[josm-dev] IPv6 problems

[Dirk Stöcker]

On Sat, 2 Jan 2016, Simon Legner wrote:

> what about an expert (for now) option (such as prefer.ipv6=jvm) to
> disable JOSM's algorithm completely and solely rely on the JVM to
> figure out the correct way to connect to a server? According to [1] …
>> IPv6 in Java is transparent and automatic. Porting is not necessary; there is no need to recompile source files.

That text is a bunch of shit. Sorry :-)

It would be automatic when Java would follow the OS, but no, they needed 
to implement their own rules. And these rules say:
  * Always use IPv4 except for IPv6 only addresses or
  * Always use IPv6 except for IPv4 only addresses and
  * You need to choose before doing any network I/O.

Only thing automatic is that you don't need to use new functions and care 
for address format as you often need to do in C/C++.

So essentially we have three options
  * reimplement nearly the whole network stuff
  * don't use IPv6
  * use the testing approach and see what troubles users have.

Currently we use method 3 and adapted detection when necessary. The magic 
in JOSM tries to detect when IPv6 is usable and does fallback tp IPv4 when 
unsure. Current implementation is relatively reliable except for the 
mentioned corner cases. I'm willing to fix the detection when issues 
occur, but not for partially broken network providers. Users should issue 
complaints with their providers in these cases.

Ciao
-- 
http://www.dstoecker.eu/ (PGP key available)