Skip to content

The Norway Problem - why StrictYAML refuses to do implicit typing and so should you

A while back I met an old coworker and he started telling me about this interesting bug he faced:

"So, we started internationalizing the website by creating a config file. We added the UK, Ireland, France and Germany at first."

countries:
- GB
- IE
- FR
- DE

"This was all fine. However, one day after a quick configuration change all hell broke loose. It turned out that while the UK, France and Germany were all fine, Norway was not..."

"While the website went down and we were losing money we chased down a number of loose ends until finally finding the root cause."

"If turned out that if feed this configuration file into pyyaml:"

countries:
- GB
- IE
- FR
- DE
- NO

"This is what you got in return:"

>>> from pyyaml import load
>>> load(the_configuration)
{'countries': ['GB', 'IE', 'FR', 'DE', False]}

It snows a lot in False.

When this is fed to code that expects a string of the form 'NO', then the code will usually break, often with a cryptic error, Typically it would be a KeyError when trying to use 'False' as a key in a dict when no such key exists.

It can be "quick fixed" by using quotes - a fix for sure, but kind of a hack - and by that time the damage is done:

countries:
- GB
- IE
- FR
- DE
- 'NO'

The most tragic aspect of this bug, however, is that it is intended behavior according to the YAML 1.2 specification. The real fix requires explicitly disregarding the spec - which is why most YAML parsers have it.

StrictYAML sidesteps this problem by ignoring key parts of the spec, in an attempt to create a "zero surprises" parser.

Everything is a string by default:

>>> from strictyaml import load
>>> load(the_configuration).data
{'countries': ['GB', 'IE', 'FR', 'DE', 'NO']}

String or float?

Norway is just the tip of the iceberg. The first time this problem hit me I was maintaining a configuration file of application versions. I had a file like this initially - which caused no issues:

python: 3.5.3
postgres: 9.3.0

However, if I changed it very slightly:

python: 3.5.3
postgres: 9.3

I started getting type errors because it was parsed like this:

>>> from ruamel.yaml import load
>>> load(versions) == [{"python": "3.5.3", "postgres": 9.3}]    # oops those *both* should have been strings

Again, this led to type errors in my code. Again, I 'quick fixed' it with quotes. However, the solution I really wanted was:

>>> from strictyaml import load
>>> load(versions) == [{"python": "3.5.3", "postgres": "9.3"}]    # that's better

The world's most buggy name

Christopher Null has a name that is notorious for breaking software code - airlines, banks, every bug caused by a programmer who didn't know a type from their elbow has hit him.

YAML, sadly, is no exception:

first name: Christopher
surname: Null
# Is it okay if we just call you Christopher None instead?
>>> load(name) == {"first name": "Christopher", "surname": None}

With StrictYAML:

>>> from strictyaml import load
>>> load(name) == {"first name": "Christopher", "surname": "Null"}

Type theoretical concerns

Type theory is a popular topic with regards to programming languages, where a well designed type system is regarded (rightly) as a yoke that can catch bugs at an early stage of development while poorly designed type systems provide fertile breeding ground for edge case bugs.

(it's equally true that extremely strict type systems require a lot more upfront and the law of diminishing returns applies to type strictness - a cogent answer to the question "why is so little software written in haskell?").

A less popular, although equally true idea is the notion that markup languages like YAML have the same issues with types - as demonstrated above.

User Experience

In a way, type systems can be considered both a mathematical concern and a UX device.

In the above, and in most cases, implicit typing represents a major violation of the UX principle of least astonishment.