F_Front: lexing bug

[skosukhin]

The Fortran frontend fails to parse the following:

subroutine a(b)
  ; &
integer b
end subroutine

"./test.f90", line 2: bad char &(0x26)
"./test.f90", line 2: syntax error

[m214089]

This is, as far as I know, invalid Fortran ...

[clementval]

What is it suppose to do? I would say that code looking like that should be clean up first :-)

[skosukhin]

This is just a small snippet that is supposed to reproduce the problem. Maybe it's too small. Let's consider the following:

subroutine a(b)
 integer b ; &
&integer c
end subroutine

I've managed to compile this with different versions of gfortran, ifort, nagfor, pgfortran, ftn (Cray), and only intel smelled something bad (but with just a warning):

./test.f90(2): warning #5151: Continuation character illegal as first non_blank in statement
 integer b ; &
-------------^

F_Front, however, chokes with this piece of code. As far as I understand, once it sees ;, it expects a valid statement afterwards. A valid statement can't start with & and we get an error. However, it looks like the treatment of line continuations should have priority.

The code does seem very unhealthy but this is a possible result of the preprocessing. Say, we have a macro that contains multiple statements separated with ; and the macro expands to a very long string:

#define MANY_CALLS \
    call banana(); \
    call apple(); \
    call orange();
MANY_CALLS

Some Fortran preprocessors (e.g. NAG) break long lines with & and instead of:

call banana();     call apple();     call orange();

we might get something like:

call banana();     call apple();   &
& call orange();

In the end, I don't really know whether it is correct to expect a successful compilation of such code from F_Fortran.

[clementval]

My guess is that the lever is waiting for something after the ; and fails if '&' is encountered. I'll have a quick look.