Re: [Bug-wget] Revisiting AI_ADDRCONFIG

[Ángel González]

 Hrvoje Niksic wrote:
> "Ángel González" <address@hidden> writes:
>
>> However, there's another easier thing to do for wget. Ulrich Drepper
>> strongly argues there that getaddrinfo() callers should use the
>> AI_ADDRCONFIG flag. wget is purposefully not using that flag [2] and
>> so even if disabling ipv6 [3] it still performs the two queries (and
>> produces the waiting behavior) whereas adding AI_ADDRCONFIG avoid it.
> Note that you can disable IPv6 in Wget itself with --inet4-only or the
> equivalent .wgetrc command.  That will cause Wget to request AF_INET
> family from getaddrinfo and (presumably) avoid the spurious lookup that
> causes problems.
I could also build a patched wget with that flag set.
The issue is, I shouldn't need to configure the network individually
on each of the hundreds of installed packages.

> AI_ADDRCONFIG does the equivalent of --inet4-only and --inet6-only
> automatically, depending on which network interfaces are configured.
> When we tried using it some years ago, it caused a number of bugs
> because of the flag having been poorly implemented, both on Linux
> (including my primary development machine) and on other operating
> systems, such as Solaris and reportedly AIX.
I am requesting to reconsider that. Hopefully, the operating
systems will have improved on these years.

> Ulrich is pushing for uniformity in application space, which is a
> reasonable position and one I have sympathy for.  But Wget is not a
> Linux-only program, nor does it cater only to systems configured in a
> specially prescribed way. 
AI_ADDRCONFIG is a POSIX option. How does it restrict you
any more than getaddrinfo() ?.

> If we're certain that AI_ADDRCONFIG wouldn't
> hurt anyone, we can add it, but given Wget's flexibility in requesting
> and sorting addresses returned by DNS, I don't see a compelling reason
> to do it right now.
Today is as good as tomorrow. You can consider it as a "would
be nice" feature if you want to.
Sadly, I don't have access to any AIX machine, so can hardly
help testing to add it on whe "broken" machines, which is
the tricky part.