[flang] Failed substitution with preprocessor

[bcornille]

Issue:

flang does not substitute a preprocessor variable in a snake case Fortran variable. In order to get substitution some codes use continuation lines so that the preprocessor variable can be inserted in-between underscores in code. flang appears to concatenate continuation lines prior to applying the preprocessor substitutions so the substitution does not occur.

Reproducer:

main.F90

program main
  implicit none
#define LETTER b
  integer, parameter :: a_&
    &LETTER&
    &_c = 1
  write(*,*) a_b_c
end program main

compilation:

>> flang-new -c main.F90
error: Semantic errors in main.F90
./main.F90:8:14: error: No explicit type declared for 'a_b_c'
    write(*,*) a_b_c
               ^^^^^

preprocessed source:

>> flang-new -E main.F90
#line "./main.F90" 1
      program main
      implicit none
      integer, parameter :: a_LETTER_c = 1
      write(*,*) a_b_c
      end program main

GNU behavior:

>> gfortran -E main.F90
# 1 "main.F90"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "main.F90"
program main
  implicit none
  integer, parameter :: a_&
    &b&
    &_c = 1
  write(*,*) a_b_c
end program main

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: Brian Cornille (bcornille)

### Issue: flang does not substitute a preprocessor variable in a snake case Fortran variable. In order to get substitution some codes use continuation lines so that the preprocessor variable can be inserted in-between underscores in code. flang appears to concatenate continuation lines prior to applying the preprocessor substitutions so the substitution does not occur.

Reproducer:

main.F90

program main
  implicit none
#define LETTER b
  integer, parameter :: a_&
    &LETTER&
    &_c = 1
  write(*,*) a_b_c
end program main

compilation:

>> flang-new -c main.F90
error: Semantic errors in main.F90
./main.F90:8:14: error: No explicit type declared for 'a_b_c'
    write(*,*) a_b_c
               ^^^^^

preprocessed source:

>> flang-new -E main.F90
#line "./main.F90" 1
      program main
      implicit none
      integer, parameter :: a_LETTER_c = 1
      write(*,*) a_b_c
      end program main

GNU behavior:

>> gfortran -E main.F90
# 1 "main.F90"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "main.F90"
program main
  implicit none
  integer, parameter :: a_&
    &b&
    &_c = 1
  write(*,*) a_b_c
end program main

[klausler]

That's indeed how it works in f18; continuation lines are combined prior to macro substitution, so that macro references can span multiple lines.

[bcornille]

So I guess due to that design decision there is no way to get macro substitution in the middle of a variable as demonstrated by the example code, correct? A macro function would probably need to be used instead to construct the composite name.

[klausler]

Token pasting with ## works. In fixed-form code, just use spaces between the parts of the name.

[klausler]

I'm looking into tweaking the interaction between continuation line joining and preprocessing a little to accommodate this case. In short, I may be able to allow the pp-tokens that constitute a macro reference to span multiple continuation lines, but not the individual identifier tokens themselves.

[klausler]

#78868

[klausler]

Patch has been merged into llvm-project/main.