This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: GCC-6.4 sys/select.h build failures with std=c++14

Am 26.09.2017 um 07:32 schrieb Marco Atzeri:

"The <sys/select.h> header shall define the fd_set type as a structure."

so if they are using it, they should have a proper include

The complete story is a bit more complicated than that, though.

The curl maintainers almost fixed this at their end, but then Cygwin reverted part of a change that had "suddenly" triggered their broken code to actually fail before they did. Before then, Cygwin's fd_set was actually in <sys/types.h>, which was then #included by <sys/select.h>. The change was to define them it <sys/select.h> where it belongs

But then the second change was made (commit ee97c4b22491b205fd3b7697e03c909e02b652d3), which reintroduced

# if    __BSD_VISIBLE
#include <sys/select.h>

in <sys/types.h>. This actually re-introduced a part of the original POSIX violation in Cygwin: #including <sys/types.h> will, again, drag in <sys/select.h> even though that was not asked for. Apparently this violation is sufficiently wide-spread that there's quite some code out there that relies on it. Curl is in this group.
All this seems to have left the Curl people under the impression that there was nothing wrong with their code, but that rather Cygwin was broken, and for a relatively short time only, too (2015-12-17 to 2016-03-18). The fragment in question is positively hideous:

#if defined(_AIX) || defined(__NOVELL_LIBC__) || defined(__NetBSD__) || \
  defined(__minix) || defined(__SYMBIAN32__) || defined(__INTEGRITY) || \
  defined(ANDROID) || defined(__ANDROID__) || defined(__OpenBSD__) || \
 (defined(__FreeBSD_version) && (__FreeBSD_version < 800000))
 #include <sys/select.h>

Now people complain about autoconf being clunky, but it's easily an order of magnitude better than _that_ pile of nonsense --- particularly given the fact that they actually use autoconf to build their code. They just don't do so in their public, installable header file.

Problem reports:
Unsubscribe info:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]