[flang] Inconsistent compilation results for programs with almost the same error

[k-arrows]

Consider the following three programs. These are all invalid.

test1.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-1111111111111111111:4)
end

test2.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-11111111111111111111:4)
end

test3.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-111111111111111111111:4)
end

Gfortran rejects the above three programs with the same message.

$ gfortran test1.f90
test1.f90:3:16:

    3 | print*, "s:", s(-1111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)
$ gfortran test2.f90
test2.f90:3:16:

    3 | print*, "s:", s(-11111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)
$ gfortran test3.f90
test3.f90:3:16:

    3 | print*, "s:", s(-111111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)

With flang-new, I got the following results.

$ flang-new test1.f90
error: Semantic errors in test1.f90
./test1.f90:3:15: error: Substring must begin at 1 or later, not -1111111111111111111
  print*, "s:", s(-1111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^
$ flang-new test2.f90
./test2.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-11111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^
$ ./a.out
 s:
$ flang-new test3.f90
error: Semantic errors in test3.f90
./test3.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: error: Substring must begin at 1 or later, not -430646668853801415
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: error: Substring must begin at 1 or later, not -430646668853801415
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^

It seems more consistent for test2.f90 and test3.f90 to be rejected with the same error message as test1.f90. Also, the error message for test3.f90 is incorrect because the number -430646668853801415 does not exist in test3.f90.

[llvmbot]

@llvm/issue-subscribers-flang-frontend

Author: None (k-arrows)

Consider the following three programs. These are all invalid.

test1.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-1111111111111111111:4)
end

test2.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-11111111111111111111:4)
end

test3.f90

character(len=5) :: s
s = "flang"
print*, "s:", s(-111111111111111111111:4)
end

Gfortran rejects the above three programs with the same message.

$ gfortran test1.f90
test1.f90:3:16:

    3 | print*, "s:", s(-1111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)
$ gfortran test2.f90
test2.f90:3:16:

    3 | print*, "s:", s(-11111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)
$ gfortran test3.f90
test3.f90:3:16:

    3 | print*, "s:", s(-111111111111111111111:4)
      |                1
Error: Syntax error in PRINT statement at (1)

With flang-new, I got the following results.

$ flang-new test1.f90
error: Semantic errors in test1.f90
./test1.f90:3:15: error: Substring must begin at 1 or later, not -1111111111111111111
  print*, "s:", s(-1111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^
$ flang-new test2.f90
./test2.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-11111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^
$ ./a.out
 s:
$ flang-new test3.f90
error: Semantic errors in test3.f90
./test3.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: error: Substring must begin at 1 or later, not -430646668853801415
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: warning: INTEGER(16) to INTEGER(8) conversion overflowed
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^
./test3.f90:3:15: error: Substring must begin at 1 or later, not -430646668853801415
  print*, "s:", s(-111111111111111111111:4)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^

It seems more consistent for test2.f90 and test3.f90 to be rejected with the same error message as test1.f90. Also, the error message for test3.f90 is incorrect because the number -430646668853801415 does not exist in test3.f90.

[klausler]

I'm looking into this. The basic outcome is correct but the mysterious integer value in the message can be improved.

[klausler]

111111111111111111111 decimal is 605F9F6DD18BC71C7 hexadecimal; the least-order 64 bits of that value are 5F9F6DD18BC71C7 hexadecimal, which is the mysterious decimal value 430646668853801415 appearing in the messages.

[klausler]

#110031

[klausler]

The fix has been merged.