Based on experience creating DFDL schemas and running them using Daffodil, there are a number of things not currently described in the DFDL v1.0 standards documents that would be helpful in modeling.
They are described here in no particular order, along with some discussion of them.
DFDL v1.0 was designed to standardize the behaviors of many ad-hoc product-specific data format description capabilities.
One set of goals for DFDL v2.0 will be to advance the state-of-the-art, and describe data formats that none of the prior-generation ad-hoc product-specific data tools have been able to address effectively.
Recursion
One of the first things people want to model in DFDL always seems to be a binary legacy document formats like RTF or older MS Word documents. These have recursive structures where a section can contain text and other sections. DFDL v1.0 was not designed with document formats in mind, but rather with more traditional "data sets" or files of data in mind.
One advantage of DFDL without recursion is that it is not a 'turing complete' language. This is helpful from a security perspective. Adding recursion may break this boundary. Of course since DFDL has a rich regular expression capability, and its own backtracking, one can still create schemas that take absurdly long to execute even without recursion, so maybe this is not an issue.
Layering - Data Source/Target Indirection
Often one needs multiple passes. The value of some element, which might be a string, a hexBinary, or an array of bytes, wants to be used as the input for more parsing.
There are a whole bunch of related issues here. One also needs related functions in the expression language to allow concatenating a node list of strings into a single string (similarly for hexBinary and arrays of bytes).
The inverse of this layering for unparsing opens its own significant can of worms.
Complex Representations for Simple Types
XML Schema's simple vs. complex type distinction is quite painful. Often you want the logical result to be some computed element (using dfdl:inputValueCalc) of simple type, but one must have a hidden sequence group of several elements that are the more complicated representation details. In DFDL v1.0, one must of necessity model such a thing as a complex type, so that you have a place where both a hidden group and the 'value' element of simple type can live side by side.
A means is needed to embed a hidden group within the definition of a simple type, so that the hidden group is implicitly laid down next to the element having that simple type.
XML Attributes
The ability to have data of simple type become XML attribute values would go a long ways to making DFDL-created-XML more human-friendly.
More XML Schema Constructs
Several things in XML Schema seem to be missing. Unless there is a clear reason not to support them, it would be helpful. This list includes at least:
- repeating sequence and choice groups (minOccurs and maxOccurs)
- complex type derivations
- attributes (already mentioned above)
XML Schema 1.1
This new standard supports richer validation rules. They are useful since XML Schema 1.0's validation capabilities are so limited.
Delimited by Next Item
The ability to say that an element or group is delimited, but that it is delimited by the boundary of finding the initiator of the next element or group would simplify the description of many formats.
Another formulation would be to specify that an element/group is delimited, but that the terminating markup is not consumed, and hence, must be consumed by whatever comes next in the model.
Character Class Entities
We badly need an entity that means 'any whitespace that is not a line ending'. This avoids the specification of separators like:
dfdl:separator="%SP; %SP;%SP; %SP;%SP;%SP; %SP;%SP;%SP;%SP; %SP;%SP;%SP;%SP;%SP;"
which is sometimes needed when %NL; is the terminator and you want to distinguish the separator and terminator. The %WSP+; entity encompasses all whitespace.
A possible good name is %LSP*; where the "L" is for "Linear" as in "within a line".
Summary Functions/Operations
For both parsing and particularly for unparsing, one often must measure something. Fixed length formats often have tabular layouts in them, and the widths of the columns need to be computed from the longest string in the data.
This is a form of multi-pass (aka Layers), but for the unparsing case, it's really just about computing the length of something from values in the infoset, a capability DFDL already has. The need is to just generalize the calculation capability with some sort of map/reduce on arrays.
Extensions with User-defined Functions
Some computations are too complex to render directly in DFDL expressions. The ability to add functions in an orderly way is necessary.
Examples of this are computing the CRC for a network packet or checksums/hashes for other data structures, or encrypt/decrypt and compress/decompress.
These functions need to be able to examine the Daffodil processor state (Infoset and data streams).
Security Features
No Network Mode: This is less a DFDL language feature than a characteristic desirable for all implementations of DFDL. Applications using DFDL must be able to execute both in an environment which has no access to the internet, and even on machines that do have such access, in a mode where they make no attempt to access anything remotely.
Regular Expression Enhancements
DFDL schemas involve some large and complex regular expressions. Even the most advanced regular expression languages lack convenient ability to define a given construct once and name it, and then reuse it by somehow referencing that name. This would dramatically ease construction of regular expressions, and it is simply basic software engineering that large and complex things need to be named and reused, not duplicated.