Re: mod_perl, mod_gzip and connection->fileno

[prev] [thread] [next] [Date index for 2005/02/19]

Stas Bekman schrieb:

> Marcel Greter wrote:
>
>> Stas Bekman schrieb:
>>
>>>>    ... my $fileno = $connection->fileno(0);
>>>>    vec(my $rin = '', $fileno, 1) = 1;
>>>>    select(my $rout = $rin, undef, undef, 0);
>>>
>>>
>>>
>>>
>>> Is this better that IO::Select in [2]?
>>>
>> It should do the same thing. IMHO IO::Select is just a frontend for 
>> the core select function, so you don't need to fiddle with vec 
>> yourself. One can save some memory (504 KB here) if you use the core 
>> select function directly and don't load IO::Select, which is always a 
>> good thing.
>
>
> How about we keep the original code using IO::Select and provide an 
> example how to do it more efficiently? The IO::Select example is much 
> more userfriendly :)
>
Maybe something like this? Don't know how much you want to annotate ??

=head1 Detecting a Client Abort

  # IsClientConnected? Might already be disconnected for busy
  # site, if a user hits stop/reload

  my $conn = $r->connection;
  my $is_connected = $conn->aborted ? 0 : 1;
  if($is_connected) {
      if (defined(my $fileno = $conn->fileno(0))) {
          my $s = IO::Select->new($fileno);
          $is_connected = $s->can_read(0) ? 0 : 1;
      } else { $is_connected = 0 }
  }

You may also implement this without using the IO::Select module

  my $is_connected = 1;
  my $conn = $r->connection;
  # check Apache::Connection abort status
  if($conn->aborted) { $is_connected = 0 }
  else {
      if (defined(my $fileno = $conn->fileno(0))) {
          vec(my $rbit = '', $fileno, 1) = 1; # enable fd in vector table
          select($rbit, undef, undef, 0); # select for readable handles
          $is_connected = vec($rbit, $fileno, 1) ? 0 : 1; # check fd in
vector table
      } else { $is_connected = 0 }
  }




>>> Also take a look at: 
>>> http://perl.apache.org/docs/1.0/guide/debug.html#Handling_the__User_pressed_Stop_button__case 
>>>
>>> It's directly related and should probably be merged into one item. 
>>> or at least xref each other.
>>>
>> I'm checking the return value of each print, so this should take care 
>> of this case.
>
>
> Wow! But most people don't. So in mp2 you no longer need to do that, 
> as print() as well as all other IO methods croak on failure.
>
I'm doing a bigger framework, so there is actually just one place to
check the print function :-)

> So could you please adjust whatever needs to be adjusted in that 
> section, and post here a patch to the POD source (or the section 
> itself), while adding the discussion why one should use 0 and not 1, 
> based on this thread?
>
Please see the addition above. I'll post a patch later, ss it might will
change. By 0/1, do you mean the fileno call?

Marcel