Hello, The code example listed at the end of this email fails to compile with gfortran 4.1.0 20051025 (experimental). Compiling like so: gfortran -c gfortran_test.f90 produces the error message: In file gfortran_test.f90:16 END INTERFACE OPERATOR (.EqualTo.) 1 Error: Expecting 'END INTERFACE OPERATOR (.compare_float.)' at (1) This problem with recognizing the end interface statement flows on and produces a whole slew of (spurious) error messages on valid code. If I change the source code line END INTERFACE OPERATOR (.EqualTo.) to simply END INTERFACE the code compiles just fine. However, I believe that using the syntax of END INTERFACE OPERATOR (.EqualTo.) is standard Fortran95. Thanks, paulv ------<<cut here for gfortran_test.f90>>------ MODULE Compare_Float_Numbers IMPLICIT NONE PRIVATE PUBLIC :: Compare_Float PUBLIC :: OPERATOR (.EqualTo.) INTERFACE Compare_Float MODULE PROCEDURE Compare_Float_Single MODULE PROCEDURE Compare_Float_Double END INTERFACE Compare_Float INTERFACE OPERATOR (.EqualTo.) MODULE PROCEDURE Is_Equal_To_Single MODULE PROCEDURE Is_Equal_To_Double END INTERFACE OPERATOR (.EqualTo.) INTEGER, PARAMETER :: Single = SELECTED_REAL_KIND(6) ! Single precision INTEGER, PARAMETER :: Double = SELECTED_REAL_KIND(15) ! Double precision CONTAINS ! -- Is Equal To ELEMENTAL FUNCTION Is_Equal_To_Single( x, y ) RESULT( Equal_To ) REAL( Single ), INTENT( IN ) :: x, y LOGICAL :: Equal_To Equal_To = ABS( x - y ) < SPACING( MAX(ABS(x),ABS(y)) ) END FUNCTION Is_Equal_To_Single ELEMENTAL FUNCTION Is_Equal_To_Double( x, y ) RESULT( Equal_To ) REAL( Double ), INTENT( IN ) :: x, y LOGICAL :: Equal_To Equal_To = ABS( x - y ) < SPACING( MAX(ABS(x),ABS(y)) ) END FUNCTION Is_Equal_To_Double ! -- General floating point comparison ELEMENTAL FUNCTION Compare_Float_Single( x, y, ulp ) RESULT( Compare ) REAL( Single ), INTENT( IN ) :: x REAL( Single ), INTENT( IN ) :: y INTEGER, OPTIONAL, INTENT( IN ) :: ulp LOGICAL :: Compare REAL( Single ) :: Rel Rel = 1.0_Single IF ( PRESENT( ulp ) ) THEN Rel = REAL( ABS(ulp), Single ) END IF Compare = ABS( x - y ) < ( Rel * SPACING( MAX(ABS(x),ABS(y)) ) ) END FUNCTION Compare_Float_Single ELEMENTAL FUNCTION Compare_Float_Double( x, y, ulp ) RESULT( Compare ) REAL( Double ), INTENT( IN ) :: x REAL( Double ), INTENT( IN ) :: y INTEGER, OPTIONAL, INTENT( IN ) :: ulp LOGICAL :: Compare REAL( Double ) :: Rel Rel = 1.0_Double IF ( PRESENT( ulp ) ) THEN Rel = REAL( ABS(ulp), Double ) END IF Compare = ABS( x - y ) < ( Rel * SPACING( MAX(ABS(x),ABS(y)) ) ) END FUNCTION Compare_Float_Double END MODULE Compare_Float_Numbers ------<<cut here for gfortran_test.f90>>------
Here's a reduced code that shows the problem. Gfortran is not handling the END INTERFACE OPERATOR (.EqualTo.) correctly. This confuses the heck out of the error recovery code. MODULE Compare_Float_Numbers IMPLICIT NONE INTERFACE Compare_Float MODULE PROCEDURE Compare_Float_Single END INTERFACE Compare_Float INTERFACE OPERATOR (.EqualTo.) MODULE PROCEDURE Is_Equal_To_Single END INTERFACE OPERATOR (.EqualTo.) CONTAINS FUNCTION Is_Equal_To_Single(x, y) RESULT(Equal_To) REAL(4), INTENT(IN) :: x, y LOGICAL :: Equal_To Equal_To = .true. END FUNCTION Is_Equal_To_Single FUNCTION Compare_Float_Single(x, y) RESULT(Compare) REAL(4), INTENT(IN) :: x, y LOGICAL :: Compare Compare = .true. END FUNCTION Compare_Float_Single END MODULE Compare_Float_Numbers
Perhaps this cures it. Index: interface.c =================================================================== RCS file: /cvs/gcc/gcc/gcc/fortran/interface.c,v retrieving revision 1.21 diff -u -3 -p -r1.21 interface.c --- interface.c 21 Oct 2005 18:50:52 -0000 1.21 +++ interface.c 26 Oct 2005 18:53:39 -0000 @@ -295,7 +295,7 @@ gfc_match_end_interface (void) /* Comparing the symbol node names is OK because only use-associated symbols can be renamed. */ if (type != current_interface.type - || strcmp (current_interface.sym->name, name) != 0) + || strcmp (current_interface.uop->name, name) != 0) { gfc_error ("Expecting 'END INTERFACE OPERATOR (.%s.)' at %C", current_interface.sym->name);
Subject: Bug 24545 Author: kargl Date: Fri Oct 28 20:05:56 2005 New Revision: 105953 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=105953 Log: PR fortran/24545 * interface.c (gfc_match_end_interface): Fix typo in INTERFACE_USER_OP case. Added: trunk/gcc/testsuite/gfortran.dg/interface_2.f90 Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/interface.c trunk/gcc/testsuite/ChangeLog
Subject: Re: gfortran bug regarding interface block with named END INTERFACE statements On Wed, Oct 26, 2005 at 06:54:10PM -0000, steven at gcc dot gnu dot org wrote: > > Index: interface.c > =================================================================== > RCS file: /cvs/gcc/gcc/gcc/fortran/interface.c,v > retrieving revision 1.21 > diff -u -3 -p -r1.21 interface.c > --- interface.c 21 Oct 2005 18:50:52 -0000 1.21 > +++ interface.c 26 Oct 2005 18:53:39 -0000 > @@ -295,7 +295,7 @@ gfc_match_end_interface (void) > /* Comparing the symbol node names is OK because only use-associated > symbols can be renamed. */ > if (type != current_interface.type > - || strcmp (current_interface.sym->name, name) != 0) > + || strcmp (current_interface.uop->name, name) != 0) > { > gfc_error ("Expecting 'END INTERFACE OPERATOR (.%s.)' at %C", > current_interface.sym->name); > Bootstrapped and regression tested on amd64-*-freebsd. I've committed the patch.
Subject: Bug 24545 Author: kargl Date: Fri Oct 28 20:57:17 2005 New Revision: 105962 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=105962 Log: PR fortran/24545 * interface.c (gfc_match_end_interface): Fix typo in INTERFACE_USER_OP case. Added: branches/gcc-4_0-branch/gcc/testsuite/gfortran.dg/interface_2.f90 Modified: branches/gcc-4_0-branch/gcc/fortran/ChangeLog branches/gcc-4_0-branch/gcc/fortran/interface.c branches/gcc-4_0-branch/gcc/testsuite/ChangeLog
Fixed.