The major, breaking change is that it is no longer possible to "take" a reply in
order to block the effect of a later addon hook.
This is patch 4/4 of the reply-ectomy.
* Partial gRPC contentview prototype, not linted, no tests, not as add-on
* Linted (flake8)
* Save dev state
* Rewrote of protobuf parser, use decoding strategy, reduced rendered data. Parser uses generators
* minor cleanup
* fix: preferred encoding was provided as function instead of value
* flake8: line length
* Backlinked message tree objects, temporary debug out
* Partial implementation of gRPC definitions. Save state to fix a cras (data invalidate in edit mode)
* hack: deal with missing exception handling for generator based content views
* gRPC/Protoparser descriptions (with test code)
* replaced manual gzip decoding with mitmproxy.net.encoding.decode
* Refactored typing imports
* Reafctoring
* distinguish request vs response definitions, separate view config from parser config
* Code cleaning, moved customized protobuf definitions to example addon
* final cleanup
* changelog
* Stubs for tests
* Fixed render_riority of addon example
* Started adding tests
* Work on tests
* mypy
* Added pseudo encoder to tests, to cover special decodings
* Example addon test added
* finalized tests, no 100 percent coverage possible, see comments un uncovered code
* minor adjustments
* fixup tests
* Typos
Co-authored-by: Maximilian Hils <git@maximilianhils.com>
A common pattern in sans-io is to essentially match on all variants of
an enum, and then have a `else: raise AssertionError` in the end.
This increases robustness as we spot incomplete coverage instead of
silently passing. However, it makes no sense to explicitly test these
asserts.
Also add a "raw" export format.
The Python and Locust exports are hard to maintain, their tests are extremely
brittle, they didn't have full test coverage, and are by my guess very rarely
used. I feel the Locust export should certainly be an externally maintained
addon. The Python/requests export can come back if someone cares enough, and it
can be structured in a way we can maintain.
Branch coverage means lots of duplication in tests to hit all cases, and I'm
not sure it's worth it. At any rate, let's aim for statement coverage first.
