Fixes

• C/C++/Java: escape newline etc. when printing String and Char literals [#449]
• Java/ANTLR: unescape String and Char literals in parser (needs Java ≥ 15) [#451]
• Java/ANTLR: fix case problem with language names like C [#455]
• Java with line numbers: compatibility with jflex ≥ 1.8 [#453]
• Haskell/GADT: generated ComposOp.hs no longer needs mtl library [#438]
• Ocaml: fixed a crash in printer with unbalanced } [#439]
• Ocaml: lex escape sequences in Char [#452]
• Ocaml: removed superfluous let rec in the printers for token categories

Tested GHC versions:

• with cabal, GHC 7.10 - 9.6
• with stack, GHC 8.2 - 9.6

What's Changed (autogenerated from PRs)

• Haskell/GADT: produce mtl-2.3 compatible code by @andreasabel in #438
• Fix #439: Ocaml/printer: catch negative indentation by @andreasabel in #440
• Fix: #437: enable doctest for GHC 9.6.1 in Haskell-CI by @andreasabel in #441
• Simplify definition of internal function columns by @ivanperez-keera in #442
• Fix for Dockerfile smell DL3059 by @grosa1 in #444
• Haskell/Printer: showString t [] is just t by @andreasabel in #446
• various: fix typos by @p-linnane in #448

New Contributors

• @ivanperez-keera made their first contribution in #442
• @grosa1 made their first contribution in #444
• @p-linnane made their first contribution in #448

Full Changelog: v2.9.4.1...v2.9.5

• C/C++: allow space characters in token definitions again [#431]
  (regression in 2.9.0)
• installation: get rid of cabal-doctest dependency [#429]

Tested GHC versions:

• with cabal, GHC 7.10 - 9.4
• with stack, GHC 8.2 - 9.4

Bug fixes and improvements

• LBNF: empty tokens types are now forbidden [#388]
• Agda: support position information via --functor [#405]
• C/C++: use size_t and -Wsign-conversion [#391]
• C++: repair broken --line-numbers [#390], regression in 2.9.2 by [#349]
• Haskell: fix a problem with layout stop words and top-level layout [#399,#413]
• Haskell: generated test parser can parse several files now [#400]
• Java: use L.valueOf() instead of deprecated new L() for literal classes L [#402]
• Ocaml: non-terminals in generated parser are now type-annotated [#407]
• Ocaml: sanitize bound variables in define
• Ocaml/Menhir: update parse error mechanism to Menhir 2021/12/30 [#414]

Contributors:

• Michał Radwański [#390,#391]
• Beatrice Vergani [#399,#400]

Bug fixes and small improvements

• BNFC now uniformly signs generated files with its version number [#373]
• C/C++: include stdio.h in parser header files [#381]
• C++: fixed parser regression in 2.9.2: missing #include <algorithm> [#377]
• Ocaml: lex CR as whitespace [see also #376]
• Ocaml: correct position in parse errors [#380]
• Ocaml/Haskell: make printer for lists categories total [#383]

64bit binaries for Linux, macOS, and Windows below.

Major improvements

• Haskell: layout keywords can now be stacked on the same line [#354],
  see https://bnfc.readthedocs.io/en/latest/lbnf.html#stacking-layout-keywords
• C: new methods free_* and clone_* to deallocate and clone syntax trees [#348]
• C/C++ backends now create reentrant parsers [#349]

Bug fixes and small improvements

• Haskell-GADT: generated code is warning free [#346]
• Haskell: fixes in layout preprocessor [#343,#344,#345,#352,#353]
• Haskell: print [Char] correctly, removed method prtList [#359]
• Haskell: added missing import [#368], regression introduced in 2.9.1 by [#331]
• C: fixed a space leak when parsing from a string in memory [#347]
• C: removed errorneous defined constructors from Skeleton.c
• C++: defined constructors now reside in Absyn [#287]
• Java: defined constructor now reside in AbsynDef.java [#287]
• Ocaml: fixed translation of nested defined constructors
• C/C++/Java: Pre/post/mixfix lists are now printed correctly [#358]
• all: defined constructors involving list expressions work now [#363]
• all: printers render braces on their own line [#366]

Cosmetical changes

• C/C++: instead of _SYMB_nnn, more readable token names in lexer & parser

Main new feature

• Haskell: the --functor option now produces position-annotated ASTs [#176,#327]. Thanks @Commelina!

Bug fixes and small improvements

• Haskell: fix generated Makefile and test parser for --glr mode [#340]
• Haskell(/GADT): generated modules import Prelude explicitly, compatible with {-# LANGUAGE NoImplicitPrelude #-}
• Haskell: generated code is warning free [#331]
• Haskell: generated printer more robust wrt. identifier clashes [#337]
• Haskell/C: handle token constructors in define expressions [#338]
• Java/ANTLR: removed more superfluous quotation in lexer character sets [#329]
• Ocaml: fix syntax error in generated printer [#330]
• LBNF: more sanity checks [#339]
• Tested with GHC 9.0

Major changes

• New license: BSD 3-clause [#308]
• LBNF: removed delimiters pragma [#308]
• Haskell: removed options --alex1, --alex2, --sharestrings, --profile, and --cnf [#265]
• C#: backend removed [#265]

Bug fixes

• LBNF: allow list categories in entrypoints pragma [#306]
• LBNF: report clashes between token and ordinary categories [#323]
• C: strdup is not part of C89 standard, _POSIX_C_SOURCE required [#318]
• C/C++: fixed buffer overrun in String literal lexer [#316]
• C++: fixed regressions (one of them #310) introduced by #288
• C/C++/OCaml: allow unicode characters in token definitions [#324]
• C/OCaml: sanitize grammar file names [#325]
• Java/ANTLR: removed superfluous quotation in lexer character sets [#319]

This release adds the OCaml/Menhir backend, a variant of the OCaml backend that uses parser generator menhir instead of ocamlyacc. Beyond that, the focus of this release has been increased robustness and predictability of the BNFC tool:

• New option --check that only performs sanity checks on the given LBNF grammar file, without calling any backend (#286).
• New sanity checks (#186, #213, #214) and deprecation warnings.
• Error messages of BNFC concerning rule or category names are now equipped with the error location.
• Generated code systematically tries to avoid clashes with language keywords or standard library names (#278, #289).
• Pragma define now supported by all maintained backends (Agda, Haskell, Haskell/GADT, C, C++, Java).
• Pragma position token now supported by all maintained backends with the exception of C++/NoSTL.
• List separators consisting just of whitespace are now accepted. They are used in the generated printers but treated like "" in the generated parsers (#70).

The Haskell backend has seen the following improvements:

• New option --text-token to use Data.Text instead of String in the generated lexer (#167).
• Deriving IsString instances for Ident-like token types (#192).
• The Err monad is now defined as Either String. Module ErrM is only generated for backwards compatibility. Imports of ErrM with an explicit import list may have to modified slightly, see a migration guide at the Haskell backend documentation.
• Block comments delimiters are no longer restricted to 2 characters.

The C-family backends have been improved as follows:

• C, C++, Java: Now multiple block comment forms per grammar are allowed (#202).
• C++/STL: The parser now throws an exception when parsing fails (#288).
• Java/JFlex: Special characters are properly escaped (#299).
• Java/ANTLR: Start rules are generated to work around ANTLR issue #2689 (#272).

Further bug fixes: #163, #169, #196, #212, #235, #256, #264, #269, #274, #276, #277, #279, #282, #283, #290.

Building of BNFC 2.8.4 requires GHC ≥ 7.10 and has been tested up to GHC 8.10.2. BNFC can be build using cabal or stack (using one of the enclosed stack-x.y.z.yaml files).

Main new feature:

• Agda backend: generates bindings to output of Haskell backend

Backend-independent features:

• GHC 8.8 compatibility
• Stack installation supported by provided .yaml files [#198]
• Unicode support in lexer, excl. C, C++ and Ocaml [#249]
• LBNF: support \r (carriage return) and \f (form feed) in token declaration [#257]
• LBNF: allow numbered categories (like Foo2) independent of coercions [#210]

C backend:

• supports now the -l option to give error locations [#238]
• correct function names in generated skeletons [#239]
• handle regular expression difference (subtraction) correctly [#237]

Haskell backend:

• generates now more fine-grained Makefile that invokes bnfc on changed .cf file
• uses qualified import of AST in generated printer [fixes #128,#228]
• printer code no longer produces deprecation warning concerning OverlappingInstances [#233]
• --cnf: fixed problem with Any type in generated code [#216]
• --cnf: generated test program now same name (Test) as without --cnf
• --glr: correct module header in .y file [#252]

Binaries (see below):

• bnfc-2.8.3-mac.binary: Binary compiled under Mac OS X Mojave, x86 64bit processor
• bnfc.exe: Binary compiled under Windows 10, x86 64bit processor

Main new feature:

• Java: An experimental ANTLR4 backend [#155]

Backend-independent features:

• GHC 8.4 compatibility [#227,#236]
• bnfc now puts current timestamp on all created files, even unchanged ones [#219]
• bnfc is now more accepting about border-line terminals [#149]
• Improved error messages [#144] in particular on undefined categories [#151]

C and C++ backends:

• C: Emit correct function prototypes [#185]
• C++: Fix buffer overrun in pretty printer [#242]
• C++: Fix regression introduced in 2.8 in Skeleton.H [#164]
• C++: Replace %name-prefix with %define api.prefix in bison files [#181]
• C++: Fix a bug that appeared if you had a category named "List"
• C, C++: Add usage output to the test program [#141]
• C, C++: Fix a bug in the parser file when using -p [#172]
• C, C++, OCaml, Java: Accept ' or " as comment delimiters [#146]

Haskell backend:

• Generated code mostly hlint-warning free
• Small fixes [#166,#170,#222]

Other backends:

• Java: Add support for passing line numbers to parser [#217,#224,#226]
• OCaml: Reserved symbols now have a higher precedence than predefined tokens as in Haskell

Further:

• Some updates of the documentation [#211,#223]
• And various small fixes [#139,#159,#195,#201,#215]

Thanks to the many contributers to this version!