Re: [Bug-tar] Bad filenames cause tar to produce bad pax archives

[Tim Kientzle]

SUSv3-2008 has a standard way of addressing this issue.  I
don't have the current reference handy, but it was first
raised on the Austin group mailing list a few years ago.
Search for 'hdrcharset' in:

http://www.opengroup.org/austin/aardvark/latest/xcubug2.txt

With this, you can store the filename as a binary
string instead of utf8 if the utf8 conversion fails.
This is supported by bsdtar/libarchive, by Heirloom tar,
and (I believe) by star.

Implementing this correctly is a little tricky, since
the hdrcharset option applies to all headers, not just
the pathname header.  So you basically have to try to
convert everything to UTF8 and if anything fails,
use hdrcharset=binary.

Tim

Lars Gustäbel wrote:
> Hello!
>
> I noticed that tar with the --format=pax option will produce invalid archives
> when it encounters filenames that it cannot convert from the user's charset to
> utf8.  In src/xheader.c the function code_string simply copies the input string
> to the output string in event of an error during utf8_convert. The problem
> seems to be known - there is a FIXME tag in the comment. However, I'd favor the
> solution from the comment (report an error) instead of silently producing a
> corrupt archive.
>
> Regards,