Hacker News Reader

> I'm convinced the author was not aware of attributes, or for some reason held a grudge against them.

That is completely unsurprising, attributes suck ass for data, what would XML-RPC even have used them for? Replacing <int> by <integer width=“32”>?

Other serialisation formats make the same choice (e.g. plists) because it’s so much simpler.

> There's only like five primitive types, so implementations that want to transmit wild and unusual values such as 64-bit integers need to define their own ad-hoc extensions.

An issue which, which very much unfortunate, is not exactly shocking. Json would have the same if it had integers in the first place.

> including binary data, including NUL

Nul is a perfectly valid UTF8 character.

  > That is completely unsurprising, attributes suck ass for data,
  > what would XML-RPC even have used them for?

Instead of this:

  <struct>
    <member>
      <name>faultCode</name>
      <value><int>-123</int></value>
    </member>
    <member>
      <name>faultString</name>
      <value><string>some fault</string></value>
    </member>
  </struct>

It could have been:

  <struct>
    <member name="faultCode"><int>-123</int></member>
    <member name="faultString"><string>some fault</string></member>
  </struct>

or even:

  <struct>
    <int name="faultCode">-123</int>
    <string name="faultString">some fault</string>
  </struct>

--

  > An issue which, which very much unfortunate, is not exactly shocking.
  > Json would have the same if it had integers in the first place.

XDR (used in Sun RPC) was defined in the '80s and had 64-bit integers. The DEC Alpha line had been out for 5 years or so when XML-RPC was published. It's not like 64-bit integers were unusual at the time.

  > Nul is a perfectly valid UTF8 character.

NUL is not permitted in XML documents. The XML-RPC spec's explicit directive to allow NUL puts it at odds with standards-conforming XML parsers.

This reminds me that Apple's Plist[1] is similar too:

    <array>
        <dict>
            <key>CFBundleTypeName</key>
            <string>ShapeEditDocument</string>
            <key>LSHandlerRank</key>
            <string>Owner</string>
            <key>LSItemContentTypes</key>
            <array>
                <string>com.example.shape-doc</string>
            </array>
        </dict>
    </array>

( Taken from[2] )

Maybe it's common to design XML schema like this at that time?

[1] https://en.wikipedia.org/wiki/Property_list

[2] https://github.com/robovm/apple-ios-samples/blob/1f6b14ef6e2...

You’re absolutely correct. That’s how XML was written in the late 1990s.

Might be useful to look at what we did for the VWRAP abstract type system in the late 2000's. We generally eschewed attributes in XML and defined a subset of XML that could be automagically converted between JSON and a previously defined legacy binary format.

I'm certainly not saying this is how you should do things, but it's interesting to look at and has the feature that messages in one format (like XML) could be automagically transmogrified into another (like JSON.)

https://datatracker.ietf.org/doc/html/draft-ietf-vwrap-type-...

I feel that the complaints about "old tech", like comparing XML-RPC to XDR or the DEC Alpha at the time isn't very helpful, and maybe lacks some context. The specification came out in 1999, which means that people were discussing the spec for months or even years beforehand. Was any Microsoft-associated engineer using a 64-bit computer during that time? Did they even care? It's a guess on my part, but I wouldn't be surprised they couldn't put 64-bit in the spec since their competitors may have had patents in that space, and some tech companies weren't being friendly with each other at the time.

Also, messages that look like a HTML web page at the time also could have made it seemingly easier for developers to work with, and be attracted to use a new platform.

NT ran on Alpha and MIPS R4000, both of which were 64b architectures, though Windows only ran in 32b until XP64 (there was a 64b 2K prototype targeted at Itanium, but since Itanium ended up delayed it was shelved without release).

According to a post on xml.com[0], development of the spec that would become XML-RPC started in 1998. Microsoft was selling (32-bit) NT for Alpha at that point, and according to Raymond Chen[1] they had an internal 64-bit build of it.

I may be dating myself here, but a lot of folks new to the industry seem to think of anything pre-2000 as a dark time full of greyscale CRTs and dot-matrix printers. While those did exist in the home market, academic and business computing (where most of the advanced tech was being developed) was pretty advanced and a modern engineer would have felt right at home on a mid-90s UNIX workstation.

If you go back further you'll start encountering things like CORBA, which were pretty bad, but had the legitimate excuse of being very early. Remember that the '90s moved fast by modern standards -- CORBA was contemporary with MS-DOS, while XML-RPC was after IE had dethroned Netscape.

[0] https://www.xml.com/pub/a/ws/2001/04/04/soap.html

[1] https://docs.microsoft.com/en-us/previous-versions/technet-m...

Windows on the DEC Alpha was still in support at the time, I think, so someone at Microsoft must have used it.

Attributes suck when you're trying to write that data into a database. That's the main reason why lots of implementations enforced no attributes, because it made writing that data into a database table very hard.

NUL characters are not allowed in XML: https://stackoverflow.com/a/27938996/623763

The rationale probably being to enable storing XML content in zero-terminated C strings.

JSON was originally a transfer syntax, not a serialization format. But then when people wanted a definition, it got documented with type semantics. So there's all sort of mistakes that were made along the way...

At the time, the point was simple, non-binary RPC calls. I'm pretty sure that the spec came from a hacked up client and server, then frozen, with all warts intact. I'd be surprised if the initial (and final) spec had more than a handful of weeks of actual work in it. At the time, it did mostly work, and it did enable an explosion of connectivity in the early weblog adjacent world.

SOAP was basically XML-RPC++, with namespaces, 4 part harmony, and 8x10 color glossy photographs. It's the codification of XML-RPC adding all of those XML features, data types, and interface definitions. It's also a miracle if there's any interop between different stacks.

Compared to SOAP, this was pretty minimalistic and a pragmatic way to talk to a server from a browser. And of course people actually used SOAP as a really convoluted way to do RPC. This was the lightweight alternative. Json rpc also briefly was a thing.

I had the "pleasure" of dealing with a few SOAP web services a few times. That stuff was just completely horrible and unusable.

HTML5 became a thing when W3C began insisting that XHTML should be a thing, complete with namespaces and other nonsense. That was peak XML basically. Developers rejected it. Browser vendors and content creators basically stepped in and fixed HTML and CSS properly because w3c could not be bothered to step over the semantics and actually define how that stuff was supposed to behave in the real world (i.e. specifiying the semantics of their semantic HTML).

Come to think about it, there was a whole bunch of things that came out of w3c that essentially was about using namespaced XML: web service architecture, RDF, semantic web, XHTML, etc. All that stuff seems to have faded away into obscurity.

XML namespaces sound like a good idea until you realize that it just means endless verbosity that doesn't really serve much purpose and that really complicates things like parsing. Thankfully, people simply refused to apply that to json. Most json continues to be free of that nonsense. Same with yaml and similar formats. Works great without a lot of namespace urls.

It wasn’t “people” who saved JSON from this nonsense. It was Douglas Crockford, alone. I have been frequently annoyed at him for not allowing comments in JSON, but the explicit reason was that people were using json comments to encode all the metadata and crap from XML like namespaces to make fully reversible JSON/XML converters. And Crockford very much did not want JSON to be compatible with XML - he wanted it to be a simpler alternative. So good on him.

I hadn’t heard that explanation for not having comments, but if true, I want to buy him a drink.

> the explicit reason was that people were using json comments to encode all the metadata and crap from XML like namespaces to make fully reversible JSON/XML converters

Oh is that why. I had heard that comments were taken out because people were using it for "parser directives" or something vague. This makes a lot more sense (although I'm sure there were all manners of hacks in the comments). Definitely a bullet dodged there.

Oh many people have tried adding namespaces to json. It's just that these people and their solutions are consistently ignored by most developers. If you consider that Json came out of the javascript world where people tend to prefer to not specify a lot of types, it's perfectly obvious that they would probably not want or need any namespaces. Never was a thing in Json. And good on Douglas for keeping Json simple. The comment thing is annoying indeed. Using a good parser, you can usually configure it to not freak out over stuff like that. But sadly that is not very common.

Elasticsearch is a positive exception. It will happily accept comments.

I don't understand why is it hard to build a JSON/XML converter? XML is represented by a DOM and DOM should be easy to convert to JSON. Actually I did just that. I made a service which uses XML for inputs and outputs and those JavaScript people kept asking me about JSON. Not sure if they were happy in the end, JSON is too verbose to my taste, but whatever.

So some simple XML like

    <soap:Envelope xmlns:soap="http://schemas.xmlsoap.org/soap/envelope/">
        <soap:Body>
            <soap:Fault>
                <faultcode>soap:Server</faultcode>
                <detail>
                    <ns2:SendMessageFault1_SendMessageFault xmlns:ns2="http://bip.bee.kz/SyncChannel/v10/Types">
                        <errorCode>SCE013</errorCode>
                    </ns2:SendMessageFault1_SendMessageFault>
                </detail>
            </soap:Fault>
        </soap:Body>
    </soap:Envelope>

is represented by the following JSON monstrosity:

    {
      "name": "soap:Envelope",
      "attributes": [
        {
          "name": "xmlns:soap",
          "value": "http://schemas.xmlsoap.org/soap/envelope/"
        }
      ],
      "children": [
        "\n    ",
        {
          "name": "soap:Body",
          "children": [
            "\n        ",
            {
              "name": "soap:Fault",
              "children": [
                "\n            ",
                {
                  "name": "faultcode",
                  "children": [
                    "soap:Server"
                  ]
                },
                "\n            ",
                {
                  "name": "detail",
                  "children": [
                    "\n                ",
                    {
                      "name": "ns2:SendMessageFault1_SendMessageFault",
                      "attributes": [
                        {
                          "name": "xmlns:ns2",
                          "value": "http://bip.bee.kz/SyncChannel/v10/Types"
                        }
                      ],
                      "children": [
                        "\n                    ",
                        {
                          "name": "errorCode",
                          "children": [
                            "SCE013"
                          ]
                        },
                        "\n                "
                      ]
                    },
                    "\n            "
                  ]
                },
                "\n        "
              ]
            },
            "\n    "
          ]
        },
        "\n"
      ]
    }

Actually, that's what an automated translation would look like. A human would likely represent this as follows:

  { "errorCode "SCE013" }

That SOAP envelope is 99% just telling you it's a SOAP envelope, which technically is completely redundant because presumably you'd be aware of the fact that you are talking to a SOAP API. And apparently they had no schema for actually telling you anything meaningful about the error. Which is why it is represented as a string with an error code. That string is the only useful bit of information in the message.

Basically the SOAP equivalent of Internal Server Error.

From a parsing point of view it's horrible because all those namespaces need to be resolved and the schemas need to be processed. And of course a lot of those namespace urls don't actually have servers behind them that serve those schemas. Most of those urls never resolved to anything useful at all. The servers that were supposed to host those schemas never existed.

Alternatively you just ignore all that crap and pick the message apart with XPath or a regular expression. That's what I used to do when dealing with SOAP APIs. Most of those APIs were just stupendously convoluted RPC wrappers anyway from some bit of enterprise crap ware. I never could be bothered to treat those systems with the respect they demanded from me. Way faster to just sidestep all that nonsense. Send a message, pick apart what comes back to determine success / failure and move on with your life.

That's why SOAP died. Because as soon as people caught up to this being the optimal way to deal with SOAP, the next logical thing was to get rid of it entirely.

Not a super fan of JSON, not least because it has no comments, but that was totally the right call on his part. We'd have a ton of semi-de-facto crap to parse out of comments by now.

My favorite is DCE RPC, which specifies in every header if the data is little endian or big endian and most importantly, how floating point values are encoded (choose one: IEEE, VAX, CRAY, IBM) [1].

1: https://pubs.opengroup.org/onlinepubs/9629399/chap14.htm

> I'm convinced the author was not aware of attributes, or for some reason held a grudge against them.

The XML heyday is a little before my time, but I think attributes were considered improper design compared to only using tags by purists.

It is arguably "good design" because tags are infinitely extensible whereas attributes are not.

For example in the case the author supplied imagine if you wanted to add a compression type to a member. Your only choice would be an attribute which means that it can only be a string.

  <struct>
    <member name="faultCode"><int>-123</int></member>
    <member name="faultString"><string>some fault</string></member>
  </struct>

If in the original you could add a new element inside the member like

  <compression>
    <algorithm>zstd</algorithm>
    <dictionary>dict7</dictionary>
  </compression>

Not the best example but the point stands. Of course the complaint is valid because while extensibility is valuable you need to weigh it against the cost of the verbosity (both in use and for developers).

Note that in the specific case of XML-RPC, extensibility is forbidden by the spec:

  A <fault> struct may not contain members other than those specified.
  This is true for all other structures. We believe the specification
  is flexible enough so that all reasonable data-transfer needs can be
  accomodated within the specified structures. If you believe strongly
  that this is not true, please post a message on the discussion group.

At one point the author did intend to allow elements to contain user-defined children[0], but it seems like their position changed between that mailing list post and the spec update 7 days later.

[0] https://web.archive.org/web/19991010205056/http://discuss.us...

My recollection of the personalities involved is that the author is absolutely not a purist. He's a pragmatist, and this was simple, understandable, and worked well enough at the time. He was doing this because he needed the functionality, not because it was the best design possible.

I also think that this design was also reasonably well aligned with the internal XML capabilities of Frontier.

> My recollection of the personalities involved is that the author is absolutely not a purist.

The opposite really, half-assed "good enough for me" is also very much visible in RSS.

This brings back bad memories. I once worked on a mobile app that used XML-RPC for communications, in the late 2000's. I remember looking at packet captures of this stuff. Ugh.