Goldilocks and the 11 Parsers

Once upon a time there was a programmer, lost in the requirements specification.

"If only I had a standard parser to communicate structured dynamic data!" she said.

As chance would have it, hidden behind a footnote in subsection 4 of paragraph 14, she spied ITU X.680 and X.690.

"Oh!" she exclaimed, "I am saved! For the system described in these documents provides near infinite flexibility to represent even the smallest semantic nuance." And off she went down to the river to code an implementation.

Weeks later she emerged from the forest disheveled and half-starved. "ASN.1 tricked me!" she complained, "What I thought was flexibility was merely underspecification wrought from the inability of a technical committee to find agreement on implementation details. If I continue on this path I will be forced to decide between adherence to an often-vague specification or an implementation with decent performance. Most likely I will code only to the subset of the specification I intend to use leading to long-term brittleness and possibly security vulnerabilities based on incomplete understanding of a confusing specification." In reward for her understanding she was named as a director of the IETF Applications Area.

Two years later the programmer returned to her specification, tired of mediating technical disputes between commercial interests who were attempting to find competitive advantage by scuttling each other's proposals while simultaneously advancing the bare minimum specification required for a viable protocol.

"Clearly XML is the way to go," she said, "with its mix of representational flexibility and optional enhancements." And away she did go, coding real-time applications using XML to transfer data between loosely coupled manufacturing-floor process-control equipment. Everything was fine until it came time to integrate a newer version of the Smerxerces XML parser that referenced a version of libboost that did not run on their MIPS-based process control system due to a compiler bug.

After reading the code for Smerxerces, trying to understand why it was coded to require behavior specific to a particular version of libboost she said to herself, "Most of my requirements for data transfer are relatively simple, not requiring the expressive power of XML. What's more, we were forced to reduce the requirement that XML input be 'well formed' instead of 'valid' because XML components produced by non-coordinating, loosely coupled data systems do not test their components with each other and frequently reference schemas or DTDs which are unavailable to the public.:

In reward for her new-found enlightenment she was named to the ACM Publications Board, a position of great respect and moderate power within her community.

After two years of defending the ACM's policy to retain it's paywall, she quietly resigned to volunteer with the Internet Archive and EFF to atone for her mercantile sins. One day at the Internet Archive she saw a web developer cursing incompatibilities between browsers with respect to JSON parsing.

"Eureka!" she exclaimed, "Clearly JSON is the transfer syntax of a new generation!"

"All hail JSON!" she said to no-one in particular, "It eschews schema, working only as a transfer syntax allowing higher layers of the system to make decisions regarding missing or extra data in a protocol data unit!"

She turned this new-found discovery into a well-regarded TED talk and was rewarded by being named AOL's Internet of Things Architect.

As the IoT Architect she encouraged manufacturers of 4 and 8 bit sensors to produce PDUs using JSON. When visiting a vendor site she happened upon a Pacific Coast Mountain Octopus crafting a small parser to run on an 8 bit micro-controller powered only by the RF-field of an NFC card reader.

"Above. Around. Within." said the Octopus, "Context and regarding; limits and failure." But alas, she did not speak Octopus. The octopus then flashed a series of patterns on its skin, communicating sub-consciously using the language of dreams. But of course humans do not listen to their dreams during the day, so she had to wait for REM sleep for her brain to attempt to convert the flashing patterns into long-term memories.

In her dreams the Octopus said "Your appreciation for the simplicity of JSON has blinded you to the type semantics embedded in it's specification. We are facing issues because there is no sub-specification in JSON to allow a consumer to identify its type semantics. JSON does a good job of delimiting protocol data units, but fails completely to handle the case where a consumer supports only 8 or 16 bit integers, or 64 bit integers for that matter."

Because it was a dream, the Octopus then changed into David Bowie and threw minnows at her.

Changing back into an Octopus, it continued, "The transfer syntax itself doesn't describe the protocol interactions required to establish the type behavior of the consumer, but it can signal that the data provider 'promises' to use values within a given range. This doesn't completely solve the problem, but it allows the consumer to decide before-hand what to do when it receives data it may not be able to semantically consume."

And with this, the programmer woke, with near-mystical understanding of the Octopus' message.

Like Goldilocks she had wandered into the forest and found ASN.1 to be too heavy, weighted down with explicit type semantics. XML eschewed type semantics, but it's extensibility allowed producers to identify semantic intent in documents or document schemas, requiring consumers of XML to check both locations. XML too, was too heavy. JSON eschewed schema, making it "lightweight." But it was too lightweight in that it ignored type semantics of the consumer.

And that is when the programmer devised a fourth parser, one with the simplicity of JSON, but without JSON's embedded type semantics. She later hired 11 bears as interns to code the new parser which consumed data that looked suspiciously like JSON, but added support for in-band directives APDU producers use to communicate to the consumer the type semantics they intend to use.

And they all lived happily ever after.

Until someone pointed out that JSON doesn't support comments and since they were changing things, maybe add support for comments?

[also… the parser in question can be seen at -Ed.]