Eli Zaretskii
eliz****@gnu*****
Thu Aug 2 01:52:00 JST 2018
> From: Keith Marshall <keith****@users*****> > Date: Wed, 1 Aug 2018 17:32:16 +0100 > > The problem is that Microsoft changed the return type, a few years ago, > from "int", (which is what the rest of the *nix world still uses), to > "intptr_t". To be consistent with spawn* functions, which can return a handle to the process they start. > The simplest solution would be to either[**]: > > - remove the offending prototypes from the OLDNAMES section of our > mingwrt <process.h>, or ... > > - leave the OLDNAMES prototypes in place, but revert the return > type to "int", (as g++ wants it, and POSIX says it should be). > > [*] Actually, I'd advise against using these Microsoft functions anyway; > if they successfully create a child process, then the calling process > terminates, without providing any mechanism for its parent to wait for, > or even identify the child. exec* functions make little sense without fork, yes. Moreover, I needed to replace exec* with the corresponding spawn*, because the former caused trouble with keyboard input, see the Gawk sources. > [**] There's actually a third option: ignore the testsuite failure! The > issue isn't even apparent, unless compiling with -Wsystem-headers, (as > the testsuite does), and even then it only raises warnings, (unless > -Werror is also enabled). If a C++ program invoking one of these functions compiles without errors, then maybe the third option is the best. I assumed, perhaps incorrectly, that the test suite failure means C++ programs using these functions (or even including <process.h>?) will fail to compile or link.
Fork