dict/mapping values: do a simple verification of the value's
STIX version, not just its type. Added a lot more unit tests to
test behavior on dicts. To make the implementation work, I had
to move the detect_spec_version() function out of the parsing
module and into utils. So that required small changes at all
its previous call sites.
level with STIX versions in the same format as is used everywhere
else in the API: "X.Y", as opposed to the "vXY" format used by
the version-specific python packages. This eliminates all of
the awkward conversion from public API format to "vXX" format.
Also a little bit of code rearranging in the registration module
to ensure that some STIX 2.1-specific checks are done whether
version 2.1 is given explicitly or is defaulted to.
In the same module I also added a missing import of
stix2.properties, since my IDE was claiming it could not find a
function from that module.
- stix2.registry, which contains the class mapping structure
and code for scanning stix2 modules for its initial population
- stix2.registration, which contains code used to register custom
STIX types with the registry
- stix2.parsing, which contains code for creating instances of
registered stix2 classes from raw dicts.
This is intended to reduce circular import problems, by giving
dependent code the ability to import a module which has exactly
the functionality it needs, without pulling a lot of other stuff
it doesn't need. Fewer imports means less chance of an import
cycle.
* new packages for graph and object-based semantic equivalence
* new method graphically_equivalent for Environment, move equivalence methods out
* object equivalence function, methods used for object-based moved here.
* new graph_equivalence methods
* add notes
* add support for versioning checks (default disabled)
* new tests to cover graph equivalence and new methods
* added more imports to environment.py to prevent breaking changes
* variable changes, new fields for checks, reset depth check per call
* flexibility when object is not available on graph.
* refactor debug logging message
* new file stix2.equivalence.graph_equivalence.rst and stix2.equivalence.object_equivalence.rst for docs
* API documentation for new modules
* additional text required to build docs
* add more test methods for list_semantic_check an graphically_equivalent/versioning
* add logging debug messages, code clean-up
* include individual scoring on results dict, fix issue on list_semantic_check not keeping highest score
* include results as summary in prop_scores, minor tweaks
* Update __init__.py
doctrings update
* apply feedback from pull request
- rename semantic_check to reference_check
- rename modules to graph and object respectively to eliminate redundancy
- remove created_by_ref and object_marking_refs from graph WEIGHTS and rebalance
* update docs/ entries
* add more checks, make max score based on actual objects checked instead of the full list, only create entry when type is present in WEIGHTS dictionary
update tests to reflect changes
* rename package patterns -> pattern
* documentation, moving weights around
* more documentation moving
* rename WEIGHTS variable for graph_equivalence
Made them consistent with _register_observable_extension, by:
- moving validation logic there from _custom_*_builder functions
- using a new function for ensuring properties are dict-like
- using the library default spec version instead of None
Fix#371, fix#372, fix#373.
TypeProperty uses a fixed value, so check() was never called. This way
also runs the check at object registration time because the wrapper
creates an instance of TypeProperty and doesn't have to wait for the
object to be instantiated so clean() can be called.
Also fix some tests.
and sometimes a STIX property name. It didn't work (caused
crashes under some circumstances). Now, attributes whose names
conflict with Mapping methods will have the Mapping
interpretation. Same-named STIX object properties will not be
accessible as attributes.
cleaned" property, but actually, the cleaning algorithm works on
a dict copy, so aborting cleaning partway through doesn't
actually affect the object in that way. It would actually cause
the extensions property to be completely uncleaned, rather than
partially cleaned.
it doesn't make sense to have a test per STIX version. The
workbench only uses the latest supported STIX version. In
order to make this work, the test suite was modified to
dynamically compute some settings like where to get demo data,
based on the value of stix2.DEFAULT_VERSION.
Switched stix2.DEFAULT_VERSION back to "2.0", since I figure it
should be sync'd up with the 'from .vxx import *' import
statement from the top level package.
additionally required:
- Removing the v21 workbench test suite and reinstating the v20
test suite
- Fixing up a few v20 unit tests to work with the workbench
monkeypatching.
- I didn't revert the analogous changes I'd previously made to
the v21 unit tests, because I think they make sense even when
the workbench monkeypatching isn't happening.
- Removed all plain python base classes (e.g. ValueError, TypeError)
- Renamed InvalidPropertyConfigurationError -> PropertyPresenceError,
since incorrect values could be considered a property config error, and
I really just wanted this class to apply to presence (co-)constraint
violations.
- Added ObjectConfigurationError as a superclass of InvalidValueError,
PropertyPresenceError, and any other exception that could be raised
during _STIXBase object init, which is when the spec compliance
checks happen. This class is intended to represent general spec
violations.
- Did some class reordering in exceptions.py, so all the
ObjectConfigurationError subclasses were together.
Changed how property "cleaning" errors were handled:
- Previous docs said they should all be ValueErrors, but that would require
extra exception check-and-replace complexity in the property
implementations, so that requirement is removed. Doc is changed to just
say that cleaning problems should cause exceptions to be raised.
_STIXBase._check_property() now handles most exception types, not just
ValueError.
- Decided to try chaining the original clean error to the InvalidValueError,
in case the extra diagnostics would be helpful in the future. This is
done via 'six' adapter function and only works on python3.
- A small amount of testing was removed, since it was looking at custom
exception properties which became unavailable once the exception was
replaced with InvalidValueError.
Did another pass through unit tests to fix breakage caused by the changed
exception class hierarchy.
Removed unnecessary observable extension handling code from
parse_observable(), since it was all duplicated in ExtensionsProperty.
The redundant code in parse_observable() had different exception behavior
than ExtensionsProperty, which makes the API inconsistent and unit tests
more complicated. (Problems in ExtensionsProperty get replaced with
InvalidValueError, but extensions problems handled directly in
parse_observable() don't get the same replacement, and so the exception
type is different.)
Redid the workbench monkeypatching. The old way was impossible to make
work, and had caused ugly ripple effect hackage in other parts of the
codebase. Now, it replaces the global object maps with factory functions
which behave the same way when called, as real classes. Had to fix up a
few unit tests to get them all passing with this monkeypatching in place.
Also remove all the xfail markings in the workbench test suite, since all
tests now pass.
Since workbench monkeypatching isn't currently affecting any unit tests,
tox.ini was simplified to remove the special-casing for running the
workbench tests.
Removed the v20 workbench test suite, since the workbench currently only
works with the latest stix object version.
partly inspired by PR #263. This resulted in some error message
format changes (an improvement, I think), which caused some
unit test breakage. Removed those asserts from the unit tests,
since tests shouldn't be testing human-targeted error messages.
Michael Chisholm