Mike Beckerle created DAFFODIL-2326:
---------------------------------------
Summary: Do not pass RuntimeData to DPathCompileInfo constructor
Key: DAFFODIL-2326
URL: https://issues.apache.org/jira/browse/DAFFODIL-2326
Project: Daffodil
Issue Type: Improvement
Components: Back End, Middle "End"
Affects Versions: 2.5.0
Reporter: Mike Beckerle
Fix For: 3.0.0
DPathCompileInfo should not be passed a RuntimeData to its constructor.
This causes a great deal of circularity between RuntimeData objects and
computations that provide the values needed in those RuntimeData objects.
This prevents simplifying the constructors of the RuntimeData objects
(DAFFODIL-2325), and causes them to hold onto large compile-time structures
unnecessarily (DAFFODIL-1879)
Any computation that uses a check of an Evaluatable if it isConstant,
(knownEncoding for example asks EncodingEv.isConstant), well.... constructing
an evaluatable requires a DPathCompileInfo object, and ..... here's the
problem: right now we pass a RuntimeData object to the DPathCompileInfo
constructor purely for use compiling type-calc-related functions.
The type-calc functions need to know about sibling elements. The DPath compile
info objects don't have sibling info. But the RuntimeData objects do. So that's
where the expression compiler gets the sibling info it needs.
If some sibling information is needed to compile these functions, then we
should compute that information and pass it to the DPathCompileInfo
constructor, not just pass in the RuntimeData object.
Note however. The whole set of type-calc related functions is an experimental
feature and subject to review. There are proposals to simplify it
significantly. It's possible these functions will just not be needed. And in
that case, the sibling info won't be needed in the DPathCompileInfo objects.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
