Forum OpenACS Development: Re: Tcl Web Services Toolkit: TWiST

Collapse
Posted by Malte Sussdorff on
<wsdl:definitions xmlns:projectm="urn:tcl:projectm" targetNamespace="urn:tcl:projectm" xmlns:tns="urn:tcl:projectm" xmlns:wsdl="http://schemas.xmlsoap.org/wsdl/" xmlns:wsoap="http://schemas.xmlsoap.org/wsdl/soap/" xmlns:whttp="http://schemas.xmlsoap.org/wsdl/http/" xmlns:xsd="http://www.w3.org/2001/XMLSchema">

In the above definition you can see there exists an "xmlns:projectm". This is not automatically added by TWiST to the definition (::wsdl::definitions::new), but I manually ammended the wsdl::definitions::new procedure and added it there using:

lappend attributes "xmlns:projectm" "$targetNamespace"

right after

lappend attributes "xmlns:tns" "$targetNamespace"

My assumption would be that TWiST is adding the Namespace which I use (projectm), especially if used in the type name for the elements in a sequence, automatically in the definition. That is not the case when using the stock examples provided by Tom and cause Eclipse to break. After I inserted the code which resulted in the insert of xmlns:projectm="..." in the WSDL definition, Eclipse worked and the partners where able to access the webservices provided.

Collapse
Posted by Tom Jackson on
Yes, you are correct about one thing: projectm shows up as a prefix in the xml schema. However, your fix does not remove this prefix. I noticed this issue when looking into the wsdlpull issue. However, as you noted yourself, my example site doesn't work with wsdlpull either. But it also doesn't have the extra prefix. (This is a slightly older version than the trunk code):

http://junom.com/ws/datetime/?WSDL

So what is the bug? TWiST creates a mapping of the xml namespace to the prefix of your choice. However, it fails to trim off the leading ::, it passes ::projectm to tWSDL when creating the map. That is probably the bug. However, tWSDL could just trim this, because the leading : are illegal. So in the future, if another user API is developed, if the same mistake is made, the mapping will still work.

Now, during writing the WSDL file, the targetNamespace is your xml namespace. When you include types which are in the targetNamespace, the prefix is replaced with 'tns'. If an incorrect map is used, the replacement doesn't take place.

In the future, multiple namespaces will be supported, and this will be handled by creating another schema in the wsdl file and importing it into the targetNamespace. But, the schema for the targetNamespace will always use either the tns prefix or the built in xml-schema namespace.

The wsdlpull issue is a problem with wsdlpull, not TWiST. The problem will go away with the next update. Hopefully very soon. wsdlpull does not understand renaming of simpleTypes, at least within the same namespace. The problem will go away because the renaming isn't necessary anymore. I plan, eventually, on moving to a directed graph construction to plan the order of writing out xml schemas, but an intermediate solution is on the way.

Your current fix with the lappend is not actually evil, just in the incorrect place. I think you can achieve the same effect by setting:

::wsdb::definitions::projectmServer::definitions::.ATTRS(xmlns:projectm)

to the targetNamespace. This will only affect the projectm wsdl.

You can also set the targetNamespace instead of accepting the default 'urn:tcl:yourns':

&lt;ws&gt;namespace init ::stock
&lt;ws&gt;namespace schema ::stock "http://junom.com/stockquoter";

But a simple fix is here:

in wsdl/tcl/wsdl-schema-procs.tcl

in proc ::wsdl::schema::new, add a trim of schemaAlias:

    if {!$initialized} {
	initDatabase
    }
    # trim schemaAlias for buggy user APIs
    set schemaAlias [string trim $schemaAlias :]
Collapse
Posted by Malte Sussdorff on
I like the idea of having the prefix default to tns: if the namespace is actually the target namespace and your fix just does that. I assume (and here I have to wait for the partners tomorrow, as we are in Europe and it is late already) that this is fixing the issue Eclipse has been having as well, as they had been able to parse correctly your code at junom.com. Thanks for that simple fix.

As for naming the schema using "<ws>namespace schema ::projectm "http://e4.cognovis.biz/twsdl/projectmanager" ", I remembered it not adding an additional xmlns::projectm="..." to the definitions, just setting the targetNamespace differently. (Look at the source code of http://e4.cognovis.biz/twsdl/wsapi/stockquoter/?WSDL which is using your example). Sadly this still is an issue Eclipse is choking at (but do not ask me why...). So, if you could add the first call (adding the Namespace as an xmlns: definition) when doing the second (defining a custom schema), I assume it would work. Though I think for most implementations the fix you provided should be just fine.

One thing in your posting caught my attention.

"wsdlpull does not understand renaming of simpleTypes, at least within the same namespace. The problem will go away because the renaming isn't necessary anymore."

I was wondering about that. If I use <ws>proc ::projectm:GetTask {task_id:integer}, you are creating the custom "simpleType" with the name "project_id" and the base="xsd:integer". When I (during testing) made sure that not the custom type but the original xsd:integer was used as the type in the xsd:element, it worked. Obviously that is not a  solution if you want to have further restrictions like enumeration or patterns. But for strictly the simple types I don't think you need to create a custom simpleType, but again, I haven't read the specs thoroughly enough to say that is allowed by the standard. I can only speak for wsdlpull behaviour in this particular instance.

Collapse
Posted by Tom Jackson on
"As for naming the schema using "&lt;ws&gt;namespace schema ::projectm "http://e4.cognovis.biz/twsdl/projectmanager"; ", I remembered it not adding an additional xmlns::projectm="..." to the definitions,..."

You are right, it doesn't fix the bug, it replaces the default targetNamespace with your choice. Note that a lot of defaults exist as variables in &lt;ws&gt;namespace.

The problem with wsdlpull is that there is another layer of derivation for the types which it isn't handling correctly by wsdlpull. I believe the schema is correct. One way to test is to remove the intermediate type and point directly to the global simpleType. I'm guessing, but I think that wsdlpull is interpreting the definition as empty:

&lt;xsd:simpleType name="DayNumber"&gt;
    &lt;xsd:restriction base="tns:dayNumber"/&gt;
&lt/xsd:simpleType&gt;
Collapse
Posted by vivek krishna on
Hello,

As the developer of wsdlpull,somebody brought my attention to
this bug. as far as i understand its a bug in simple types.One such bug is caused by the creating another type DayNumber which is just a wrapper for "daynumber" note the case. Due to this the check for valid integers in the enumeration fails for the 3rd operation in the http://junom.com/ws/datetime/?WSDL because wsdlpull does not check validity recursively up the type tree yet. I am workin on fixing this but an immediate fix would be to remove the (redundant ,IMHO)wrapper
type "DayNumber". Feel free to contact me if you do have issues in wsdlpull

vivek Krishna mailto:krishna.vivek@gmail.com

Collapse
Posted by Tom Jackson on
Vivek,

Thanks for keeping track of this discussion. Personally I like wsdlpull, which is why I provided links to your software. I immediately recognized limitations for the built in client 'wsdl'. These limitations are to be expected in this field, especially with a simple example use of the API. The point is that the WSDL is correct, but individual clients may not take into account every conceivable representation of data types. This is not necessarily a bug for the client. In my opinion, XML Schema provides so many ways to represent the same structures, that it is unreasonable to expect independent client and server software to 'understand' each other.

I agree with Vivek that there is an unnecessary, although legal, layer (at the moment) in the type definitions of TWiST. This will be removed as I have stated before. But this fix will not make wsdlpull interoperate fully with TWiST. There will still be limitations with types based upon patterns and complexTypes beyond what a simple array could represent. My understanding is that the 'wsdl' client of wsdlpull is merely an example use of the underlying API. I like it, and it helps me check for certain errors. But it is unreasonable to expect that any client or server will ever support much of the XML Schema standard for type definition. There are a handful of companies investing billions into this problem with little success. XML Schema is something of a disaster in flexibility. In response, I have placed well defined limits on how types are represented in tWSDL.

As Vivek has mentioned, a tool in the cure is to represent types as a tree going back to the built in types. I wish to use a tree so that the XML Schema is as simple as possible, but this might also help the clients. At least clients can apply a rational and consistent process in reading the WSDL and developing the type validation code. However, as an intermediate step, or maybe a final step I have decided to flatten the type hierarchy. Every complexType is at a global level, allowing it to be reused as a child of other complexTypes. Elements with a simpleType content will be protected inside their parent elements, but their simpleType will be exposed as a global type in the case that the type is a derived type. But as I said, even this simplification will not guarantee that the wsdlpull client will interoperate in every case.