Re: GNUmail on NetBSD

[Richard Frith-Macdonald]

On 12 Apr 2012, at 20:16, Fred Kiefer wrote:

> On 12.04.2012 11:54, Riccardo Mottola wrote:
>> On 04/07/12 11:44, Wolfgang Lux wrote:
>>> Fred Kiefer wrote:
>>> 
>>>> 
>>>> For NSASCIIStringEncoding the code in GSFromUnicode() should never try
>>>> to use iconv. We have the conversion for this format hard coded.
>>>> Could you please add a breakpoint in this function and step through
>>>> it? I really would like to understand what goes on here.
>>> The problem, as explained by the reference attached to Stefan's email,
>>> is that NetBSD's iconv does not support the //TRANSLIT option that
>>> GNUstep uses to implement lossy transformations (as you also can see
>>> from the backtrace provided by Riccardo). I have no idea how we could
>>> fix that in GNUstep, but on the other hand there is an easy solution
>>> for Riccardo: Just install the gnu libiconv package on NetBSD and make
>>> sure it gets used instead of the default one.
>>> 
>> I tried to install the gnu libiconv and reconfigure base and now it
>> works without warnings.
>> Is this the only choice however?
> 
> No, I don't think so. But we should wait for Richard to decide whether the 
> current code is a bug or not. We have conversion code for 
> NSASCIIStringEncoding in place and should try to use that, when no iconv 
> conversion is found.
> The other chance you have is to modify the code in 
> -[NSString(PantomimeStringExtensions) stringFromModifiedUTF7]. No idea why 
> this is converting the hard coded strings in such a complicated way.

Looking at the email trail, I *think* the problem here was due to an error in 
checking for iconv's ability to do a lossy conversion (which should now be 
fixed in svn trunk).

That being said, the best solution here is to install libiconv rather the use 
the BSD iconv ... since using BSD iconv means you can't do a decent lossy 
conversion (the fallback mechanism in base should be to output a '?' marker in 
place of any character which can't be converted).

I've modified configure.ac so we should check for  the presence of an iconv 
library with lossy conversion support when base is built, and have the option 
to enable/disable it there.