Gregory Szorc <gregory.szorc@gmail.com> [Wed, 07 Feb 2018 20:19:06 -0800] rev 36067
wireproto: use maybecapturestdio() for push responses (API)
The "pushres" and "pusherr" response types currently call
proto.restore() in the HTTP protocol. This completes the pairing
with proto.redirect() that occurs in the @wireprotocommand
functions. (But since the SSH protocol has a no-op redirect(),
it doesn't bother calling restore() because it would also be
a no-op.)
Having the disconnect between these paired calls is very confusing.
Knowing that you must use proto.redirect() if returning a "pushres"
or a "pusherr" is even wonkier.
We replace this confusing code with our new context manager for
[maybe] capturing output.
The "pushres" and "pusherr" types have gained an "output" argument
to their constructor and an attribute to hold captured data. The
HTTP protocol now retrieves output from these objects.
.. api::
``wireproto.pushres`` and ``wireproto.pusherr`` now explicitly
track stdio output.
Differential Revision: https://phab.mercurial-scm.org/D2082
Gregory Szorc <gregory.szorc@gmail.com> [Wed, 07 Feb 2018 20:17:47 -0800] rev 36066
wireprotoserver: add context manager mechanism for redirecting stdio
Today, proto.redirect() sets up redirecting stdio and proto.restore()
undoes that. The API is a bit wonky because restore() is only
implemented on the HTTP protocol. Furthermore, not all calls to
redirect() are obviously paired with calls to restore(). For
example, the call to restore() for "unbundle" requests is handled
by the response handler for the HTTP protocol.
This commit introduces a new method on the protocol handler interface
to maybe capture stdio. It emits a file object or None depending on
whether stdio capture is used by the transport.
To prove it works, the "pushkey" wire protocol command has been
updated to use the new API.
I'm not convinced this is the best mechanism to capture stdio. I may
need to come up with something better once the new wire protocol
emerges into existence. But it is strictly better than before because
it removes variance in the wire protocol handler interface. It
therefore gets us closer to a unified interface between the SSH and
HTTP transports.
Differential Revision: https://phab.mercurial-scm.org/D2081
Gregory Szorc <gregory.szorc@gmail.com> [Wed, 07 Feb 2018 20:17:05 -0800] rev 36065
wireprotoserver: split ssh protocol handler and server
We want to formalize the interface for protocol handlers. Today,
server functionality (which is domain specific) is interleaved
with protocol handling functionality (which conforms to a generic
interface) in the sshserver class.
This commit splits the ssh protocol handling code out of the
sshserver class.
Differential Revision: https://phab.mercurial-scm.org/D2080