNode is added, to support -> { it }.
Parse integer and float literal values onto the tree.
Warn on duplicated hash keys and duplicated when clauses.
Ship much improved RBI and RBS types.
Support for the -p, -n, -a, and -l command line switches.
Warn on integer literals in flip-flops.
Support BSD make.
Add Prism::WhenNode#then_keyword_loc.
Support custom allocation functions through the PRISM_XALLOCATOR define.
Warn for certain keywrods at the end of the line.
Provide pm_visit_node, a C visitor API.
Prism::parse_stream is added, which works for any Ruby IO object.
Provide flags for regular expression literals for their derived encoding.
Provide flags for whether or not an interpolated string literal is frozen.
Add Prism::StringNode.mutable? for when a string is explicitly mutable, to support delineating chilled strings.
Warn for incorrect character literal syntax.
Warn for chained comparison operators.
Warn for ** interpreted as an argument prefix.
Warn for & interpreted as an argument prefix.
Prism::ShareableConstantNode added to support ractors.
Warn for frozen string literals found after tokens.
Support PRISM_BUILD_MINIMAL to provide only the minimal necessary functionality to reduce the binary size.
Handle CLRF inside heredocs, strings, and regular expressions.
Mark inner strings in interpolated strings as frozen.
Support the -x command line switch.
Error messages now much more closely mirror CRuby.
Provide syntax errors for invalid block exits (break, next, retry, and yield).
Warn on unused local variables.
Do not syntax error on default parameter values that only write to the parameter.

Changed

Many improvements to the compatibility with the whitequark/parser translation.
Accept newlines before pattern terminators ) or ].
Prism::Node#start_offset and Prism::Node#end_offset are now much more efficient.
Read files using fread instead of mmap when we’re going to keep around the source through the Ruby API.
Fix Sexp#line_max setting in the seattlerb/ruby_parser translation layer.
Allow spaces before the encoding comment.

0.24.0 - 2024-02-15

Added

More support for Prism::Translation::Ripper is added.
Support multiple versions for Prism::Translation::Parser.
Improved memory usage in the FFI backend.
Very large speed and memory improvements for creating the Ruby AST.

Changed

Fix location for empty symbol in hash key.
Fix parsing a rescue modifier on the value of an assignment when the LHS is a method call with arguments and no parentheses.

0.23.0 - 2024-02-14

Added

More support for Prism::RipperCompat is added.
A significantly faster offset cache for Prism::Translation::Parser is added for files with multibyte characters.
Prism::Translation::RubyParser is added.
Prism::ConstantPathTarget#full_name is added.
version: "3.4.0" is added as an option that is an alias for version: "latest".
Four new APIs are added to Prism::Location:
Prism::Location#start_code_units_offset
Prism::Location#end_code_units_offset
Prism::Location#start_code_units_column
Prism::Location#end_code_units_column
Invalid multibyte characters are now validated within strings, lists, and heredocs.

Changed

When defining def !@, the name_loc was previously only pointing to !, but now includes the @. The name is the same.
Prism::RipperCompat has been moved to Prism::Translation::Ripper.
Many of the error messages that prism produces have been changed to match the error messages that CRuby produces.

0.22.0 - 2024-02-07

Added

More support for Prism::RipperCompat is added.
Support for Ruby 2.7 has been added, and the minimum Ruby requirement has been lowered to 2.7.

Changed

The error for an invalid source encoding has a new :argument level to indicate it raises an argument error.
BeginNode nodes that are used when a class, singleton class, module, method definition, or block have an inline rescue/ensure/else now have their opening locations set to the beginning of the respective keyword.
Improved error messages for invalid characters.
Prism.parse_file and similar APIs will raise more appropriate errors when the file does not exist or cannot be mapped.
Correctly handle the recover parameter for Prism::Translation::Parser.

0.21.0 - 2024-02-05

Added

Add the pm_constant_pool_find API for finding a constant.

Changed

Fixes for Prism::Translation::Parser.
Ensure all errors flow through parser.diagnostics.process.
Fix the find pattern node.
Fix block forwarding with NumberedParametersNode.
Ensure we can parse strings with invalid bytes for the encoding.
Fix hash pairs in pattern matching.
Properly reject operator writes on operator calls, e.g., a.+ -= b.
Fix multi-byte escapes.
Handle missing body in begin within the receiver of a method call.

0.20.0 - 2024-02-01

Added

String literal hash keys are now marked as frozen as static literal.
IndexTargetNode now always has the ATTRIBUTE_WRITE flag.
Call*Node nodes now have an IGNORE_VISIBILITY flag.
We now support the it default parameter.
Errors and warnings now have levels associated with them.
Symbols now have correct encoding flags.
We have now merged parser-prism in, which provides translation to the whitequark/parser AST.
We now emit errors for invalid method definition receivers.

Changed

We now emit errors on invalid pinned local variables.
When passed scopes, it is now assumed that the innermost scope is the current binding.
We now provide better error recovery for non terminated heredocs.
Fix for RationalNode#value for non-decimal integers.
Unary symbols !@ and ~@ now unescape to ! and ~, respectively.
frozen_string_literal: false now works properly.

Removed

We’ve removed the locals_body_index field.
We’ve removed the verbose option on the various parse APIs. Warnings are now always emitted with their associated level so that consumers can decide how to handle them.

0.19.0 - 2023-12-14

Added

ArrayNode now has a contains_splat? flag if it has a splatted element in it.
All of the remaining encodings have been implemented.
Allow forwarding & in a method that has a ... parameter.
Many statements that are found in non-statement positions are being properly rejected now.
Void values are now properly checked.
Referencing a parameter in its own default value is now properly rejected.
DATA/__END__ is now parsed as its own field on parse result (data_loc) as opposed to as a comment.
Blank * now properly forwards into arrays.
ImplicitRestNode is introduced to represent the implicit rest of a destructure.
We now support negative start lines.
StringNode#heredoc?, InterpolatedStringNode#heredoc?, XStringNode#heredoc?, and InterpolatedXStringNode#heredoc? are introduced.
NumberedParametersNode is introduced to represent the implicit set of parameters when numbered parameters are used.
Prism::parse_success? and Prism::parse_failure? are introduced to bypass reifying the AST.
We now emit a warning for constant assignments in method definitions.
We now provide flags on strings and xstrings to indicate the correct encoding.
The hash pattern rest field now more accurately parses ** and **nil.
The equality operators are now properly parsed as non-associative.

Changed

BREAKING: Many fields have changed positions within their nodes. This impacts the C API and the Ruby API if you are manually creating nodes through the initializer.
BREAKING: Almost all of the error messages have been updated to begin with lowercase characters to match ruby/spec.
Unterminated strings with only plain content are now always StringNode as opposed to InterpolatedStringNode
BREAKING: Call node has been split up when it is in the target position into CallTargetNode and IndexTargetNode.

0.18.0 - 2023-11-21

Added

The ParametersNode#signature method is added, which returns the same thing as Method#parameters.
Visitor functionality has been added to the JavaScript API.
The Node#to_dot API has been added to convert syntax trees to Graphviz digraphs.
IfNode and UnlessNode now have a then_keyword_loc field.
Many more encodings are now supported.
Some new Location APIs have been added for dealing with characters instead of bytes, which are: start_character_offset, end_character_offset, start_character_column, and end_character_column.
A warning has been added for when END {} is used within a method.
ConstantPathNode#full_name{,_parts} will now raise an error if the receiver of the constant path is not itself a constant.
The in keyword and the => operator now respect non-associativity.
The .. and ... operators now properly respect non-associativity.

Changed

Previously ... in blocks was accepted, but it is now properly rejected.
BREAKING: librubyparser.* has been renamed to libprism.* in the C API.
We now properly reject floats with exponent and rational suffixes.
We now properly reject void value expressions.
BREAKING: The --disable-static option has been removed from the C extension.
The rescue modifier keyword is now properly parsed in terms of precedence.
We now properly reject defining a numbered parameter method.
BREAKING: MatchWriteNode now has a list of targets, which are local variable target nodes. This is instead of locals which was a constant list. This is to support writing to local variables outside the current scope. It has the added benefit of providing location information for the local variable targets.
BREAKING: CaseNode has been split into CaseNode and CaseMatchNode, the latter is used for case ... in expressions.
BREAKING: StringConcatNode has been removed in favor of using InterpolatedStringNode as a list.

0.17.1 - 2023-11-03

Changed

Do not use constant nesting in RBI files.

0.17.0 - 2023-11-03

Added

We now properly support forwarding arguments into arrays, like def foo(*) = [*].
We now have much better documentation for the C and Ruby APIs.
We now properly provide an error message when attempting to assign to numbered parameters from within regular expression named capture groups, as in /(?<_1>)/ =~ "".

Changed

BREAKING: KeywordParameterNode is split into OptionalKeywordParameterNode and RequiredKeywordParameterNode. RequiredKeywordParameterNode has no value field.
BREAKING: Most of the Prism:: APIs now accept a bunch of keyword options. The options we now support are: filepath, encoding, line, frozen_string_literal, verbose, and scopes. See the pull request for more details.
BREAKING: Comments are now split into three different classes instead of a single class, and the type field has been removed. They are: InlineComment, EmbDocComment, and DATAComment.

0.16.0 - 2023-10-30

Added

InterpolatedMatchLastLineNode#options and MatchLastLineNode#options are added, which are the same methods as are exposed on InterpolatedRegularExpressionNode and RegularExpressionNode.
The project can now be compiled with wasi-sdk to expose a WebAssembly interface.
ArgumentsNode#keyword_splat? is added to indicate if the arguments node has a keyword splat.
The C API pm_prettyprint has a much improved output which lines up closely with Node#inspect.
Prism now ships with RBS and RBI type signatures (in the /sig and /rbi directories, respectively).
Prism::parse_comments and Prism::parse_file_comments APIs are added to extract only the comments from the source code.

Changed

BREAKING: Multi{Target,Write}Node#targets is split up now into lefts, rest, and rights. This is to avoid having to scan the list in the case that there are splat nodes.
Some bugs are fixed on Multi{Target,Write}Node accidentally creating additional nesting when not necessary.
BREAKING: RequiredDestructuredParameterNode has been removed in favor of using MultiTargetNode in those places.
BREAKING: HashPatternNode#assocs has been renamed to HashPatternNode#elements. HashPatternNode#kwrest has been renamed to HashPatternNode#rest.

0.15.1 - 2023-10-18

Changed

Fix compilation warning on assigning to bitfield.

0.15.0 - 2023-10-18

Added

BackReferenceReadNode#name is now provided.
Index{Operator,And,Or}WriteNode are introduced, split out from Call{Operator,And,Or}WriteNode when the method is [].

Changed

Ensure PM_NODE_FLAG_COMMON_MASK into a constant expression to fix compile errors.
super(&arg) is now fixed.
Ensure the last encoding flag on regular expressions wins.
Fix the common whitespace calculation when embedded expressions begin on a line.
Capture groups in regular expressions now scan the unescaped version to get the correct local variables.
* and & are added to the local table when ... is found in the parameters of a method definition.

0.14.0 - 2023-10-13

Added

Syntax errors are added for invalid lambda local semicolon placement.
Lambda locals are now checked for duplicate names.
Destructured parameters are now checked for duplicate names.
Constant{Read,Path,PathTarget}Node#full_name and Constant{Read,Path,PathTarget}Node#full_name_parts are added to walk constant paths for you to find the full name of the constant.
Syntax errors are added when assigning to a numbered parameter.
Node::type is added, which matches the Node#type API.
Magic comments are now parsed as part of the parsing process and a new field is added in the form of ParseResult#magic_comments to access them.

Changed

BREAKING: Call*Node#name methods now return symbols instead of strings.
BREAKING: For loops now have their index value considered as part of the body, so depths of local variable assignments will be increased by 1.
Tilde heredocs now split up their lines into multiple string nodes to make them easier to dedent.

0.13.0 - 2023-09-29

Added

BEGIN {} blocks are only allowed at the top-level, and will now provide a syntax error if they are not.
Numbered parameters are not allowed in block parameters, and will now provide a syntax error if they are.
Many more Ruby modules and classes are now documented. Also, many have been moved into their own files and autoloaded so that initial boot time of the gem is much faster.
PM_TOKEN_METHOD_NAME is introduced, used to indicate an identifier that if definitely a method name because it has an ! or ? at the end.
In the C API, arrays, assocs, and hashes now can have the PM_NODE_FLAG_STATIC_LITERAL flag attached if they can be compiled statically. This is used in CRuby, for example, to determine if a duphash/duparray instruction can be used as opposed to a newhash/newarray.
Node#type is introduced, which returns a symbol representing the type of the node. This is useful for case comparisons when you have to compare against multiple types.

Changed

BREAKING: Everything has been renamed to prism instead of yarp. The yp_/YP_ prefix in the C API has been changed to pm_/PM_. For the most part, everything should be find/replaceable.
BREAKING: BlockArgumentNode nodes now go into the block field on CallNode nodes, in addition to the BlockNode nodes that used to be there. Hopefully this makes it more consistent to compile/deal with in general, but it does mean it can be a surprising breaking change.
Escaped whitespace in %w lists is now properly unescaped.
Node#pretty_print now respects pretty print indentation.
Dispatcher was previously firing _leave events in the incorrect order. This has now been fixed.
BREAKING: Visitor has now been split into Visitor and Compiler. The visitor visits nodes but doesn’t return anything from the visit methods. It is suitable for taking action based on the tree, but not manipulating the tree itself. The Compiler visits nodes and returns the computed value up the tree. It is suitable for compiling the tree into another format. As such, MutationVisitor has been renamed to MutationCompiler.

0.12.0 - 2023-09-15

Added

RegularExpressionNode#options and InterpolatedRegularExpressionNode#options are now provided. These return integers that match up to the Regexp#options API.
Greatly improved Node#inspect and Node#pretty_print APIs.
MatchLastLineNode and InterpolatedMatchLastLineNode are introduced to represent using a regular expression as the predicate of an if or unless statement.
IntegerNode now has a base flag on it.
Heredocs that were previously InterpolatedStringNode and InterpolatedXStringNode nodes without any actual interpolation are now StringNode and XStringNode, respectively.
StringNode now has a frozen? flag on it, which respects the frozen_string_literal magic comment.
Numbered parameters are now supported, and are properly represented using LocalVariableReadNode nodes.
ImplicitNode is introduced, which wraps implicit calls, local variable reads, or constant reads in omitted hash values.
YARP::Dispatcher is introduced, which provides a way for multiple objects to listen for certain events on the AST while it is being walked. This is effectively a way to implement a more efficient visitor pattern when you have many different uses for the AST.

Changed

BREAKING: Flags fields are now marked as private, to ensure we can change their implementation under the hood. Actually querying should be through the accessor methods.
BREAKING: AliasNode is now split into AliasMethodNode and AliasGlobalVariableNode.
Method definitions on local variables is now correctly handled.
Unary minus precedence has been fixed.
Concatenating character literals with string literals is now fixed.
Many more invalid syntaxes are now properly rejected.
BREAKING: Comments now no longer include their trailing newline.

0.11.0 - 2023-09-08

Added

Node#inspect is much improved.
YARP::Pattern is introduced, which can construct procs to match against nodes.
BlockLocalVariableNode is introduced to take the place of the locations array on BlockParametersNode.
ParseResult#attach_comments! is now provided to attach comments to locations in the tree.
MultiTargetNode is introduced as the target of multi writes and for loops.
Node#comment_targets is introduced to return the list of objects that can have attached comments.

Changed

BREAKING: GlobalVariable*Node#name now returns a symbol.
BREAKING: Constant*Node#name now returns a symbol.
BREAKING: BlockParameterNode, KeywordParameterNode, KeywordRestParameterNode, RestParameterNode, DefNode all have their name methods returning symbols now.
BREAKING: ClassNode#name and ModuleNode#name now return symbols.
BREAKING: Location#end_column is now exclusive instead of inclusive.
Location#slice now returns a properly encoded string.
CallNode#operator_loc is now CallNode#call_operator_loc.
CallOperatorAndWriteNode is renamed to CallAndWriteNode and its structure has changed.
CallOperatorOrWriteNode is renamed to CallOrWriteNode and its structure has changed.

0.10.0 - 2023-09-01

Added

InstanceVariable*Node and ClassVariable*Node objects now have their name returning a Symbol. This is because they are now part of the constant pool.
NumberedReferenceReadNode now has a number field, which returns an Integer.

Changed

BREAKING: Various operator_id and constant_id fields have been renamed to operator and name, respectively. See 09d0a144 for details.
%w, %W, %i, %I, %q, and %Q literals can now span around the contents of a heredoc.
BREAKING: All of the public C APIs that accept the source string now accept const uint8_t * as opposed to const char *.

0.9.0 - 2023-08-25

Added

Regular expressions can now be bound by \n, \r, and a combination of \r\n.
Strings delimited by %, %q, and %Q can now be bound by \n, \r, and a combination of \r\n.
IntegerNode#value now returns the value of the integer as a Ruby Integer.
FloatNode#value now returns the value of the float as a Ruby Float.
RationalNode#value now returns the value of the rational as a Ruby Rational.
ImaginaryNode#value now returns the value of the imaginary as a Ruby Complex.
ClassNode#name is now a string that returns the name of just the class, without the namespace.
ModuleNode#name is now a string that returns the name of just the module, without the namespace.
Regular expressions and strings found after a heredoc declaration but before the heredoc body are now parsed correctly.
The serialization API now supports shared strings, which should help reduce the size of the serialized AST.
*Node#copy is introduced, which returns a copy of the node with the given overrides.
Location#copy is introduced, which returns a copy of the location with the given overrides.
DesugarVisitor is introduced, which provides a simpler AST for use in tools that want to process fewer node types.
{ClassVariable,Constant,ConstantPath,GlobalVariable,InstanceVariable,LocalVariable}TargetNode are introduced. These nodes represent the target of writes in locations where a value cannot be provided, like a multi write or a rescue reference.
UntilNode#closing_loc and WhileNode#closing_loc are now provided.
Location#join is now provided, which joins two locations together.
YARP::parse_lex and YARP::parse_lex_file are introduced to parse and lex in one result.

Changed

When there is a magic encoding comment, the encoding of the first token’s source string is now properly reencoded.
Constants followed by unary & are now properly parsed as a call with a passed block argument.
Escaping multi-byte characters in a string literal will now properly escape the entire character.
YARP.lex_compat now has more accurate behavior when a byte-order mark is present in the file.
BREAKING: AndWriteNode, OrWriteNode, and OperatorWriteNode have been split back up into their 0.7.0 versions.
We now properly support spaces between the encoding and =/: in a magic encoding comment.
We now properly parse -> foo: bar do end.

0.8.0 - 2023-08-18

Added

Some performance improvements when converting from the C AST to the Ruby AST.
Two rust crates have been added: yarp-sys and yarp. They are as yet unpublished.

Changed

Escaped newlines in strings and heredocs are now handled more correctly.
Dedenting heredocs that result in empty string nodes will now drop those string nodes from the list.
Beginless and endless ranges in conditional expressions now properly form a flip flop node.
% at the end of files no longer crashes.
Location information has been corrected for if/elsif chains that have no else.
__END__ at the very end of the file was previously parsed as an identifier, but is now correct.
BREAKING: Nodes that reference &&=, ||=, and other writing operators have been consolidated. Previously, they were separate individual nodes. Now they are a tree with the target being the left-hand side and the value being the right-hand side with a joining AndWriteNode, OrWriteNode, or OperatorWriteNode in the middle. This impacts all of the nodes that match this pattern: {ClassVariable,Constant,ConstantPath,GlobalVariable,InstanceVariable,LocalVariable}Operator{And,Or,}WriteNode.
BREAKING: BlockParametersNode, ClassNode, DefNode, LambdaNode, ModuleNode, ParenthesesNode, and SingletonClassNode have had their statements field renamed to body to give a hint that it might not be a StatementsNode (it could also be a BeginNode).

0.7.0 - 2023-08-14

Added

We now have an explicit FlipFlopNode. It has the same flags as RangeNode.
We now have a syntax error when implicit and explicit blocks are passed to a method call.
Node#slice is now implemented, for retrieving the slice of the source code corresponding to a node.
We now support the utf8-mac encoding.
Predicate methods have been added for nodes that have flags. For example CallNode#safe_navigation? and RangeNode#exclude_end?.
The gem now functions on JRuby and TruffleRuby, thanks to a new FFI backend.
Comments are now part of the serialization API.

Changed

Autotools has been removed from the build system, so when the gem is installed it will no longer need to go through a configure step.
The AST for foo = *bar has changed to have an explicit array on the right hand side, rather than a splat node. This is more consistent with how other parsers handle this.
BREAKING: RangeNodeFlags has been renamed to RangeFlags.
Unary minus on number literals is now parsed as part of the literal, rather than a call to a unary operator. This is more consistent with how other parsers handle this.

0.6.0 - 2023-08-09

Added

🎉 Initial release! 🎉