Text 2675, 397 rader
Skriven 2005-02-22 07:57:50 av Ellen K. (1:379/45)
Kommentar till text 2667 av Rich (1:379/45)
Ärende: Re: ESB / XML / Unicode vs 8-bit characters ?
=====================================================
From: Ellen K. <72322.1016@compuserve.com>
On Mon, 21 Feb 2005 20:52:26 -0800, "Rich" <@> wrote in message
<421aba24@w3.nls.net>:
> I don't understand why oracle using UTF-8 wouldn't work and oracle using
UTF-16 would. That sounds like an oracle problem.
That would be no surprise to me, I am not an Oracle fan. It drove me nuts when
I had to use it at Kaiser. That parallel query processor they are so proud of
crashed if you looked at it cross-eyed. My SQL Server
OLTP databases each have over 600 users now, one on a single-processor box, the
other on a dual-processor one, zero performance problems and almost no tuning
ever required. If they were on Oracle we would need more CPUs and a full-time
DBA just to babysit it. If I had a monster
multi-terabyte database that was too big for SQL Server I would go with DB2 or
possibly Terabyte.
>
> There are two XML issues I see.
>
> One, and the only bug in what we have discussed, is that the encoding
declared in the ?xml PI must match the actual encoding of the file. You can
use any encoding that is supported as long as your XML document is correctly
encoded in whatever is declared.
>
> The second issue is what encodings does your software support. I believe
all XML implementations to claim compliance must support UTF-8 and UTF-16. Most
should support more. I haven't tried Microsoft SQL Server but I would expect
it to handle any of the encodings with support installed on that host machine.
Windows supports many encodings. I just checked my machine and see more than
100 supported encodings.
Where do I look to see what encodings are supported?
>
> My suggestion is to stick with UTF-8 for the stuff you generate if you can.
For what you have to consume from other software simply has to work. If not
you can go back to the folks responsible for creating it and find a way to make
it valid.
For stuff I will have to generate, the FOR XML clause doesn't have a parameter
for the encoding and I'm not finding anything in BooksOnline about how to set
the encoding, or about a default encoding for output generated using FOR XML.
??? Clearly this is something about which I
need to learn much more, until now I have only been consuming XML, not
outputting it. Somewhere I saw that SS2005 can automatically create
SOAP objects, maybe that is something to investigate.
Based on Sunday's experiments and the fact that Oracle's default is UTF-16,
it's looking to me like that would be the most foolproof, but OTOH being that
every character will take 2 bytes it could slow things down. I had originally
planned to communicate directly with Oracle but
I will likely have to use Sonic for situations requiring the equivalent of a
distributed transaction where related data need to be written to SQL Server,
Oracle and one or more of the purchased apps. Our Oracle guy's strength is the
Oracle Financials apps as opposed to stuff like interoperability. Sonic and
the purchased apps are being installed and configured by an outside consulting
outfit -- I don't wanna get involved with trying to make either the outside
consultants or the vendors fix stuff that doesn't work, I want it to work from
day one and continue to work going forward. I am going to try to talk with
them about this but based on what I've seen so far I can't say my expectations
are very high.
>
>Rich
>
> "Ellen K." <72322.1016@compuserve.com> wrote in message
news:qt9l115cio5kc8ijjumaubpvb6n9ig50ir@4ax.com...
> Argggh, I replied from work using Dale's mirror, it doesn't show up
> here, waitaminnit...
>
> OK, here we go:
>
> First of all I have to say you have once again earned my eternal
> gratitude, I am so glad you are here. :)
>
> I still have to make a decision regarding the standard I want to
> request. Are you saying that if for example I get XML generated by say
> SalesForce with UTF-8 specified and it includes these characters, I
> would not have a problem? Stuff I write myself I'm not worried about
> because I have the opportunity to tweak it, what I'm worried about is
> what the ESB will try to feed my databases. Since even the hand-typed
> XML was accepted when UTF-16 was specified, I'm kind of leaning toward
> that, especially since the Oracle guy told me that the TJ accounting
> manager used to complain that the Spanish characters weren't coming
> out on reports when we were on Oracle 8.x which used UTF-8 but as soon
> as we moved to 9.x which uses UTF-16 there were no more problems. What
> (if any) do you see as a potential downside to UTF-16?
>
> Thanks again. :)
>
>
>
> On Mon, 21 Feb 2005 00:24:53 -0800, "Rich" <@> wrote in message
> <42199a9e@w3.nls.net>:
>
> > The problem is not that UTF-8 won't work for what you call Spanish
characters. UTF-8 can encode anything that you consider a character. The
problem is that when these characters are present they are not being encoded in
UTF-8. There are two straight forward solutions I see. One is to encode the
XML correctly in UTF-8. The other is to encode the XML in the Windows ANSI
encoding I suspect you are using and to tag it correctly. On a Windows U.S.
English system this would be "Windows-1252". I would expect SQL Server to
support this. Other applications may or may not. You could also use UTF-16 as
long as you generate your XML in UTF-16. I don't know if this is simple for
your application or not. If you are using VB 7.0 it should be.
> >
> >Rich
> >
> > "Ellen K." <72322.1016@compuserve.com> wrote in message
news:of2j11lqf4r4h1dsnvok8i1dv7c9e3rlb2@4ax.com...
> > Well. You just helped me learn a lot. For one thing, I guess I
> > thought that because the Spanish characters can be expressed as 8 bits
> > they were ASCII.
> >
> > I just made a little test stored procedure taking an XML document as a
> > parameter, created the document manually in Notepad with UTF-8 specified
> > in the header and tried including some of the Spanish characters... and
> > it failed. SQL Server could not execute sp_xml_preparedocument because
> > "an invalid character was found in text content". Just to make sure
> > that was the problem I substituted non-Spanish characters for the
> > Spanish ones and it executed fine. However, I can manually type the
> > text with the Spanish characters into the varchar field if I open the
> > table in EnterpriseManager and SQL Server is perfectly happy. OTOH, if
> > I specify UTF-16, I get "Switch from current encoding to specified
> > encoding not supported." Next thing I tried was cloning the sproc to
> > write to the table with the nvarchar column, still no joy, same error
> > message... but on changing the datatype of the input parameter from text
> > to ntext it worked fine. BUT here's the surprise (OK, to me it was a
> > surprise): If I again clone the sproc to point to the table with the
> > varchar column, but leave the input parameter as ntext and specify
> > UTF-16 in the document header, it works. In other words, a varchar (and
> > presumably a char) column can successfully accept unicode data even
> > though char and varchar are explicitly defined as non-unicode datatypes!
> >
> > Now I have to understand whether I have a problem at work. I never
> > experienced this problem in real life because none of the data we
> > currently send using XML includes any of the Spanish characters. Is
> > the problem only going to occur if the XML document is constructed using
> > the concatenated-string method? Or would it happen any time an XML
> > document specified as UTF-8 included Spanish characters? (The data
> > sent by SalesForce to the ODS is likely to include Spanish characters,
> > but it probably creates the XML some other way.) Do I need to tell the
> > consulting outfit to specify all XML as UTF-16?
> >
> > For the ETL from the ODS to the data warehouse I am not planning to use
> > Sonic, but rather probably to link the databases and use a bunch of
> > stored procedures controlled by some VB code, IOW I will not need XML
> > because all the extract sprocs will look like INSERT INTO....SELECT
> > FROM.
> >
> > I don't yet understand why UTF-8 can't work for the Spanish
> > characters... (unless it only doesn't work when the characters are
> > manually typed into the document). If I correctly understand the
> > document to which you referred me, an 8-bit character can't fit in one
> > UTF-8 byte because the first bit is reserved for indicating which is the
> > first byte of a UTF-8 multi-byte character. (This was your point about
> > not greater than 0x7F.) But why wouldn't it just make two bytes out of
> > the Spanish characters then? The documentation says UTF-8 uses
> > multiple bytes for the characters that it can't fit into one byte.
> >
> > ???
> >
> >
> >
> > On Sun, 20 Feb 2005 13:10:40 -0800, "Rich" <@> wrote in message
> > <4218fc91@w3.nls.net>:
> >
> > > The Spanish accented characters are not part of ASCII. They are part
of Windows calls ANSI of which ASCII is the subset (0x00 to 0x7F). Any
character in the 0x80 to 0xFF range is not compatible between ANSI and UTF-8.
> > >
> > >Rich
> > >
> > > "Ellen K." <72322.1016@compuserve.com> wrote in message
news:7ouh119ivmuk26icg3mqqqk2ss1lfm5c10@4ax.com...
> > > Should not have any non-ASCII characters, as previously noted all the
> > > special Spanish characters are available in the ASCII character set.
> > > And since the company is built on our understanding of the Hispanic
> > > market, I don't see any use of, say, pictograph-based languages in the
> > > foreseeable future. If 10 years down the road something like that
> > > happens, well, by then we will no longer need compatibility with the
> > > current legacy system because it will long since have been replaced.
> > >
> > > On Sun, 20 Feb 2005 12:52:25 -0800, "Rich" <@> wrote in message
> > > <4218f849$1@w3.nls.net>:
> > >
> > > > From what you describe below, if the values you emit to XML have
non-ASCII characters I would expect you to have a problem.
> > > >
> > > >Rich
> > > >
> > > > "Ellen K." <72322.1016@compuserve.com> wrote in message
news:eanh11h4vv6b9v21fiaounii3f5dunjl3g@4ax.com...
> > > > On Sat, 19 Feb 2005 23:32:37 -0800, "Rich" <@> wrote in message
> > > > <42183ccd@w3.nls.net>:
> > > >
> > > > > The UTF in UTF-8/16/32 stands for Unicode Transformation
Format. You can find these defined in section 2.5 of
http://www.unicode.org/versions/Unicode4.0.0/ch02.pdf.
> > > >
> > > > THANK YOU SO MUCH!!! :)
> > > > >
> > > > > It's not clear to me how you are creating the XML from the
templates. If ANSI data is emitted into an XML document declared as UTF-8 then
you would have problems only for non-ASCII characters. UTF-8 and Windows-1252
are identical for 0x00 to 0x7F which is ASCII in both.
> > > >
> > > > I don't have a copy of a template here at home, but I have them
create
> > > > it by string concatenation because that seems to be the only way to
be
> > > > able to have CDATA attributes, which I have to have because in the
> > > > legacy data numeric-appearing identifiers are actually 10-character
> > > > strings with leading spaces, and if these are not specified as
CDATA
> > > > the spaces go lost even with "xml:space="preserve"" included in the
> > > > header. Here is a code snippet from one of my apps that creates an
XML
> > > > document which is passed as a parameter to a SQL Server stored
> > > > procedure:
> > > >
> > > > > strXM = "<?xml version =" & Chr(34) & "1.0" & Chr(34) & "
encoding=" & Chr(34) & "UTF-8" & Chr(34) & "?>" & vbCrLf _
> > > > > & "<ROOT xml:space=" & Chr(34) & "preserve" & Chr(34) &
">" & vbCrLf
> > > > >
> > > > > Do While Not .EOF
> > > > > strXM = strXM & "<M><A>" & !Ofc & "</A><B><![CDATA[" &
!Contract & "]]></B><C>" & !TCode & "</C><D>" & !Date & "</D>" _
> > > > > & "<E><![CDATA[" & !TransNo & "]]></E></M>" &
vbCrLf
> > > > > .MoveNext
> > > > > Loop
> > > > >
> > > > > strXM = strXM & "</ROOT>"
> > > >
> > > > (The vbCrLf's are there so if there is a problem the document can
be
> > > > printed to a text file and be easier for humans to read -- SQL
Server
> > > > ignores them. The single-character aliases for entity and
attribute
> > > > names are for performance -- for most of the stuff we use these for
it
> > > > doesn't really matter because we are only sending a few rows, but
the
> > > > first time I did it it was for something that was sending about
5000
> > > > rows and there it made a huge difference, so I stuck with it. We
> > > > comment both the front-end code and the stored procedure with the
> > > > mappings of these aliases.)
> > > >
> > > > > I do not know how SQL Server maps from char to nchar,
specifically what conversion is performed. Also, in some (maybe all released)
versions of SQL Server nchar and nvarchar are encoded in UCS-2. UCS-2 is a
16-bit encoding like UTF-16. It dates back to when Unicode was defined as
having 2**16 characters instead of the 2**20+ that it has now. You can not
express characters >= U+10000 in UCS-2 not that you care about these.
> > > >
> > > > Thankfully, no. :)
> > > > >
> > > > > I don't know if whether those systems you describe being
written in java make a difference. They can do what they want. The native
java string is Unicode though I don't remember if it is UCS-2 or UTF-16. My
guess is that it was once the former and is now the latter. One of the
documents on this on sun's site suggests that java used UCS-2 until the
recently released 1.5 which is the first to use UTF-16.
> > > >
> > > > The Java native string being unicode is exactly what made me start
> > > > worrying -- when I was learning Java a couple of years ago (because
I
> > > > wanted to port an app to it so as to be able to run it right on the
Unix
> > > > box where the Oracle database was) I was horrified the first time I
> > > > tried reading back what I had written to a text file when I saw
spaces
> > > > between all the characters.
> > > > >
> > > > >Rich
> > > > >
> > > > > "Ellen K." <72322.1016@compuserve.com> wrote in message
news:aqag115606i9g8bmh3lst66une1f1sotth@4ax.com...
> > > > > UTF-8 is unicode?!? Sheesh, all this time I thought it meant
8-bit.
> > > > > In fact I could swear I read that somewhere.
> > > > >
> > > > > My question was coming from the database perspective, where I
always use
> > > > > char and varchar, as opposed to nchar and nvarchar. I give the
> > > > > front-end guys little templates for creating the XML documents
for all
> > > > > my SQL Server stored procedures that take XML input, and I
always
> > > > > specify UTF-8 in the header... and my char and varchar columns
always
> > > > > end up normal, so since you're now telling me UTF-8 is really
unicode, I
> > > > > guess that would answer my question for XML data I would be
getting from
> > > > > the apps...? Or would the answer be different if the incoming
XML is
> > > > > some other encoding?
> > > > >
> > > > > To simulate getting nvarchar data from somewhere, I just tried
creating
> > > > > two dummy tables, one with an nvarchar column and the other with
a
> > > > > varchar column, typed stuff into the nvarchar one, then inserted
to the
> > > > > varchar one select from the nvarchar one and it looks normal.
> > > > >
> > > > > If all this means I was worrying about nothing, excellent!
OTOH, is
> > > > > there something I should be worrying about that I didn't ask?
> > > > >
> > > > > The only pieces whose names I know so far are Sonic and
SalesForce, both
> > > > > of which are written in Java, if that makes any difference. I
know
> > > > > there is at least one other external piece but I think that is
the next
> > > > > phase.
> > > > >
> > > > > On Sat, 19 Feb 2005 21:37:15 -0800, "Rich" <@> wrote in message
> > > > > <421821c1$1@w3.nls.net>:
> > > > >
> > > > > > You need to be more specific than "8-bit characters". There
are many 8-bit character encodings. If you are using Windows to generate your
data you most likely are using Windows-1252 which is the default 8-bit
character set for U.S. English in Windows. Windows supports many 8-bit
encodings so you could be using something else too.
> > > > > >
> > > > > > Unicode is a character set not an encoding. There are
multiple encodings the main ones being UTF-8, UTF-16, and UTF-32. You can use
any of these for XML as well as non-Unicode encodings. For interoperability
you should use Unicode preferably UTF-8.
> > > > > >
> > > > > > What comes out when the XML is parsed depends on the XML
parser. XML is logically expressed in Unicode. The Windows XML parsers
provide a Unicode interface. Other parsers could do differently.
> > > > > >
> > > > > >Rich
> > > > > >
> > > > > >
> > > > > > "Ellen K." <72322.1016@compuserve.com> wrote in message
news:4o2g11pu048kafbdilg46u77vs5ls0be55@4ax.com...
> > > > > > Our new enterprise system is going to be built around an
Enterprise
> > > > > > Service Bus. I don't have the full specs yet but as I
understand it the
> > > > > > main apps (starting with SalesForce) are going to be out on
the internet
> > > > > > and the Sonic ESB will be the messaging piece. There will
be an
> > > > > > Operational Data Store in house that will get updated every
night on a
> > > > > > batch basis from the main apps.
> > > > > >
> > > > > > My data warehouse will continue to be the data warehouse and
will remain
> > > > > > in house. The dimensions will stay the same but I might have
to create
> > > > > > separate measures for the data from the new apps and then
create views
> > > > > > to keep everything transparent to the users.
> > > > > >
> > > > > > I'm thinking if we're going to have an ODS in house already,
I may as
> > > > > > well do the ETL from there. But I'm worrying that the new
data will
> > > > > > probably be unicode (because Java defaults to that and
SalesForce is
> > > > > > written in Java). Right now I am storing everything (except
our blobs
> > > > > > of course) in 8-bit characters.
> > > > > >
> > > > > > Anyone here who's up on this stuff, can the XML that goes
back and forth
> > > > > > convert between unicode and 8-bit characters, or am I gonna
have to
> > > > > > redefine all my data? For example, if unicode data is put
into an XML
> > > > > > document that specifies UTF-8, what comes out when the
document is
> > > > > > parsed? How about vice versa? If this is too simplistic to
work, what
> > > > > > is needed?
> > > > > >
> > > > > > (We actually have no substantive need for unicode -- we are
bilingual
> > > > > > Spanish but all the special Spanish characters exist in the
ascii
> > > > > > character set.)
--- BBBS/NT v4.01 Flag-5
* Origin: Barktopia BBS Site http://HarborWebs.com:8081 (1:379/45)
|