Re: [PATCH] gdb: add a file event mask parameter to add_file_handler

[Patrick Monnerat <patrick@monnerat.net>]

On 8/23/22 21:09, Tom Tromey wrote:
>>>>>> "Patrick" == Patrick Monnerat via Gdb-patches <gdb-patches@sourceware.org> writes:
> Thanks for the patch.
>
> Patrick> +#if defined(POLLRDHUP)
>
> Space before the '('.
Sure: my bad !
> Patrick>  void
> Patrick>  add_file_handler (int fd, handler_func *proc, gdb_client_data client_data,
> Patrick> -		  std::string &&name, bool is_ui)
> Patrick> +		  std::string &&name, bool is_ui, int mask)
> Patrick>  {
> Patrick> +  if (fd < 0 || (mask & (GDB_READABLE | GDB_WRITABLE | GDB_EXCEPTION)) == 0)
> Patrick> +    return;
>
> Is there ever a situation where this can happen and it is not a bug in
> the caller?  I am wondering if this ought to be an assert instead.
Yes. In Insight, command input from stdin is disabled by setting 
main_ui->input_fd to -1. A later call to register_file_handler from 
async_enable_stdin causes add_file_handler to be called with fd = -1. 
Maybe I should add a comment for this.
>
> Patrick> +/* Tell add_file_handler what events we are interested in. */
> Patrick> +
> Patrick> +#define GDB_READABLE	(1<<1)
> Patrick> +#define GDB_WRITABLE	(1<<2)
> Patrick> +#define GDB_EXCEPTION	(1<<3)
>
> I tend to think it would be better to use enum_flags here.

I thought so too and did try to use enum_flags! Unfortunately the same 
int field (file_handler::mask) stores either these bits in case of 
select use, or a poll mask as defined by the system when poll is used.

We then have a type conflict: it can be resolved but is rather 
cumbersome (a simple cast would be awful). Your opinion?


Thanks for the review.

Patrick