In the C code of Python 3.5, there are still preprocessor commands checking for defines:

- __BORLANDC__: Borland C compiler
- __WATCOMC__: Watcom C compiler
- __DJGPP__: MS-DOS port of GCC

In 2014, it's time to drop this old code. OS/2 support was already removed in Python 3.4 (see the PEP 11).

I'm quite sure that Python 3.5 cannot be compiled on MS-DOS and/or will not work on MS-DOS. I don't think that anyone tried Python 3 on MS-DOS. The MS-DOS support was simply inherited from Python 2, Python 3 source code is based on Python 2.

I don't want to drop support of custom compilers, only removing code specific to MS-DOS, see:
http://bugs.python.org/issue22579#msg228908

AFAIK Borland and Watcom compilers supported Windows.

> AFAIK Borland and Watcom compilers supported Windows.

It looks like Borland C++ Builder 5.5 was released in 2000, 14 years ago. Yes, it supports Windows and MS-DOS, but do we really want to support this very old proprietary compiler?

For Watcom, it's a different story. The compiler was released under an open source license. It's now hosted at github and the project is actively developed (last commit a few hours ago):
https://github.com/open-watcom/open-watcom-v2/

It would be interesting to support OpenWatcom compiler on Windows to not depend on the proprietary Microsoft compiler.

> AFAIK Borland and Watcom compilers supported Windows.

AFAIU Victor only proposes to remove the MSDOS-specific conditionals. But it's hard to tell, since he didn't post a patch ;-)

Here is my huge patch :-) It only changes osdefs.h.

In fact, I expected more code checking for __DJGPP__.

For __WATCOMC__, it's unclear for me yet if the code is specific to MS-DOS or to the Watcom compiler. For example, timemodule.c includes <i86.h> to get the delay() function. On Linux, I would prefer to use select(), on Windows, I would prefer to use the Windows code. We should check if OpenWatcom provides select() on Linux. I'm trying to build OpenWatcom for this ;)

I created the issue #22592 to propose to drop the support of the Borland C compiler.

I hope that "MS_WINDOWS" is also defined when Python is compiled by the Borland C compiler or OpenWatcom. IMO if it's not the case, it's a bug and it should be fixed.

It means that my patch drop_msdos_support.patch only drops support for the DJGPP compiler in fact. DJGPP is (was) the MS-DOS port of GCC.

So, what do you think of drop_msdos_support.patch? Is it ok for Python 3.5?

Yes!

New changeset a1605d2508af by Victor Stinner in branch 'default':
Issue #22591: Drop support of MS-DOS
https://hg.python.org/cpython/rev/a1605d2508af