Although unexpected, I think this behavior is actually correct. Since a docstring is a Python string, a \n inside the string will be parsed as a newline character by the Python parser, before the docformat parser even sees the string. If this is not wanted, a raw string can be used instead:

r"""
Code::

    x = 'a\nb'
"""

It would be worth documenting this though, because it is not intuitive.

The original ticket asked for both a public/private toggle and for a way to see what is private when everything is visible. We have the former, but not the latter. Do we want the latter as well?

For users that want to see everything, maybe we can add an argument to the URL's query? That would allow linking to the "show private definitions" version.

It would be most useful if these could be merged for the next release, we're also stuck with issues due to idna dependencies.

Motivation:

We have new ways of handling RPCs on the server, the server codec and state machine should use them!

Modifications:

• In the state machine: try to handle an RPC using the new API, falling back to the old way
• In the codec: store a mode, i.e. how we're handling the RPC.
• Flushing has become the responsibility of the server codec (as opposed to the RPC-specific channel handler): now we only flush at the end of 'channelReadComplete' (if a flush is pending), or on calls to 'flush(context:)' if we aren't currently reading.

Result:

The server supports the new RPC handlers.

@alex @tiran : Thanks for your feedback. The version cffi -->1.1 is not an explicit version mentioned in requirement.txt file in my py project. However, this version is getting installed due to dependent py module installation, i guess. @alex does cffi version 1.12 is handled to fix the AttributeError: Enter ?

When creating a secure websocket connection(wss) using websocket create_connection() api call in pyopenssl - openssl/ssl.py (20.0.1), AttributeError: enter error is thrown.

Complete error flow is given below :
line 514, in create_connection
websock.connect(url, **options)
line 226, in connect
self.handshake_response = handshake(self.sock, *addrs, **options)
line 76, in handshake
send(sock, header_str)
line 157, in send
return _send()
line 139, in _send
return sock.send(data)
line 1644, in send
with _from_buffer(buf) as data:
AttributeError: enter

Websocket call flow : websocket.create_connection() -> websock.connect->handshake() -> send(sock, header_str) -> sock.send(data)

Surprisingly, when using 19.1.0 of pyopenssl version, there was no such error observed.

Between these 2 versions, send() method in SSL.py file does not have any changes. surprised to see the error when i execute the test case after the update of library.

with _from_buffer(buf) as data:             
    if len(buf) > 2147483647:                 
        raise ValueError("Cannot send more than 2**31-1 bytes at once.")
    result = _lib.SSL_write(self._ssl, data, len(data))
    self._raise_ssl_error(self._ssl, result)
    return result

You'll need to update to cffi >= 0.12.

@alex cffi version installed is 1.1

That's very old. cffi 1.1.0 was released in 2015.

@alex cffi version installed is 1.1

We are using request's latest version but it uses certifi as depedency which is under MPL 2.0 license. This makes certifi a plugin with a copyleft license. Our company does not want to use any copyleft license plugin. I know that requests added certifi as separate plugin dependency after v 2.16 and onwards but I am not sure if its okay to use v2.15 because a lot of other bugs would also be fixed in latest versions. Is there a way we can exclude certifi depedency to avoid using a copyleft license plugin?

The MPL-licensed code included in certifi is not modified so doesn't trigger the MPL license terms. Please search for existing closed issues before opening a new one.

Motivation:

Manually wiring up the new handlers to run the rest of the test suite highlighted a few rough edges.

Modifications:

• Make sure compression is correctly set: it must be enabled on the server, in the call context, and - if applicable - on the individaul message. Add tests for this.
• Call the error delegate in the right place.
• Add a 'protocol violation' error.

Result:

Fewer bugs.

Motivation:

We have new ways of handling RPCs on the server, the server codec and state machine should use them!

Modifications:

• In the state machine: try to handle an RPC using the new API, falling back to the old way
• In the codec: store a mode, i.e. how we're handling the RPC.
• Flushing has become the responsibility of the server codec (as opposed to the RPC-specific channel handler): now we only flush at the end of 'channelReadComplete' (if a flush is pending), or on calls to 'flush(context:)' if we aren't currently reading.

Result:

The server supports the new RPC handlers.

SPMs for SwiftNIO builds in wrong order

Steps to reproduce:

1. Create new project
2. File -> Swift packages -> add package dependency Use https://github.com/grpc/grpc-swift.git exact Swift 1.0.0-alpha.21
3. build in XCode. Build succeeded
4. close xcode
5. run terminal
6. Goto folder with project cd ~/projects/<use your path>/
7. Run xcodebuild

Actual result: Library/Developer/Xcode/DerivedData/MultiScreen-gstsxmaaiuzazmcmkxnwlwsczybn/SourcePackages/checkouts/swift-nio-transport-services/Sources/NIOTransportServices/NIOFilterEmptyWritesHandler.swift:15:8: error: no such module 'NIO' import NIO

BUILD FAILED **

xcode 12.3

We are using request's latest version but it uses certifi as depedency which is under MPL 2.0 license. This makes certifi a plugin with a copyleft license. Our company does not want to use any copyleft license plugin. I know that requests added certifi as separate plugin dependency after v 2.16 and onwards but I am not sure if its okay to use v2.15 because a lot of other bugs would also be fixed in latest versions. Is there a way we can exclude certifi depedency to avoid using a copyleft license plugin?

…and codec

Motivation:

The HTTP2ToRawGRPCStateMachine returns a somewhat generic 'Action' which the called is meant to act on. This pattern was used wholesale for every interaction with the statemachine. This is a little silly: in some cases the cases any action can be specified just fine with a Bool, in others a Result is fine. The other downside to this approach is that it makes it harder to see at the callsite what could happen as a result of interacting with the state machine.

Modifications:

• Specify an action to take for each different interaction with the state machine.
• Update tests.

Result:

A handful of reductions in instructions in benchmarks:

• unary_10k_small_requests: -0.5%
• embedded_server_unary_10k_small_requests: -2.6%
• embedded_server_client_streaming_1_rpc_10k_small_requests: -4.0%
• embedded_server_client_streaming_10k_rpcs_1_small_requests: -2.6%
• embedded_server_server_streaming_1_rpc_10k_small_responses: -1.7%
• embedded_server_server_streaming_10k_rpcs_1_small_response: -2.6%
• embedded_server_bidi_1_rpc_10k_small_requests: -3.1%
• embedded_server_bidi_10k_rpcs_1_small_request: -2.4%

Breaks python ... -bb ..., until this is fixed, can be worked around by downgrading in pip:

urllib3==1.25.11

On the current latest documentation, there is a cookies link under the quickstart option.

We can see that this section should be displayed between Response Headers and Redirection History.

This section does not currently exist, and the link in the sidebar list is dead - clicking it does nothing.

Is this section missing, or should this link be removed?

Looks like the problem I originally reported is fixed now, and I can see it at the links you gave! Perhaps GitHub was serving an outdated cache at the time?

I am happy to close this, but will leave it for you guys to decide since it sounds like some other people might be having issues too.

Thanks for getting back to me.

It looks like there is a rendering error in the purple admonition... looks like raw HTML is not supported by pydoctor.

Everything looks good I don't know what you'r talking about. Here is the source code of the purple admonition patr:

.. raw:: html
    <style>
        .rst-admonition-purple {
            background-color: plum ! important;
        }
        .rst-admonition-purple p.rst-admonition-title{
            color: purple ! important;
        }
    </style>
.. admonition:: Purple
   This needs additionnal CSS for the new "rst-admonition-purple" class. 
   Include additional CSS by customizing the ``apidocs.css`` temlate file or by defining a raw block::
    .. raw:: html
    <style>
        .rst-admonition-purple {
            background-color: plum ! important;
        }
        .rst-admonition-purple p.rst-admonition-title{
            color: purple ! important;
        }
    </style>
   .. note:: The ``! important`` is required to overrride ``apidocs.css``. 

Excellent work.

Happy to see the RST demo. We might talk forever about minor improvements in the demo, but I think that it's a great start and can be merged ...and later have separate PR to fix minor issues.

For reference, maybe add it to PR description

https://pydoctor--345.org.readthedocs.build/en/345/docformat/restructuredtext/restructuredtext_demo.html

I think that this is a great start. I would prefer to see the demo in a separate PR so that we can merge it without waiting for admonition support.

It looks like there is a rendering error in the purple admonition... looks like raw HTML is not supported by pydoctor.

Thanks for creating the demo!

I guess that if you want to kill py2 for good, you can run https://github.com/PyCQA/modernize and have all the code updated in one go.