Internal state not reset between calls

[bnoordhuis]

Internal state not reset between calls

[aklt]

This seems to be fixed by this patch: https://gist.github.com/896558

[aklt]

Hmm, I was a bit quick there, only some of the problem went away. The output is now:

+AO +AO +AO

but a final 'c' is missing.

[papandreou]

There's similar, yet subtly different breakage the other way around:

var converter = new (require('iconv')).Iconv('utf-7', 'utf-8'),
    utf7cedilla = new Buffer([0x2b, 0x41, 0x4f, 0x63]); // "+AOc" in ASCII
converter.convert(utf7cedilla).toString(); // 'ç'
converter.convert(utf7cedilla).toString(); // EINVAL, Incomplete character sequence.
converter.convert(utf7cedilla).toString(); // 'ç'
converter.convert(utf7cedilla).toString(); // EINVAL, Incomplete character sequence.
// etc.

[bnoordhuis]

I wrote a test case in plain C that exhibits the same behaviour so it seems you've uncovered a bug in libiconv.

I'm trying to check it against the bleeding edge but it's proving exceedingly difficult to build, one of the in-tree build tools dies with a buffer overrun if FORTIFY_SOURCE is enabled.

To be continued.

[aklt]

I guess one could argue that this is a bug with libiconv since the
last parameter (outbufsize) to iconv() is not updated correctly.

Still

echo -n 'ç'  | ./iconv_no_i18n -f 'UTF-8' -t 'UTF-7'

Does the right thing because outbufsize is not used to determine
the size of output in iconv.c. Instead the difference between the two
pointers is used to fwrite() the correct amount of bytes (iconv.c line 707).

In node-iconv the difference between these is not correct for this example. I think the reason may be that iconvctl is used to set some hooks in iconv.c,
specifically the update_line_column seems to be called....

Now I need to get some sleep :-)

[aklt]

Ups, accidentally closed the issue but it seems it cannot be reopened, hmm,...

[bnoordhuis]

No problem, I've reopened it.

[bnoordhuis]

My bad, it turns out that this is indeed a bug in node-iconv: the trailer isn't written correctly. I'll try to push out a fix tomorrow.

[bnoordhuis]

Fixed in commit 58764b2.

Can I have one or more 'works for me too', please? :-)

[papandreou]

WFM, too :)

[bnoordhuis]

Cool, thanks. I've uploaded 1.0.1 to npm just now.

[aklt]

Cool, thanks for the quick update :-)