- poetry: upgrade to 1.7.1
- poetry: python 3.7 support removed
- poetry: unnecessary tox tool removed
- poetry: deps version update

Fixes #900

Dashes can't be present in normal identifiers in Lua,
so we switch to a different syntactic sugar for the same thing.

manager: nits, cleanups, fixes

See merge request !1496

Removes references to pyright, which is not in use anymore. Also removes
warning suppressions and instead properly resolves the warnings.

This fixes a problem on some systems, where the `build.py` file
conflicts with the `build` module required by Poetry.

See <https://github.com/python-poetry/poetry/issues/7576>.

manager: cache-clear command via HTTP API

Closes #876

See merge request !1491

Aleš Mrázek authored 1 year ago and Oto Šťáva committed 1 year ago

- set: there is no need to specify the input data format
- get: YAML is now the default format for output data

Oto Šťáva authored 1 year ago and Aleš Mrázek committed 1 year ago

This commit adds a special JSON mode for control sockets.

The mode is activated by issuing a special `__json` command to the
socket, resulting in all Lua objects returned by all subsequent commands
to be serialized into JSONs, prepended by a 32-bit unsigned integer
byte-length value.

This JSON mode is now exclusively utilized by Manager, removing the need
to hackily strip single-quotes from the output and to read the output by
lines. Instead, it can always just read the 32-bit length value and
subsequently the whole JSON-formatted message, which is now
automatically deserialized into a Python object.

- only with libknot >= 3.4 though (which is not released yet)
- use stack instead of static buffer (saves RAM; see code comment)

There were some nontrivial conflicts to resolve, NEWS + the line
    ctx->vld_limit_crypto = KR_VLD_LIMIT_CRYPTO_DEFAULT;
(I had this resolution prepared for a long time.)

DNSSEC verification complexity could be exploited to exhaust CPU resources and stall DNS resolvers.

Solution boils down mainly to limiting crypto-validations per packet.

in a separate commit, as it will tend to conflict if patching

Improve: don't retry in this case.

Keep the first error in case priorities are equal.

At least with the current KeyTrap topic that should work better,
but blaming a single error is alchemy anyway, at least in some cases.