Re: UTF8 vs. locale in error messages (bug 5715)

[Guy Harris <guy () alum mit edu>]

On Jun 28, 2011, at 6:10 AM, Stig Bjørlykke wrote:

> On Tue, Jun 28, 2011 at 2:58 AM, Guy Harris <guy () alum mit edu> wrote:
> >        1) UN*Xes where LANG etc. aren't set to a locale with UTF-8 as the encoding (are you seeing the issue with 
> > Norwegian characters on your system?  If so, what's the setting of LANG?);
>
> I only had issues with Norwegian characters in file names reported via
> simple_dialog(), and my LANG is empty.

OK, what OS are you using?  If it's a UN*X, try compiling and running the attached C program; does it print your name 
correctly on your terminal/terminal emulator (it writes it out in UTF-8), and does the file it creates (your name is 
its name - yeah, complete with a space between "Stig" and "Bjørlykke", and with no ".txt" at the end) have a name that 
shows up correctly if you do "ls"?  If it's Windows, then you're probably just seeing bug 5715.

> Another problem is that we still have issues regarding UTF-8 strings
> in packets.  We should really fix that...

We have an issue regarding strings in packets in general.  Strings might be in a number of encodings, including ASCII 
(meaning that any byte with the 8th bit set is something that shouldn't be there), other national variants of ISO 646, 
UTF-8, UTF-16, UCS-2 (meaning "only the Basic Multilingual plane, with no surrogate pairs"), ISO 8859/x for various 
values of x, various ISO 2022-based encodings (e.g., the EUC encodings), various national standards, various DOS and 
Windows code pages, various Mac OS encodings, EBCDIC, whatever encodings are used for SMS, etc., etc., etc, etc.:

        http://en.wikipedia.org/wiki/Template:Character_encoding

I don't know whether all of the encodings in question can be mapped to Unicode without information loss.  An arbitrary 
string of octets definitely can't be mapped to UTF-8 without information loss; consider a putatively UTF-8-encoded 
string that contains an octet sequence that's not valid in UTF-8.

Perhaps, in the Wireshark dissection engine, we should initially store string values as a pair {encoding, counted octet 
string} (counted so that octets with the value 0 don't cause problems), and:

        when putting them into a textual representation of the protocol tree or into columns or something else to be 
shown to humans, map them to UTF-8, with anything that can't be mapped to UTF-8 - including, if the encoding is 
putatively UTF-8, octet sequences that aren't valid UTF-8 sequences - shown as the Unicode replacement character U+FFFD;

        when comparing them in a display filter, attempt to map them to UTF-8 (and save the result), and:

                if the mapping fails, treat *all* comparisons except for inequality as failing, and treat comparisons 
for inequality as succeeding;

                if the mapping succeeds, compare the two strings;

        when making them available to software inside *Shark (C/C++ code, Lua code, Python code, etc.), attempt to 
convert them to whatever the appropriate representation is (presumably UTF-8), and have the routines to fetch those 
values support returning a "conversion failed" indication (or perhaps offer both a "convert for display to humans" 
version that uses U+FFFD for failure and a "convert for processing" version that returns "can't do it" for failure).

Here's the program I mentioned above:

Attachment: norsk.c
Description:

___________________________________________________________________________
Sent via:    Wireshark-dev mailing list <wireshark-dev () wireshark org>
Archives:    http://www.wireshark.org/lists/wireshark-dev
Unsubscribe: https://wireshark.org/mailman/options/wireshark-dev
             mailto:wireshark-dev-request () wireshark org?subject=unsubscribe