11 KiB
Implementation
Behavior Driven Development
In Behavior Driven Development (BDD), the important role of software integration and unit tests is extended to serve both the purposes of designing the human-machine interaction flow (user journey in UX terms) and of laying down a common ground for interaction between designers and stakeholders. In this Agile software development methodology the software testing suite is based on natural language units that grant a common understanding for all participants and observers.
To implement BDD the first step is that of mapping a series of combinable, cascading sentences to actual source code; this implementation is usually done manually by programmers that have knowledge of the higher level application protocol interface (API) that grants communication between the backend and the frontend of a software application. The BDD implementation can then be seen as an alternative frontend whose purpose is that of lowering the distance between expression and execution by means of utterances expressed in human language.
Far from giving an exhaustive description of BDD implementations and characteristics, this brief chapter intends to summarize the features of this approach where they specifically apply to the development goals of Zencode (previously stated) and the solution provided.
Referring to the Cucumber implementation of BDD, arguably the most popular in use by the industry to day and de-facto standard, the grammar of utterances is very simple and definible as a "cascading" flow indeed, since the fixed sequence of lines can follow only one fixed order:
- Given
- and*
- When
- and*
- Then
This sequence is fixed and in simple terms consists of an extendable initialisation of states "Given (and)" followed by an extendable transformation of states "When (and)" and concluded by a non-extendable enunciation of states in their final form "Then".
The Zenroom implementation is kept simple at this stage and does not takes any "fuzzy" approach to the parsing, but simply defines fixed sequences of strings and variables that are expected to occur within them: the variables are what is ultimately possible to change by users and are marked by a repeating sequence of two adjacent single quotes ('').
The underlying parser acts upon a positive, unique and so far non-flexible match of the whole phrase minus the variables, then executes a function that takes as many arguments as the variables present in the lines across the utterance. As a result, every single non-repeating line of the utterance has a declared function that interacts with the underlying implementation of Zenroom, whose actions are defined in its LUA subset language.
Brief examples of this implementation follow:
Given("I introduce myself as ''", function(name) whoami = name end)
Given("I am known as ''", function(name) whoami = name end)
The above definition of two lines possibly occurring within the utterances in Zencode are demonstrating how a state "who am I" basically my own name can be set using two different phrases, leading to the execution of the same function which basically operates a simple assignment to the variable whoami. This simple demonstration is a hint to the fact that multiple patterns can be defined also in different ways, making the Zencode DSL implementation very easy to translate across different spoken languages as well contextualized within specific idiolects adopted by humans.
Furthermore, another example of implementation:
Given("that '' declares to be ''",function(who, decl)
-- declaration
if not declared then declared = decl
else declared = declared .." and ".. decl end
whois = who
end)
Given("declares also to be ''", function(decl)
ZEN.assert(who ~= "", "The subject making the declaration is unknown")
-- declaration
if not declared then declared = decl
else declared = declared .." and ".. decl end
end)
Shows how is possible to accept multiple variables and process them through more complex transformations that also contemplate the concatenation of contents to previous states. States are in fact permanent within the scope of the execution of a single utterance and will be modified in the same deterministic order by which they are expressed across lines. What is also visible within this example implementation, which we intend to facilitate by customisations made by people who have a simple knowledge of Zenroom's API and LUA scripting, is that the ZEN. namespace makes available a number of utility functions to easily check states (asserts) and propagate meaningful error messages that are then part of a traceback output given to the calling application (host) on occurrence of an error.
The full implementation of Zencode available at the time of publishing this document is inside the sourcecode files zenroom/src/lua/zencode_* and is relatively easy to maintain for the pilots analysed in our project, as well easy to extend to more usecases. At the dawn of piloting sessions, due to the lack of actual feedback so far given in field trials, this implementation does not addresses specific schemes beyond a simple Diffie-Helman asymmetric key encryption (AES-GCM) and an even simplier symmetric encryption of ciphertext by means of a PIN and KDF transformations. On top of that, perhaps the most complex implementation of Zencode so far is the "implicit certificate" crypto scheme (Qu-Vanstone, ECQV) still limited to first order curve transformations, which applies widely to pilots requiring simple certification schemes and is illustrated in more detail in the following chapters[^ecqv].
[^ecqv] It is important to note that while the ECQV scheme was not examined by other partners in our project, it has been choosen for its stable role in the industry and for its augmented complexity within an approachable implementation, complexity which could better inform the Zencode implementation. Without that complexity and without implementation feedback by other partners, it wouldn't have been possible to work on Zencode and bring it to what it is today, since both the Petition contract and the Coconut implementation in Zenroom are not available as of today and need to be completed in a later stage of the DECODE project.
Declarative Schema Validation
In order to make the processing of Zencode more robust, all data used as input and output for its computations is validated according to predefined schemas. This makes the Zencode DSL a declarative language in which data recognition is operated before processing.
The data schemas are added on a per-usecase basis: they refer to specific cryptographic implementations as they are added in Zencode. Careful evaluation regarding their addition is made to realise if old schemas can be extended to include new requirements.
Schemas are expressed in a simple format using Lua scripting syntax, for example:
-- zencode_keypair
keypair = S.record {
schema = S.Optional(S.string),
private = S.Optional(S.hex),
public = S.ecp
}
The schema above is the smallest and most commonly used one, composed by one required field and two optional ones, used to validate the input and output of public/private keypairs to be used in transformations.
The only required field in the schema is the public key which is validated using the ECP type (S. is an abbreviation for the SCHEMA. namespace). The validation of S.ECP is an actual cryptographic validation: Zenroom will check that the big integer number represented by the field corresponds to a valid point on the curve. In case the validation is not passed, the execution of the Zencode script will not take place and Zenroom will return a meaningful error message indicating the wrong field.
The other optional field is the private key which can correspond to any sequence of values, therefore no cryptographic validation is possible for it; in this case then the validation used is one that refers to the encoding of the field: S.hex is verifying that the value is encoded with a sequence of characters that express only hexadecimal numbers (that is, 0..9 numbers and case-insensitive letters from A to Z). Other encoding tests are also available, for instance S.base64 if that is the encoding used in the specific implementation.
Another more complex example follows:
-- packets encoded with AES GCM
AES-GCM = S.record {
checksum = S.hex,
iv = S.hex,
schema = S.Optional(S.string),
text = S.hex,
zenroom = S.Optional(S.string),
encoding = S.string,
curve = S.string,
pubkey = S.ecp
}
In this example no new validations are being used and in fact it just adds fields compared to the previous: it defines a portable packet of ciphertext data that is returned as output of AES-GCM asymmetric encryption as well is accepted as input to AES-GCM decryption. A similarity between these two examples is evident: the presence of the schema field. This field is a sort of "introspective" indication matching the data structure to its schema specification. If this field is not present (as it is always optional) then no validation on the data structure will take place, meaning the Zencode implementation leaves the risk (and hopefully the validation task) to the host.
This chapter ends with the current implementation of schema validation data types that are currently implemented for symmetric and asymmetric encryption of ciphertexts as well for implicit certificates. The schema implementation for Zencode is maintained into the sourcecode within the source file src/lua/zencode_schemas.lua and can be accessed by the function ZEN.validate(data,'schema','error') which is a wrapper of ZEN.assert(validate(data,schemas['schema']),'error').
_G['schemas'] = {
-- packets encoded with AES GCM
AES-GCM = S.record {
checksum = S.hex,
iv = S.hex,
schema = S.Optional(S.string),
text = S.hex,
zenroom = S.Optional(S.string),
encoding = S.string,
curve = S.string,
pubkey = S.ecp
},
-- zencode_keypair
keypair = S.record {
schema = S.Optional(S.string),
private = S.Optional(S.hex),
public = S.ecp
},
-- zencode_ecqv
certificate = S.record {
schema = S.Optional(S.string),
private = S.Optional(S.big),
public = S.ecp,
hash = S.big,
from = S.string,
authkey = S.ecp
},
certificate_hash = S.Record {
schema = S.Optional(S.string),
public = S.ecp,
requester = S.string,
statement = S.string,
certifier = S.string
},takes
declaration = S.record {
schema = S.Optional(S.string),
from = S.string,
to = S.string,
statement = S.string,
public = S.ecp
},
declaration_keypair = S.record {
schema = S.Optional(S.string),
requester = S.string,
statement = S.string,
public = S.ecp,
private = S.hex
}
}