Re: Failure with libidn2 on OpenBSD.

[Simon Josefsson]

Mats Erik Andersson <mats.andersson@gisladisker.se> writes:

> Måndag den 25:e januari 2021, klockan 08:47, skrev simon--- via Bug
> reports for the GNU Internet utilities detta:
>> Indeed, on a modern system with glibc, everything works via getaddrinfo
>> without need to link with libidn/libidn2.  The code in inetutils is
>> buggy today (it passes on non-domains to IDNA functions), and if anyone
>> wants IDN-enabled applications on non-GNU systems, the best place to fix
>> this would be in gnulib's getaddrinfo.  Then it would work automatically
>> in inetutils too.  Fixing the bugs and supporting the build complexity
>> just to cater for non-GNU platforms is not worth the maintainence cost
>> for us, IMHO.
>> 
>> Given the discussion above, I propose to remove all uses of
>> libidn/libidn2 from inetutils, and place a note in gnulib's getaddrinfo
>> that it is missing IDN-functionality to mimic glibc's IDN behaviour.  I
>> might even get around to adding IDN-functionality to gnulib, but no
>> promises.
>> 
>> Any objections?  I'll test the rest of the apps in InetUtils and remove
>> this code if there aren't any objections.
>
> As can be expected, I do object to the removal of our IDN-code, as long
> as no other provider of the corresponding functionality is present.
> My Sisyphos task of keeping Inetutils useful beyond glibc-systems is
> truely a never ending mission, expected to last until my demise or to
> my retirement, whichever comes first.
>
> Why is the action simply to disable IDN-linkage for glibc-systems so
> repulsive that you want to cripple all other users?

Hi Mats!  Please re-read the thread to see the original problem -- the
current InetUtils IDN-related code is problematic, in two ways:

  1) Linking directly to libidn/libidn2 by simple applications is not
  recommended any more and few applications ended up doing so.  Simple
  applications should use getaddrinfo() which is IDN-enabled on
  glibc-systems (via libidn2).  By having libidn/libidn2-specific code
  in InetUtils we will trigger bugs that we wouldn't otherwise
  have... such as:

  2) The current IDN-related code in InetUtils is buggy: it shouldn't
  pass strings such as ::1 that aren't domain names to IDNA functions.

Solving 2) may sound like a step forward, but it will hide all other
problems stemming from 1) and it adds even more complexity, which will
in all likelyhood introduce a new set of bugs.

The place to add portability improvements that cater to non-GNU systems
should be in gnulib, in my opinion.

I'll avoid committing anything here to not upset the status quo and
focus on getting an alpha out.  I think getting the release out has
priority over any particular improvement or bugfix right now, including
my own pet issues.

/Simon

signature.asc
Description: PGP signature