Re: [PATCH] gdbsupport: Remove some unnecessary ifdef HAVE_POLL judgments

[Youling Tang <tangyouling@loongson.cn>]

Hi, Pedro

On 05/13/2022 05:28 PM, Pedro Alves wrote:
> On 2022-05-13 08:49, Youling Tang wrote:
>> By searching event-loop.cc we found that there are only the following 2
>> places to assign values to use_poll,
>> $ grep -nr use_poll gdbsupport/event-loop.cc
>>    88:static unsigned char use_poll = USE_POLL;
>>    263:	use_poll = 0;
>>
>> This USE_POLL is defined as follows,
>>   #ifdef HAVE_POLL
>>   #define USE_POLL 1
>>   #else
>>   #define USE_POLL 0
>>   #endif
>>
>> So use_poll may be 1 only if HAVE_POLL is defined. Removed "ifdef
>> HAVE_POLL" judgment in use_poll control block.
>>
>> Signed-off-by: Youling Tang <tangyouling@loongson.cn>
> Maybe I'm missing something, but ISTM this can't possible work on hosts that don'
>   have poll at all.  Like e.g., mingw:
>
>   $ grep POLL *
>   config.h:/* #undef HAVE_POLL */
>   config.h:/* #undef HAVE_POLL_H */
>   config.h:/* #undef HAVE_SYS_POLL_H */
>
> Surely they'll fail to compile after the patch?
You are right my oversight.

But we can remove these internal_error handling, similar to the 
following modification:

  add_file_handler (int fd, handler_func *proc, gdb_client_data client_data,
                   std::string &&name, bool is_ui)
  {
-#ifdef HAVE_POLL
-  struct pollfd fds;
-#endif
-
    if (use_poll)
      {
  #ifdef HAVE_POLL
+      struct pollfd fds;
+
        /* Check to see if poll () is usable.  If not, we'll switch to
          use select.  This can happen on systems like
          m68k-motorola-sys, `poll' cannot be used to wait for `stdin'.
@@ -264,19 +263,13 @@ add_file_handler (int fd, handler_func *proc, 
gdb_client_data client_data,
        fds.events = POLLIN;
        if (poll (&fds, 1, 0) == 1 && (fds.revents & POLLNVAL))
         use_poll = 0;
-#else
-      internal_error (__FILE__, __LINE__,
-                     _("use_poll without HAVE_POLL"));
-#endif /* HAVE_POLL */
+#endif

Thanks,
Youling.