Scraping the EBNF grammar from the ISO Spec

Fortran Standards Documents and Resources

• Distilled from https://gcc.gnu.org/wiki/GFortranStandards
• All drafts in PDFs contain LHS margin with line numberings.

Fortran 90

• https://wg5-fortran.org/N001-N1100/N692.pdf

Fortran 95

• https://wg5-fortran.org/N1151-N1200/N1191.pdf

Fortran 2003

• https://j3-fortran.org/doc/year/04/04-007.pdf

Fortran 2008

• https://j3-fortran.org/doc/year/10/10-007.pdf

Fortran 2018

• https://j3-fortran.org/doc/year/18/18-007r1.pdf

Fortran 2023

• https://j3-fortran.org/doc/year/23/23-007r1.pdf

Official ISO Spec

• https://www.iso.org/standard/82170.html

Testsuites

• https://github.com/fujitsu/compiler-test-suite/tree/main/Fortran
• https://fortran-lang.discourse.group/t/fortran-compiler-testing-framework/1573
• https://github.com/scivision/fortran2018-examples
• https://github.com/fortran-lang/test-drive
• https://github.com/llvm/llvm-test-suite/tree/main/Fortran/gfortran
• https://github.com/OpenFortranProject/open-fortran-parser

Notes on scraping

• From ISO/IEC 1539-1:2023: The syntax rules are not a complete and accurate syntax description of Fortran, and cannot be used to generate a Fortran parser automatically; where a syntax rule is incomplete, it is restricted by corresponding constraints and text.
• The main scraping script is extract.sh. It calls tritext to yank out text from a PDF file. Then, it calls a custom-written program just for extracting the EBNF from the text, extraction. The script also uses an ENBF in Antlr to parse and modify the scraped Fortran EBNF.
• There are many rules that end in -list. These are currently enumerated in the extract.sh script, but the rules should be generated.
• The grammar generated is based on https://github.com/AkhilAkkapelli/Fortran2023Grammar. But, there are several issues that need explanation.
  • NAME is an identifer. PROGRAM is a keyword. There is no distinguishment between keywords and non-keywords, but there should be. In other words, NAME can remain NAME because it is that way in the spec, but PROGRAM should be KW_PROGRAM in order to distinguish the two uses.
  • NAME is defined here: https://github.com/AkhilAkkapelli/Fortran2023Grammar/blob/553123a023f70e9a524e2a4036be128978834c42/Fortran2023Lexer.g4#L502. But, keyword "NAME" is define using NAAM here. https://github.com/AkhilAkkapelli/Fortran2023Grammar/blob/553123a023f70e9a524e2a4036be128978834c42/Fortran2023Lexer.g4#L152. It needs standardization.
  • Beyond camel-case naming, program : program_unit ( program_unit )*; contains useless parentheses. These need to be removed.
  • The order of the parser rules does not correspond to the spec.
  • mult_op : ASTERIK | SLASH ;. Terminal symbols should use the established Unicode character name. '*' is not named ASTERIK, but Asterisk.