[Bug fortran/64517] New: Inconsistent behavior when mixing -E and -M

[Bug fortran/64517] New: Inconsistent behavior when mixing -E and -M

[thfanning at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64517

            Bug ID: 64517
           Summary: Inconsistent behavior when mixing -E and -M
           Product: gcc
           Version: 4.9.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thfanning at gmail dot com

According to documentation, "Passing -M to the driver implies -E". However an
explicit `-E` produces different behavior.

  touch test.F90
  gfortran -M test.F90

produces:

  test.o: test.F90

However,

  gfortran -E -M test.F90

produces

  # 1 "test.F90"
  # 1 "<built-in>"
  # 1 "<command-line>"
  # 1 "test.F90"

  test.o: test.F90

As a result, preprocessor output is mixed with dependency output. A kludge is
to add `-o /dev/null`, but that is ugly.

The description for `-E` states: "The output is in the form of preprocessed
source code, which is sent to the standard output." But shouldn't `-M` change
*what* is sent to stdout?

[Bug fortran/64517] Inconsistent behavior when mixing -E and -M

[thfanning at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64517

--- Comment #1 from Thomas Fanning <thfanning at gmail dot com> ---
There is a bigger problem that I didn't immediately recognize with the empty
file: Using `-M` results in gfortran trying to compile the source. I.e., it
does *not* stop after preprocessing, which it should if `-E` is implied.

[Bug preprocessor/64517] Inconsistent behavior when mixing -E and -M

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64517

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2015-01-09
          Component|fortran                     |preprocessor
     Ever confirmed|0                           |1

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Fist the -E and -M options are not gfortran specific (see
https://gcc.gnu.org/onlinedocs/gcc/Preprocessor-Options.html#Preprocessor-Options).

For the following C code

 cat vbmi.c
        typedef char __v64qi __attribute__ ((__vector_size__ (64)));
        __v64qi
        _mm512_multishift_epi64_epi8 (__v64qi __X, __v64qi __Y)
        {
          return (__v64qi) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi)
__X,
                                                                 (__v64qi) __Y,
                                                                 (__v64qi) __Y,
                                                                 -1);
        }

gcc vbmi.c -M

gives

vbmi.o: vbmi.c

gcc vbmi.c -E

gives

# 1 "vbmi.c"
# 1 "<built-in>"
# 1 "<command-line>"
# 1 "vbmi.c"
        typedef char __v64qi __attribute__ ((__vector_size__ (64)));
        __v64qi
        _mm512_multishift_epi64_epi8 (__v64qi __X, __v64qi __Y)
        {
          return (__v64qi) __builtin_ia32_vpmultishiftqb512_mask ((__v64qi)
__X,
                                                                 (__v64qi) __Y,
                                                                 (__v64qi) __Y,
                                                                 -1);
        }

and

gcc vbmi.c -E -M

gives

vbmi.o: vbmi.c

So it seems that the sentence

Passing -M to the driver implies -E, ...

is not accurate. Thus I am moving the PR to the 'preprocessor' component.

> There is a bigger problem that I didn't immediately recognize with
> the empty file: Using `-M` results in gfortran trying to compile
> the source. I.e., it does *not* stop after preprocessing, which
> it should if `-E` is implied.

I don't see that (at least using -E on a code generating *.mod files when
compiled does not generate these files).

Note that the use of preprocessor with fortran leads to several PRs: see
pr47485, pr47486, pr47720, and pr49149.

[Bug preprocessor/64517] Inconsistent behavior when mixing -E and -M

[thfanning at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64517

--- Comment #3 from Thomas Fanning <thfanning at gmail dot com> ---
In gcc-4.9.2, using both -M and -E produce the expected output: dependencies
only. But in gfortran, both preprocessed output and dependencies are produced.
This is incorrect.

It seems to me that the gfortran driver is not setting up the preprocessing
environment correctly, so I'm not sure this is strictly a preprocessor problem.

For the following C code: preprocess.c

  // It's me!
  #include "UTIL_System.h"

  int main()
  {
    return 0;
  }

gcc -M preprocess.c
--> preprocess.o: preprocess.c UTIL_System.h  <-- dependencies

gcc -M -E preprocess.c
--> preprocess.o: preprocess.c UTIL_System.h  <-- dependencies

Since GNU make defines CPP = '$(CC) -E', it is essential that combining -E and
-M are compatible.

For the following gfortran file however: preprocess.F90

  ! It's me!
  program test
  #include "UTIL_System.h"
  print *, "Hello, World!"
  end program

gfortran -M preprocess.F90
--> preprocess.o: preprocess.F90 UTIL_System.h  <-- dependencies

But gfortran -M -E preprocess.F90
--> # 1 "preprocess.F90"
--> # 1 "<built-in>"
--> # 1 "<command-line>"
--> # 1 "preprocess.F90"
--> ! It's me!
--> program test
--> 
--> # 1 "UTIL_System.h" 1
--> # 4 "preprocess.F90" 2
--> print *, "Hello, World!"
--> end program
--> preprocess.o: preprocess.F90 UTIL_System.h  <-- dependencies