CreateProcess No such file or directory

[Jonathan Wakely]

On Thu, 22 Sept 2022 at 07:45, LIU Hao via Gcc-help
<gcc-help@gcc.gnu.org> wrote:
>
> 在 2022/9/21 00:02, mizo 91 via Gcc-help 写道:
> > Hello,
> >
> > I'm having trouble compiling simple test program on windows 10 with long
> > list of includes provided via '@response_file' argument
> >
> >
>
> Greetings. mingw-w64 developer speaking.
>
> As far as I can see, there are at least two issues about your report:
>
>
> The first, obvious issue is that the error message is incorrect. The reason for that is, if we take
> a look at 'libiberty/pex-win32.c' we see the following:
>
>    853   /* Create the child process.  */
>    854   pid = win32_spawn (executable, (flags & PEX_SEARCH) != 0,
>    855                      argv, env, dwCreationFlags, &si, &pi);
>    856   if (pid == (pid_t) -1)
>    857     pid = spawn_script (executable, argv, env, dwCreationFlags,
>    858                         &si, &pi);
>    859   if (pid == (pid_t) -1)
>    860     {
>    861       *err = ENOENT;
>    862       *errmsg = "CreateProcess";
>    863     }
>
> We also notice this is the only place where `"CreateProcess"` appears as a sole part of an error
> message.
>
> The cause of this issue is apparent: libiberty tries `win32_spawn`, and if for whatever reason it
> fails, it makes another attempt with `spawn_script`, and if it fails again, `*err` is always set to
> `ENOENT` i.e. `No such file or directory`, no matter why.

It seems to me that pex-win32.c should use GetLastError() (and maybe
FormatMessage) to get a Windows error code, rather than assuming
ENOENT.

Using C++ that would look something like:

std::string errmsg = "CreateProcess:" +
std::system_category().message(GetLastError());

Or maybe win32_spawn should call GetLastError() and convert that to an
errno value and store it in errno:

std::error_condition ec =
std::system_category().default_error_condition(GetLastError());
if (ec.category() == std::generic_category())
  errno = ec.value();
else
  errno = ENOENT; // or EINVAL might be better?