Hello everyone

When I try to compile the code from the master branch on my Manjaro, one of the Linux system based on the Arch and based on the glibc-2.31 && gcc-10.1.0 . the compiler show me that the fcntl module has been failed to be compiled

Here's the message 

/home/manjusaka/Documents/project/cpython/Modules/fcntlmodule.c:618:33: error: ‘I_PUSH’ undeclared (first use in this function)
  618 |     if (PyModule_AddIntMacro(m, I_PUSH)) return -1;
      |                                 ^~~~~~
./Include/modsupport.h:146:67: note: in definition of macro ‘PyModule_AddIntMacro’
  146 | #define PyModule_AddIntMacro(m, c) PyModule_AddIntConstant(m, #c, c)
      |                                                                   ^
/home/manjusaka/Documents/project/cpython/Modules/fcntlmodule.c:618:33: note: each undeclared identifier is reported only once for each function it appears in
  618 |     if (PyModule_AddIntMacro(m, I_PUSH)) return -1;
      |                                 ^~~~~~
./Include/modsupport.h:146:67: note: in definition of macro ‘PyModule_AddIntMacro’
  146 | #define PyModule_AddIntMacro(m, c) PyModule_AddIntConstant(m, #c, c)
      |                                                                   ^
/home/manjusaka/Documents/project/cpython/Modules/fcntlmodule.c:619:33: error: ‘I_POP’ undeclared (first use in this function)
  619 |     if (PyModule_AddIntMacro(m, I_POP)) return -1;
      |                                 ^~~~~
./Include/modsupport.h:146:67: note: in definition of macro ‘PyModule_AddIntMacro’
  146 | #define PyModule_AddIntMacro(m, c) PyModule_AddIntConstant(m, #c, c)
      |                                                                   ^
/home/manjusaka/Documents/project/cpython/Modules/fcntlmodule.c:620:33: error: ‘I_LOOK’ undeclared (first use in this function); did you mean ‘F_LOCK’?
  620 |     if (PyModule_AddIntMacro(m, I_LOOK)) return -1;
      |                                 ^~~~~~


I have figured out the reason because the stropts.h has been deprecated since the glic-2.30, but some of the distribution of Linux keep an empty file on the /usr/include, so the configure process will recognize stropts.h is existed and open HAVE_STROPTS_H flag.

So should we add an extra content check in the configure process to avoid the empty file problem?

Here's the reference

https://sourceware.org/legacy-ml/libc-alpha/2019-08/msg00029.html

Yes, I have the same problem. The empty file is needed to avoid compilation errors in certain builds, but in this case it creates an error...

I think adding this code in the `configure` script may fix it:

as_ac_Header=`$as_echo "ac_cv_header_stropts.h" | $as_tr_sh`
ac_fn_c_check_header_mongrel "$LINENO" "stropts.h" "$as_ac_Header" "$ac_includes_default"
if eval test \"x\$"$as_ac_Header"\" = x"yes" && [ -s stropts.h ]; then :
  cat >>confdefs.h <<_ACEOF
#define `$as_echo "HAVE_stropts.h" | $as_tr_cpp` 1
_ACEOF
fi

here: https://github.com/akulakov/cpython/blob/8f24b7dbcbd83311dad510863d8cb41f0e91b464/configure#L8144

.. but I wasn't able to test it. I have a MacOS system and for me even with a non-empty stropts.h, configure does not pick it up.

The configure script is auto-generated from configure.ac. Any chance must be applied to configure.ac.

We could replace the AC_CHECK_HEADERS() for stropt.h with a more elaborate AC_COMPILE_IFELSE() block that checks for presence of I_PUSH.

Could you please report the issue to your Linux vendor, too? The empty file seems to break Open Group Unix 98 standard.

Rather of using AC_COMPILE_IFELSE() in configure.ac, would not be easier to use #ifdef in the C file?