Hi,
On Mon, Jan 20, 2014 at 06:22:37PM +0100, Martin Kaiser wrote:
> if I have a tvbuff that starts with 0x86 and I call
>
> a = tvb_get_string_enc(tvb, 0, ENC_ASCII)
> proto_tree_add_string(..., a);
>
> I can trigger the DISSECTOR_ASSERT since a is not a valid unicode string.
>
> Comments in the code suggest that tvb_get_string() should replace
> chars>=0x80 with the unicode replacement char, which is two bytes long.
> This would look like
> [...]
>
> The resulting string would still contain len+1 chars but not necessarily
> len+1 bytes. Would that be a problem, i.e. is it ok to do sth like
>
> b = tvb_get_string(NULL, tvb, offset, len_b);
> copy_of_b = g_malloc(len_b+1);
> memcpy(copy_of_b, b, len_b+1);
If you just want to duplicate string you should definitely use g_strdup() ;-)
> If that should work, we'd need a separate function for get string &
> replace 8bit chars.
I think we don't need, tvb_get_string_enc(, ENC_ASCII) should return valid UTF-8 string,
and all callers assuming it's just 1:1 copy are buggy.
Maybe we should add: ENC_STRING_DONT_CONVERT, if people want just to
have NUL terminated string?
btw. I really wonder if current way of using a replacement character is good one.
Maybe we should escape it to some: \x86.
