CHANGES.md

grmtools 0.13.8 (2024-11-07)

• %parse-param can now use types that implement Clone (i.e. relaxing the previous stringent requirement that types were Copy).

• Document start states in the grmtools book.

• Allow lrlex and nimbleparse to read from stdin if the path is -.

grmtools 0.13.7 (2024-06-14)

• Catch incorrectly terminated productions. Previously this led to a confusing situation where productions could be merged together.

• Give better warnings if lexer/grammar can't be read at build time.

grmtools 0.13.6 (2024-05-30)

• Allow %prec to define a new token in the grammar. Previously if %prec 'x' was the first mention of "x" then an error would be raised.

• Tell the user where an incomplete action in a grammar started, not finished (since it always "finishes" at the end of the file).

grmtools 0.13.5 (2024-05-04)

• Improve error messages for conflicts and the like, giving the span of input related to the error.

grmtools 0.13.4 (2024-01-04)

• Change generated code to avoid errors about unsafe action code in input grammars.

• Hide unstable parts of the API behind an _unstable_api feature that is off by default.

grmtools 0.13.3 (2023-09-21)

• lrlex now explicitly raises an error when a rule in an input file has leading space. There is a small chance of this breaking existing input files, but it brings lrlex into line with POSIX lex where leading space indicates verbatim code (a concept for which lrlex currently has no support), making porting errors less likely.

• Report errors on %epp declarations in terms of the input file (rather than pretending they're all at line 1, column 1).

• Reorganise internal testing framework.

grmtools 0.13.2 (2023-08-01)

• Add CTLexerBuilder options for configuring regex behavior.

grmtools 0.13.1 (2023-01-27)

• Support %empty in productions. This Bison-ism can be used as a signal to readers that a production really is meant to be empty.

• Allow rules to be repeated, with each being treated as a separate production(s). In other words this grammar:

  A: 'x';
  A: 'y';
  

  is now equivalent to:

  A: 'x' | 'y';
  

grmtools 0.13.0 (2023-01-19)

This release contains a number of new features and breaking changes. Most of the breaking changes are in advanced/niche parts of the API, which few users will notice. However, four breaking changes might affect a more substantial subset of users: those are highlighted in the "breaking changes (major)" section below.

New features

• Improved error messages, with various parts of grammar and lexer files now carrying around Span information to help pinpoint errors.

• A single error can be related to multiple Spans. For example, if you duplicate a rule name in a grammar, all duplicates are reported in a single error.

• The new cfgrammar::NewlineCache struct makes it easier to store the minimal information needed to convert byte offsets in an input into logical line numbers.

• lrlex now supports start states.

• Unused tokens / rules in a grammar are now detected and, by default, reported as errors. The %expect-unused can suppress such warnings on a per-token/rule basis.

Breaking changes (major)

• Start states mean that lrlex now interprets < in the regular expression differently than before: to restore the previous behaviour, escape < with \. For example, the lrlex rule < "<" now appears to lrlex as an incomplete start state: replacing it with \< "<" fixes the problem.

• grmtools now bundles many of its type parameters into a LexexTypes trait, to avoid forcing users to endlessly repeat multiple arguments. If you specified a custom StorageT with lrlex/lrpar then you will need to change idioms such as DefaultLexeme<StorageT>, StorageT> to DefaultLexerTypes<StorageT>. For example, you might need to change:

  use lrlex::{DefaultLexeme, LRNonStreamingLexer};
  ...
  lexer: &LRNonStreamingLexer<DefaultLexeme<StorageT>, StorageT>,
  

  to:

  use lrlex::{DefaultLexerTypes, LRNonStreamingLexer};
  ...
  lexer: &LRNonStreamingLexer<DefaultLexerTypes<StorageT>>,
  

• Unused tokens / rules in a grammar are now detected and, by default, reported as errors. For example, the common "trick" to turn lexing errors into parsing errors suggests adding the following to a grammar:

  Unmatched -> ():
    "UNMATCHED" { } 
  ;
  

  Both the Unmatched rule and the UNMATCHED token will be reported as unused. You can tell grmtools that you expect this to happen with the %expect-unused directive:

  %expect-unused Unmatched "UNMATCHED"
  

• StorageT is now used to represent parser states (whereas before it was hard-coded to a u16). If you used a custom StorageT, it may no longer be big enough: if this happens, an error will be reported while the grammar is being built. You will then need to increase the size of your StoargeT (e.g. you might need to change StorageT from u8 to u16).

Breaking changes (minor)

• Serde support for lrpar now requires enabling the serde feature.

• cfgrammar::yacc::grammar::YaccGrammar::token_span now returns Option<Span> rather than Option<&Span>.

• cfgrammar::yacc::ast::{Production, Symbol} no longer derive Eq, Hash, and PartialEq. Since both now carry a Span, it's easy to confuse "two {productions, symbols} have the same name" with "at the same place in the input file."

• cfgrammar::yacc::ast::add_rule's signature has changed from:

    pub fn add_rule(&mut self, name: String, actiont: Option<String>) {
  

  to:

    pub fn add_rule(&mut self, (name, name_span): (String, Span), actiont: Option<String>) {
  

• GrammarValidationError and YaccParserError have been combined into a struct YaccGrammarError (which replaces the previous enum of that name). The new YaccGrammarError has a private enum, so will mean fewer semver-breaking changes.

• LexBuildResult returns on failure Err(Vec<LexBuildError>) rather than Err(LexBuildError).

• The lrlex::lexemes module has been renamed to lrlex::defaults to better describe what it is providing.

Deprecations

• Span has moved from lrpar to cfgrammar. The import is still available via lrpar but is deprecated (though due to rust-lang/rust#30827 this unfortunately does not show as a formal deprecation warning).

• cfgrammar::yacc::grammar::YaccGrammar::rule_name has been renamed to rule_name_str. The old name is still available but is deprecated.

grmtools 0.12.0 (2022-04-14)

• Move to Rust 2021 edition.

• Various minor clean-ups.

grmtools 0.11.1 (2021-12-07)

• Explicitly error if the users tries to generate two or more lexers or parsers with the same output file name. Previously the final lexer/parser created "won the race", leading to a confusing situation where seemingly correct code would not compile. Users can explicitly set an output path via output_path that allows multiple lexers/parsers to be generated with unique names.

grmtools 0.11.0 (2021-11-18)

An overview of the changes in this version:

• Lexeme is now a trait not a struct, increasing flexibility, but requiring some changes in user code.
• The build API has slightly changed, requiring some changes in user code.
• %parse-param is now supported.
• lrlex provides a new API to make it easy to use simple hand-written lexers instead of its default lexer.

Breaking changes

Lexeme is now a trait not a struct

lrpar now defines a Lexeme trait not a Lexeme struct: this allows the parser to abstract away from the particular data-layout of a lexeme (allowing, for example, a lexer to attach extra data to a lexeme that can be accessed by parser actions) but does add an extra type parameter LexemeT to several interfaces. Conventionally the LexemeT type parameter precedes the StorageT type parameter in the list of type parameters.

lrlex defaults to using its new DefaultLexeme struct, which provides a generic lexeme struct similar to that previously provided by lrlex (though note that you can use lrlex with a lexeme struct of your own choosing).

The precise effects of these changes will depend on how you use grmtools' libraries but in general:

• You will need to change your lexeme imports from:

    use lrpar::Lexeme;

  to:

    use lrlex::DefaultLexeme;
    use lrpar::Lexeme;

• Most references to Lexeme will need to refer to DefaultLexeme.

• Any references to LRNonStreamingLexer will need to change from:

    LRNonStreamingLexer<Lexeme<u32>>

  to:

    LRNonStreamingLexerDef<DefaultLexeme<u32>, u32>

  where u32 is the StorageT of your choice.

One of the additional benefits to this change is that it allows lrpar and other lexers (e.g. lrlex) to be clearly separated: lrpar now only defines traits which lexers have to conform to.

Build API changes

Several of the functions / structs surrounding the compile-time construction of grammars have changed: more details are given below, but in most cases a build.rs that looks as follows:

  use cfgrammar::yacc::YaccKind;
  use lrlex::LexerBuilder;
  use lrpar::CTParserBuilder;

  fn main() -> Result<(), Box<dyn std::error::Error>> {
      let lex_rule_ids_map = CTParserBuilder::<u8>::new_with_storaget()
          .yacckind(YaccKind::GrmTools)
          .process_file_in_src("calc.y")?;
      LexerBuilder::new()
          .rule_ids_map(lex_rule_ids_map)
          .process_file_in_src("calc.l")?;
      Ok(())
  }

can be changed to:

  use cfgrammar::yacc::YaccKind;
  use lrlex::CTLexerBuilder;
  
  fn main() -> Result<(), Box<dyn std::error::Error>> {
      CTLexerBuilder::new()
          .lrpar_config(|ctp| {
              ctp.yacckind(YaccKind::Grmtools)
                  .grammar_in_src_dir("calc.y")
                  .unwrap()
          })
          .lexer_in_src_dir("calc.l")?
          .build()?;
      Ok(())
  }

In more detail:

• lrlex's LexerBuilder has been renamed to CTLexerBuilder for symmetry with lrpar.

• CTLexerBuilder now provides the lrpar_config convenience function which removes some of the grottiness involved in tying together an lrlex lexer and lrpar parser. lrpar_config is passed a CTParserBuilder instance to which normal lrpar compile-time options can be applied.

• CTParserBuilder::process_file_in_src is deprecated in favour of CTParserBuilder::grammar_in_src_dir and CTParserBuilder::build. The latter method consumes the CTParserBuilder returning a CTParser which exposes a token_map method whose result can be passed to lexers such as lrlex.

  The less commonly used process_file function is similarly deprecated in favour of the grammar_path, output_path, and build functions.

• LexerBuilder::process_file_in_src is deprecated in favour of LexerBuilder::lexer_in_src_dir and LexerBuilder::build.

  The less commonly used process_file function is similarly deprecated in favour of the lexer_path, output_path, and build functions.

• CTLexerBuilder's and CTParserBuilder's build methods both consume the builder, producing a CTLexer and CTParser respectively, which can be queried for additional information.

The conflicts API has moved

• The unstable CTParserBuilder::conflicts method has moved to CTParser. This interface remains unstable and may change without notice.

New features

• Yacc grammars now support the %parse-param <var>: <type> declaration. The variable <var> is then visible in all action code. Note that <type> must implement the Copy trait. The generated parse function then takes two parameters (lexer: &..., <var>: <type>).

• lrlex now exposes a ct_token_map function which creates a module with a parser's token IDs, and allows users to call LRNonStreamingLexer::new directly. This makes creating simple hand-written lexers much easier (see the new calc_manual_lex example to see this in action).

grmtools 0.10.2 (2021-08-09)

• Optimise NonStreamingLexer::span_lines_str from O(n) to O(log n).

• Deprecate GrammarAST::add_programs in favour of GrammarAST::set_programs.

grmtools 0.10.1 (2021-05-11)

• Add support for Yacc's %expect and Bison's %expect-rr declarations. These allow grammar authors to specify how many shift/reduce and reduce/reduce conflicts they expect in their grammar, and to error if either quantity is different, which is a more fine-grained check than the error_on_conflicts boolean.

• Generate code with "correct" camel case names to avoid Clippy warnings in code that uses grmtools' output.

grmtools 0.10.0 (2021-03-29)

Breaking changes

• A number of previously public items (functions, structs, struct attributes) have been made private. Most of these were either not externally visible, or only visible if accessed in undocumented ways, but some were incorrectly public.

grmtools 0.9.3 (2021-01-28)

• Optimise NonStreamingLexer::line_col from O(n) to O(log n) (where n is the number of lines in the file).

• Document more clearly the constraints on what Spans one can safely ask a lexer to operate on. Note that it is always safe to use a Span generated directly by a lexer: the constraints relate to what happens if a user derives a Span themselves.

• Suppress Clippy warnings about unnecessary_wraps, including in code generated from grammar files.

grmtools 0.9.2 (2020-11-29)

• Export the Visibility enum from lrlex.

• Ensure that lrpar rebuilds a grammar if its visibility is changed.

grmtools 0.9.1 (2020-08-21)

• Fix a handful of Clippy warnings.

grmtools 0.9.0 (2020-07-06)

Breaking changes

• The MF and Panic recoverers (deprecated, and undocumented, since 0.4.3) have been removed. Please change to RecoveryKind::CPCTPlus (or, if you don't want error recovery, RecoveryKind::None).

Minor changes

• The stategraph is no longer stored in the generated grammar, leading to useful savings in the generated binary size.

grmtools 0.8.1 (2020-06-11)

• The modules generated for compile-time parsing by lrlex and lrpar have private visibility by default. Changing this previously required a manual alias. The visibility function in lrlex and lrpar's compile-time builders allows a different visibility to be set (e.g. visibility(Visibility::Public). Rust has a number of visibility settings and the Visibility enums in lrlex and lrpar reflect this.

grmtools 0.8.0 (2020-05-28)

Breaking changes

• lrlex now uses a LexerDef which all lexer definitions must impl. This means that if you want to call methods on a concrete lexer definition, you will almost certainly need to import lrlex::LexerDef. This opens the possibility that lrlex can seamlessly produce lexers other than LRNonStreamingLexerDefs in the future.

Deprecations

• lrlex::NonStreamingLexerDef has been renamed to lrlex::LRNonStreamingLexerDef; use of the former is deprecated.

• The lrlex::build_lex function has been deprecated in favour of LRNonStreamingLexerDef::from_str.

Bug fixes

• The statetable and other elements were previously included in the user binary with include_bytes!, but this could cause problems with relative path names. We now include the statetable and other elements in generated source code to avoid this issue.

grmtools 0.7.0 (2020-04-30)

Breaking changes

• The Lexer trait has been broken into two: Lexer and NonStreamingLexer. The former trait is now only capable of producing Lexemes: the latter is capable of producing substrings of the input and calculating line/column information. This split allows the flexibility to introduce streaming lexers in the future (which will not be able to produce substrings of the input in the same way as a NonStreamingLexer).

  Most users will need to replace references to the Lexer trait in their code to NonStreamingLexer.

• NonStreamingLexer takes a lifetime 'input which allows the input to last longer than the NonStreamingLexer itself. Lexer::span_str and Lexer::span_lines_str had the following definitions:

  fn span_str(&self, span: Span) -> &str;
  fn span_lines_str(&self, span: Span) -> &str;
  

  As part of NonStreamingLexer their definitions are now:

  fn span_str(&self, span: Span) -> &'input str;
  fn span_lines_str(&self, span: Span) -> &'input str;
  

  This change allows users to throw away the Lexer but still keep around structures (e.g. ASTs) which reference the user's input.

  rustc infers the 'input lifetime in some situations but not others, so if you get an error:

  error[E0106]: missing lifetime specifier
  

  then it is likely that you need to change a type from NonStreamingLexer to NonStreamingLexer<'input>.

grmtools 0.6.2 (2020-03-22)

• Fix two Clippy warnings and suppress two others.

• Prefer "unmatched" rather than "unknown" when using the "turn lexing errors into parsing errors" trick.

grmtools 0.6.1 (2020-03-02)

• Deprecate Lexeme::len, Lexeme::start, and Lexeme::end. Each is now replaced by Lexeme::span().len() etc. An appropriate warning is generated if the deprecated methods are used.

• Avoid use of the unit return type in action code causing Clippy warnings.

• Document the "turn lexing errors into parsing errors" technique and extend lrpar/examples/calc_ast to use it.

grmtools 0.6.0 (2020-02-10)

Breaking changes

• Introduce the concept of a Span which records what portion of the user's input something (e.g. a lexeme or production) references. Users can turn a Span into a string through the Lexer::span_str function. This has several API changes:

  • lrpar now exports a Span type.
  • Lexemes now have a fn span(&self) -> Span function which returns the Lexeme's `Span.
  • Lexer::span_str replaces Lexer::lexeme_str function. Roughly speaking this:

    let s = lexer.lexeme_str(&lexeme);

    becomes:

    let s = lexer.span_str(lexeme.span());

  • Lexer::line_col now takes a Span rather than a usize and, since a Span can be over multiple lines, returns ((start line, start column), (end line, end column)).
  • Lexer::surrounding_line_str is removed in favour of span_lines_str which takes a Span and returns a (possibly multi-line) &str of the lines containing that Span.
  • The $span special variable now returns a Span rather than (usize, usize).

  In practise, this means that in many cases where you previously had to use Lexeme<StorageT>, you can now use Span instead. This has two advantages. First, it simplifies your code. Second, it enables better error reporting, as you can now point the user to a span of text, rather than a single point. See the (new) AST evaluator section of the grmtools book for an example of how code using Span looks.

• The $span special variable is now enabled at all times and no longer needs to be turned on with CTBuilder::span_var. This function has thus been removed.

grmtools 0.5.1 (2020-01-02)

• If called as a binary, lrlex now exits with a return code of 1 if it could not lex input. This matches the behaviour of lrpar.

• Module names in generated code can now be optionally configured with mod_name. The names default to the same naming scheme as before.

• Fully qualify more names in generated code.

grmtools 0.5.0 (2019-12-17)

Breaking changes

• lrlex_mod and lrpar_mod now take strings that match the paths of process_file_in_src. In other words what was:

  ...
    .process_file_in_src("a/b/grm.y");
  
  ...
  
  lrpar_mod!(grm_y);

  is now:

  ...
    .process_file_in_src("a/b/grm.y");
  
  ...
  
  lrpar_mod!("a/b/grm.y");

  and similarly for lrlex_mod. This is hopefully easier to remember and also allows projects to have multiple grammar files with the same name.

• The Lexer API no longer requires mutability. What was:

  trait Lexer {
    fn next(&mut self) -> Option<Result<Lexeme<StorageT>, LexError>>;
    fn all_lexemes(&mut self) -> Result<Vec<Lexeme<StorageT>>, LexError> {
      ...
    }
    ...
  }

  has now been replaced by an iterator over lexemes:

  trait Lexer {
    fn iter<'a>(&'a self) ->
      Box<dyn Iterator<Item = Result<Lexeme<StorageT>, LexError>> + 'a>;
    ...
  }

  This enables more ergonomic use of the new zero-copy feature, but does require changing structs which implement this trait. lrlex has been adjusted appropriately.

  In practise, the only impact that most users will notice is that the following idiom:

  let (res, errs) = grm_y::parse(&mut lexer);

  will produce a warning that the mut part of &mut is no longer needed.

Major changes

• Add support for zero-copying user input when parsing. A special lifetime 'input is now available in action code and allows users to extract parts of the input without calling to_owned() (or equivalent). For example:

    Name -> &'input str: 'ID' { $lexer.lexeme_str(&$1.map_err(|_| ())?) } ;
  

  See lrpar/examples/calc_ast/src/calc.y for a more detailed example.

Minor changes

• Generated code now uses fully qualified names so that name clashes between user action code and that created by grmtools is less likely.

• Action types can now be fully qualified. In other words this:

  R -> A::B:
    ...
    ;

  means that the rule R now has an action type A::B.

grmtools 0.4.3 (2019-11-21)

• Deprecate the MF recoverer: CPCT+ is now the default and MF is now undocumented. For most people, CPCT+ is good enough, and it's quite a bit easier to understand. In the longer term, MF will probably disappear entirely.

• License as dual Apache-2.0/MIT (instead of a more complex, and little understood, triple license of Apache-2.0/MIT/UPL-1.0).

grmtools 0.4.2 (2019-06-26)

• Action code uses $ as a way of denoting special variables. For example, the pseudo-variable $2 is replaced with a "real" Rust variable by grmtools. However, this means that $2 cannot appear in, say, a string without being replaced. This release uses $$ as an escaping mechanism, so that one can write code such as "$$1" in action code; this is rewritten to "$1" by grmtools.

grmtools 0.4.1 (2019-06-10)

• Newer versions of rustc produce "deprecated" warnings when trait objects are used without the dyn keyword. This previously caused a large number of warnings in generated grammar code from lrpar. This release ensures that generated grammar code uses the dyn keyword when needed, removing such warnings.

grmtools 0.4.0 (2019-05-09)

Breaking changes

• Lexeme::empty() has been renamed to Lexeme::inserted(). Although rare, there are grammars with empty lexemes that aren't the result of error recovery (e.g. DEDENT tokens in Python). The previous name was misleading in such cases.

• Lexeme insertion is no longer explicitly encoded in the API for lexemes end/length. Previously these functions returned None if a lexeme had been inserted by error recovery. This has proven to be more effort than it's worth with variants on the idiom lexeme.end().unwrap_or_else(|| lexeme.start()) used extensively. These definitions have thus been simplified, changing from:

  pub fn end(&self) -> Option<usize>
  pub fn len(&self) -> Option<usize>

  to:

  pub fn end(&self) -> usize
  pub fn len(&self) -> usize

Major changes

• A new pseudo-variable $span can be enabled within parser actions if CTBuilder::span_var(true) is called. This pseudo-variable has the type (usize, usize) where these represent (start, end) offsets in the input and allows users to determine how much input a rule has matched.

Minor changes

• Some dynamic assertions about the correct use of types have been converted to static assertions. In the unlikely event that you try to run grmtools on a platform with unexpected type sizes (which, in practise, probably only means 16 bit machines), this will lead to the problems being determined at compile-time rather than run-time.

grmtools 0.3.1 (2019-04-10)

Minor changes

• Document lrpar more thoroughly, in particular hiding the inner modules, whose location might change one day in the future: all useful structs (etc.) are explicitly exposed at the module level.

grmtools 0.3.0 (2019-03-25)

Breaking changes

• Have the process_file functions in both LexerBuilder and CTParserBuilder place output into a named file (whereas previously CTParserBuilder expected a directory name).

• Rename offset_line_col to line_col and have the latter return character offsets (whereas before it returned byte offsets). This makes the resulting numbers reported to humans much less confusing when multi-byte UTF-8 characters are used.

Minor changes

• Add surrounding_line_str helper function to lexers. This is helpful when printing out error messages to users.

• Add a comment with rule names when generating grammars at compile-time. Thus if user action code contains an error, it's much easier to relate this to the appropriate point in the .y file.

grmtools 0.2.1 (2019-01-16)

• Documentation fixes.

grmtools 0.2.0 (2019-01-14)

Breaking changes

• Previously users had to specify the YaccKind of a grammar and then the ActionKind of actions. This is unnecessarily fiddly, so remove ActionKind entirely and instead flesh out YaccKind to deal with the possible variants. For example ActionKind::CustomAction is now, in essence, YaccKind::Original(YaccOriginalActionKind::UserAction). This is a breaking change but one that will make future evolution much easier.

• The %type directive in grammars exposed by YaccKind::Original(YaccOriginalActionKind::UserAction) has been renamed to %actiontype to make it clear what type is being referred to. In general, most people will want to move to the YaccKind::Grmtools variant (see below), which doesn't require the %actiontype directive.

Major changes

• grmtools has moved to the 2018 edition of Rust and thus needs rustc-1.31 or later to compile.

• Add YaccKind::Grmtools variant, allowing grammar rules to have different action types. For most practical use cases, this is much better than using %actiontype.

• Add %avoid_insert directive to bias ranking of repair sequences and make it more likely that parsing can continue.

Minor changes

• Add -q switch to nimbleparse to suppress printing out the stategraph and conflicts (some grammars have conflicts by design, so being continually reminded of it isn't helpful).

• Fix problem where errors which lead to vast (as in hundreds of thousands) of repair sequences being found could take minutes to sort and rank.

• Add YaccKind::Original(YaccOriginalActionKind::NoAction) variant to generate a parser which simply tells the user where errors were found (i.e. no actions are executed, and not even a parse tree is created).

• lrlex no longer tries to create Rust-level identifiers for tokens whose names can't be valid Rust identifiers (which led to compile-time syntax errors in the generated Rust code).

grmtools 0.1.1 (2018-12-18)

• Fix bug where %epp strings with quote marks in caused a code-generation failure in compile-time mode.

grmtools 0.1.0 (2018-12-13)

First stable release.