Re: providing a libidn compatibility API

[Nikos Mavrogiannopoulos]

On Fri, Feb 10, 2017 at 5:14 PM, Tim Rühsen <address@hidden> wrote:
>> > The advantage would be to not hard-code flags within idn2_to_ascii_*().
>> > That seems to be more flexible. If you agree, I'll commit the docs from
>> > your patch and amend the arg names of idn2_to_ascii_4i.
>>
>> Replying to list, that this indeed makes sense. One thing I identified
>> is that libidn2 no longer compiles in mingw environment, or in other
>> environments where libunistring is missing (i.e., the internal
>> unistring is not sufficient).
>> Adding the missing items in bootstrap.conf (were already there in a
>> fixme block), allows compilation if one performs it as make -C lib &&
>> make
>> For some reason lib wouldn't compile, even though the order in Makefile.am
>> is: SUBDIRS = lib . src doc examples tests po
>>
>> Example:
>>
>> $ make
>> cat ./idna-tables-6.3.0.txt | ./gen-tables-from-iana.pl > data.c.new
>> mv data.c.new data.c
>>   CC       gentr46map.o
>> gentr46map.c:37:20: fatal error: unistr.h: No such file or directory
>>  #include <unistr.h>
>>
>> while
>> $ make -C lib && make
>>
>> works.
>
> We already had a commit that worked without linking libunistring - that adds
> of course extra code / binary to libidn2.so/.a.
>
> It turned out that if libunistring is linked, this extra code is still added
> to libidn2, just blowing up the size without any advantage. Simon and I didn't
> have (fast) solution to that, so we commented the extra gnulib modules in
> bootstrap.conf.

Interesting, that would affect gnutls as well since it is also using
libunistring that way.

> So right now, libunistring is a prerequisite for building libidn2. And someone
> building libidn2 should have no problem building & installing libunistring -
> it basically just is gnulib modules compiled into a library.

Would it make sense to make that explicit, and have configure fail if
libunistring is not available?


regards,
Nikos