Loading ...
Sorry, an error occurred while loading the content.
 

Re: [soapbuilders] Multiple WS-Addresses in multiple namespaces

Expand Messages
  • noah_mendelsohn@us.ibm.com
    ... One of my great regrets about the SOAP Recommendation is that it does not make crystal clear an aspect of the design that I considered to be very
    Message 1 of 14 , Feb 6, 2007
      Paul Downey writes:

      > I raised this very issue in my position paper [1] to the W3C Workshop
      > on Enterprise computing, and see it as a fundamental consequence
      > of SOAP not having a "stack", but a "bag" (actually a graph with WSS)
      > which requires meta-data such as the much trumpeted WS-Policy to
      > unravel.

      One of my great regrets about the SOAP Recommendation is that it does not
      make crystal clear an aspect of the design that I considered to be very
      important. Still, the crucial function is there IMO:

      First of all, it's not true that what SOAP has is a "bag"; the header
      elements are siblings in an envelope Infoset, and the Infoset is ordered.
      Also, headers may have interacting semantics [1] :

      "Mandatory SOAP header blocks are presumed to somehow modify the semantics
      of other SOAP header blocks or SOAP body elements."

      Furthermore, and this is crucial, while SOAP itself mandates no fixed
      order when processing, headers can be defined to control the order [1]:

      "The processing of one or more SOAP header blocks MAY control or determine
      the order of processing for other SOAP header blocks and/or the SOAP body.
      For example, one could create a SOAP header block to force processing of
      other SOAP header blocks in lexical order. In the absence of such a
      controlling SOAP header block, the order of header and body processing is
      at the discretion of the SOAP node. Header blocks MAY be processed in
      arbitrary order. Header block processing MAY precede, MAY be interleaved
      with, or MAY follow processing of the SOAP body. For example, processing
      of a "begin transaction" header block would typically precede body
      processing, a "logging" function might run concurrently with body
      processing and a "commit transaction" header block might be honored
      following completion of all other work."

      This is not an accident. It's why we require that all mustUnderstand
      checking be done before any other work. That's what ensures you that if
      you have an mU header that says "do the headers in reverse order" or
      "alphabetical order" or more likely "do signatures first" then that
      header or those headers will necessarily be checked in time to determine
      an order.

      Furthermore, and this is the part I really wish had been highlighted a bit
      more, nothing says these must be separate headers. So, IMO, if you want
      to say >in the specification for wsa:To< "if there are multiple wsa:To
      headers, here's the rule for how to process them all in the presence of
      the others", you can do so. If they are marked mU then you can be sure
      that, at least per the SOAP spec, their specifications can conspire to
      determine an order.

      In fact, what I really wanted to see in the recommendation would be a
      statement along the lines of: "The specifications for headers that may
      coexist in a SOAP message must collectively describe the correct
      interpretation of the headers in combination as well as in isolation.
      Thus, specifications may be written for families of headers designed to be
      used together, to determine operation in the case where multiple instances
      of the same header appear etc. Such specifications may call for a given
      header to be "understood" only when accompanied or only if not accompanied
      by certain other headers. etc."

      I think that is in fact implicit in what is there, but obviously many
      users have missed it. Whether the typical stacks out there provide much
      help in supporting such combined interpretation and ordering is a
      different question, but the spec definitely anticipates it IMO.

      Noah

      [1] http://www.w3.org/TR/soap12-part1/#muprocessing
      [2] http://www.w3.org/TR/soap12-part1/#procsoapmsgs
      --------------------------------------
      Noah Mendelsohn
      IBM Corporation
      One Rogers Street
      Cambridge, MA 02142
      1-617-693-4036
      --------------------------------------
    • Steve Loughran
      On 2/7/07, noah_mendelsohn@us.ibm.com wrote: . ... aah. Soap1.2. Soap 1.1 says nothing about where the headers are validated,
      Message 2 of 14 , Feb 8, 2007
        On 2/7/07, noah_mendelsohn@... <noah_mendelsohn@...> wrote:
        ."
        >
        > Furthermore, and this is crucial, while SOAP itself mandates no fixed
        > order when processing, headers can be defined to control the order [1]:
        >
        > "The processing of one or more SOAP header blocks MAY control or determine
        > the order of processing for other SOAP header blocks and/or the SOAP body.
        > For example, one could create a SOAP header block to force processing of
        > other SOAP header blocks in lexical order. In the absence of such a
        > controlling SOAP header block, the order of header and body processing is
        > at the discretion of the SOAP node. Header blocks MAY be processed in
        > arbitrary order. Header block processing MAY precede, MAY be interleaved
        > with, or MAY follow processing of the SOAP body. For example, processing
        > of a "begin transaction" header block would typically precede body
        > processing, a "logging" function might run concurrently with body
        > processing and a "commit transaction" header block might be honored
        > following completion of all other work."
        >
        > This is not an accident. It's why we require that all mustUnderstand
        > checking be done before any other work. That's what ensures you that if
        > you have an mU header that says "do the headers in reverse order" or
        > "alphabetical order" or more likely "do signatures first" then that
        > header or those headers will necessarily be checked in time to determine
        > an order.
        >

        aah. Soap1.2. Soap 1.1 says nothing about where the headers are
        validated, only that they must be validated

        "A env:mustUnderstand value of "true" means that the SOAP node must
        process the header with the semantics described in that header's
        specification, or else generate a SOAP fault. Processing the header
        appropriately may include removing the header from any generated SOAP
        message, reinserting the header with the same or altered value, or
        inserting a new header. The inability to process a mandatory header
        requires that all further processing of the SOAP message cease, and a
        SOAP fault be generated. The message is not forwarded any further."

        This is why the release of Axis (1.0?) that didnt do mU checking until
        after the message had been handled was within the spirit of the spec,
        and not the law.

        SOAP1.2 implies that I must check all mu headers before having any
        side effect at all. hmm. My current stack lets you declare handlers in
        a chain, as axis has done forever, and sun's stack has done since last
        week:

        AddressedEchoEndpoint extends AlpineEndpoint {
        name "wsa-echo";
        handlers [
        LogCurrentMessageHandler:classname,
        AddressHandler:classname,
        MustUnderstandChecker:classname,
        EchoHandler:classname,
        LogCurrentMessageHandler:classname
        ];
        }

        There's no way with this chain-of-resposibility design to do advance
        checking of handling, or guarantee that there are no side effects (I
        log the message -is that a side effect?), before the message is handed
        for processing.

        I suppose I could modify the handler interface to add an extra method
        in which every handler indicates if it will process a specific mU
        header. But if every handler is empowered to transform the incoming
        message during its work phase, things get complex. You need to create
        a provisionally transformed doc (transform, without side effects) just
        to make sure the mu headers get processed right.

        Or I ignore that part of the spec on the grounds that its pretty much
        impossible for interop tests to catch and retain my existing
        SOAP1.1-compatible mU processing algorithm.

        > Furthermore, and this is the part I really wish had been highlighted a bit
        > more, nothing says these must be separate headers. So, IMO, if you want
        > to say >in the specification for wsa:To< "if there are multiple wsa:To
        > headers, here's the rule for how to process them all in the presence of
        > the others", you can do so. If they are marked mU then you can be sure
        > that, at least per the SOAP spec, their specifications can conspire to
        > determine an order.
        >

        well, its a shame they dont.

        -steve
      • noah_mendelsohn@us.ibm.com
        ... Well, my reading of the ws-i basic profile is that it says the same thing regarding SOAP 1.1 [1]: R1025 A RECEIVER MUST handle messages in such a way
        Message 3 of 14 , Feb 8, 2007
          Steve Loughran writes:

          > On 2/7/07, noah_mendelsohn@...
          > <noah_mendelsohn@...> wrote: ."
          >
          > > Furthermore, and this is crucial, while SOAP
          > > itself mandates no fixed order when processing,
          > > headers can be defined to control the order [1]:
          >
          > > "The processing of one or more SOAP header
          > > blocks MAY control or determine the order of
          > > processing for other SOAP header blocks and/or
          > > the SOAP body. For example, one could create a
          > > SOAP header block to force processing of other
          > > SOAP header blocks in lexical order. In the
          > > absence of such a controlling SOAP header block,
          > > the order of header and body processing is at
          > > the discretion of the SOAP node. Header blocks
          > > MAY be processed in arbitrary order. Header
          > > block processing MAY precede, MAY be interleaved
          > > with, or MAY follow processing of the SOAP
          > > body. For example, processing of a "begin
          > > transaction" header block would typically
          > > precede body processing, a "logging" function
          > > might run concurrently with body processing and
          > > a "commit transaction" header block might be
          > > honored following completion of all other work."
          >
          > > This is not an accident. It's why we require
          > > that all mustUnderstand checking be done before
          > > any other work. That's what ensures you that if
          > > you have an mU header that says "do the headers
          > > in reverse order" or "alphabetical order" or
          > > more likely "do signatures first" then that
          > > header or those headers will necessarily be
          > > checked in time to determine an order.
          > >
          >
          > aah. Soap1.2. Soap 1.1 says nothing about where
          > the headers are validated, only that they must be
          > validated

          Well, my reading of the ws-i basic profile is that it says the same thing
          regarding SOAP 1.1 [1]:

          "R1025 A RECEIVER MUST handle messages in such a way that it appears
          that all checking of mandatory header blocks is performed before any
          actual processing. "

          Like so much in the basic profile, it takes the SOAP 1.2 refinements to
          the processing model, and mandates their use with SOAP 1.1.

          > > "A env:mustUnderstand value of "true" means that
          > > the SOAP node must process the header with the
          > > semantics described in that header's
          > > specification, or else generate a SOAP
          > > fault. Processing the header appropriately may
          > > include removing the header from any generated
          > > SOAP message, reinserting the header with the same
          > > or altered value, or inserting a new header. The
          > > inability to process a mandatory header requires
          > > that all further processing of the SOAP message
          > > cease, and a SOAP fault be generated. The message
          > > is not forwarded any further."
          >
          > This is why the release of Axis (1.0?) that didnt
          > do mU checking until after the message had been
          > handled was within the spirit of the spec, and not
          > the law.
          >
          > SOAP1.2 implies that I must check all mu headers
          > before having any side effect at all. hmm. My
          > current stack lets you declare handlers in a
          > chain, as axis has done forever, and sun's stack
          > has done since last week:

          Well, I'm not sure what to say. The spec is clear that you must inspect
          the headers first, and I think there are defensible reasons for that. I
          believe most of the commercial implementations (not speaking for my
          employer, IBM here) are moving toward a model in which headers are
          buffered so that mU checking can be done and header processing
          appropriately ordered, but the body is not necessarily parsed and
          processed in advance. Thus, the body can be streamed, but not the
          headers.

          > AddressedEchoEndpoint extends AlpineEndpoint {
          > name "wsa-echo";
          > handlers [
          > LogCurrentMessageHandler:classname,
          > AddressHandler:classname,
          > MustUnderstandChecker:classname,
          > EchoHandler:classname,
          > LogCurrentMessageHandler:classname
          > ];
          > }
          >
          > There's no way with this chain-of-resposibility
          > design to do advance checking of handling, or
          > guarantee that there are no side effects (I log
          > the message -is that a side effect?), before the
          > message is handed for processing.
          >
          > I suppose I could modify the handler interface to
          > add an extra method in which every handler
          > indicates if it will process a specific mU
          > header. But if every handler is empowered to
          > transform the incoming message during its work
          > phase, things get complex. You need to create a
          > provisionally transformed doc (transform, without
          > side effects) just to make sure the mu headers get
          > processed right.

          To be precise, headers are empowered to change the interpretation of other
          parts of the message, or to affect whether you can "understand" other
          headers in the message, but NOT to transform the message. In particular,
          I don't see anything that says a header can cause parts of the message to
          be decrypted or decompressed in a manner that would lead to the appearance
          of new headers that were processed as if they were there from the start.

          What I do think is there implicitly for such cases is for the
          specification of a header to say: when I am present, the SOAP processing
          model is to be rerun from the start on a new infoset constructed in the
          following way (e.g. by decrypting parts of the current.) In a given
          single use of the SOAP processing model, the input envelope infoset is
          invariant, but the interpretations of the headers present may interact.
          Indeed, it's crucial that they do. Surely a digital signature header
          affects the semantics of that which is signed, in the sense that if the
          dsig doesn't check, you may want to act as if none of the other headers
          are there (since they may be the result of tampering.) SOAP is designed
          to enable such scenarios, as long as the specification for the header (in
          this case a dsig header) is written properly.

          > Or I ignore that part of the spec on the grounds
          > that its pretty much impossible for interop tests
          > to catch and retain my existing SOAP1.1-compatible
          > mU processing algorithm.

          Whether you wish your implementation to expose that full power is up to
          you. HOWEVER, if you do not, then the only conforming response is for you
          to act as if you do not in fact "understand" any headers that might
          conceivably have such interacting semantics that you can't handle. So,
          unless you know that a header doesn't affect others, then if it's mU
          you'll either have to mU fault or be nonconforming to ws-i basic profile
          and/or SOAP 1.2.


          > > Furthermore, and this is the part I really wish
          > > had been highlighted a bit more, nothing says
          > > these must be separate headers. So, IMO, if you
          > > want to say >in the specification for wsa:To<
          > > "if there are multiple wsa:To headers, here's
          > > the rule for how to process them all in the
          > > presence of the others", you can do so. If they
          > > are marked mU then you can be sure that, at
          > > least per the SOAP spec, their specifications
          > > can conspire to determine an order.
          > >
          >
          > well, its a shame they dont.

          Yes, well, I have many concerns about how the higher level WS* specs were
          written, and whether they took sufficient care to use SOAP's details
          properly. I'm sorry that's in fact causing you trouble.

          > -steve
          >

          Noah

          [1]
          http://www.ws-i.org/Profiles/BasicProfile-1.0-2004-04-16.html#refinement16468312



          --------------------------------------
          Noah Mendelsohn
          IBM Corporation
          One Rogers Street
          Cambridge, MA 02142
          1-617-693-4036
          --------------------------------------
        • Steve Loughran
          ... Sometimes I suspect complexity is the underlying problem. Like WSDL. Its almost impossible for humans to write, so what you get is a mess, compared to,
          Message 4 of 14 , Feb 8, 2007
            On 2/8/07, noah_mendelsohn@... <noah_mendelsohn@...> wrote:
            > Steve Loughran writes:
            >
            > > On 2/7/07, noah_mendelsohn@...
            > > <noah_mendelsohn@...> wrote: ."
            >
            >
            > > > Furthermore, and this is the part I really wish
            > > > had been highlighted a bit more, nothing says
            > > > these must be separate headers. So, IMO, if you
            > > > want to say >in the specification for wsa:To<
            > > > "if there are multiple wsa:To headers, here's
            > > > the rule for how to process them all in the
            > > > presence of the others", you can do so. If they
            > > > are marked mU then you can be sure that, at
            > > > least per the SOAP spec, their specifications
            > > > can conspire to determine an order.
            > > >
            > >
            > > well, its a shame they dont.
            >
            > Yes, well, I have many concerns about how the higher level WS* specs were
            > written, and whether they took sufficient care to use SOAP's details
            > properly. I'm sorry that's in fact causing you trouble.
            >

            Sometimes I suspect complexity is the underlying problem. Like WSDL.
            Its almost impossible for humans to write, so what you get is a mess,
            compared to, say, COM IDL interfaces.

            As for the higher order specs, well, I relish their inconsistency,
            epecially WSRF, that has explicit dependencies on two different draft
            WSA versions, and punts on the whole problem of whether not bulk
            attribute read/write operations are atomic or not.

            But that's an OASIS problem so I won't be giving my local TAG
            representative a hard time, about it, or you. Liked your W3C
            submission to the web of enterprisey services; interesting contrast to
            the IBM 'we want a single stack' story.

            -steve
          Your message has been successfully submitted and would be delivered to recipients shortly.