[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v3 01/15] python: create qemu packages

From: John Snow
Subject: Re: [PATCH v3 01/15] python: create qemu packages
Date: Wed, 28 Oct 2020 15:54:48 -0400
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.3.1

On 10/28/20 12:39 PM, Cleber Rosa wrote:
On Wed, Oct 28, 2020 at 11:21:19AM -0400, John Snow wrote:
On 10/28/20 10:46 AM, Cleber Rosa wrote:
On Tue, Oct 20, 2020 at 03:35:41PM -0400, John Snow wrote:
move python/qemu/*.py to python/qemu/[machine, qmp]/*.py and update import
directives across the tree.

This is done to create a PEP420 namespace package, in which we may
create subpackages. To do this, the namespace directory ("qemu") should
not have any modules in it. Those files will go into new 'machine' and 'qmp'
subpackages instead.

Implement machine/__init__.py making the top-level classes and functions
from its various modules available directly inside the package. Change
qmp.py to qmp/__init__.py similarly, such that all of the useful QMP
library classes are available directly from "qemu.qmp" instead of

Signed-off-by: John Snow <jsnow@redhat.com>
   python/{qemu => }/.isort.cfg                |  0
   python/qemu/__init__.py                     | 11 ------
   python/qemu/{ => machine}/.flake8           |  0
   python/qemu/machine/__init__.py             | 41 +++++++++++++++++++++
   python/qemu/{ => machine}/accel.py          |  0
   python/qemu/{ => machine}/console_socket.py |  0
   python/qemu/{ => machine}/machine.py        | 16 +++++---
   python/qemu/{ => machine}/pylintrc          |  0
   python/qemu/{ => machine}/qtest.py          |  3 +-
   python/qemu/{qmp.py => qmp/__init__.py}     | 12 +++++-
   tests/acceptance/boot_linux.py              |  3 +-
   tests/qemu-iotests/297                      |  2 +-
   tests/qemu-iotests/300                      |  4 +-
   tests/qemu-iotests/iotests.py               |  2 +-
   tests/vm/basevm.py                          |  3 +-
   15 files changed, 71 insertions(+), 26 deletions(-)
   rename python/{qemu => }/.isort.cfg (100%)
   delete mode 100644 python/qemu/__init__.py
   rename python/qemu/{ => machine}/.flake8 (100%)
   create mode 100644 python/qemu/machine/__init__.py
   rename python/qemu/{ => machine}/accel.py (100%)
   rename python/qemu/{ => machine}/console_socket.py (100%)
   rename python/qemu/{ => machine}/machine.py (98%)
   rename python/qemu/{ => machine}/pylintrc (100%)
   rename python/qemu/{ => machine}/qtest.py (99%)
   rename python/qemu/{qmp.py => qmp/__init__.py} (96%)

diff --git a/python/qemu/.isort.cfg b/python/.isort.cfg
similarity index 100%
rename from python/qemu/.isort.cfg
rename to python/.isort.cfg
diff --git a/python/qemu/__init__.py b/python/qemu/__init__.py
deleted file mode 100644
index 4ca06c34a4..0000000000
--- a/python/qemu/__init__.py
+++ /dev/null
@@ -1,11 +0,0 @@
-# QEMU library
-# Copyright (C) 2015-2016 Red Hat Inc.
-# Copyright (C) 2012 IBM Corp.
-# Authors:
-#  Fam Zheng <famz@redhat.com>
-# This work is licensed under the terms of the GNU GPL, version 2.  See
-# the COPYING file in the top-level directory.
diff --git a/python/qemu/.flake8 b/python/qemu/machine/.flake8
similarity index 100%
rename from python/qemu/.flake8
rename to python/qemu/machine/.flake8
diff --git a/python/qemu/machine/__init__.py b/python/qemu/machine/__init__.py
new file mode 100644
index 0000000000..27b0b19abd
--- /dev/null
+++ b/python/qemu/machine/__init__.py
@@ -0,0 +1,41 @@
+QEMU development and testing library.
+This library provides a few high-level classes for driving QEMU from a
+test suite, not intended for production use.
+- QEMUMachine: Configure and Boot a QEMU VM
+ - QEMUQtestMachine: VM class, with a qtest socket.
+- QEMUQtestProtocol: Connect to, send/receive qtest messages.
+- list_accel: List available accelerators
+- kvm_available: Probe for KVM support
+- tcg_available: Probe for TCG support
+# Copyright (C) 2020 John Snow for Red Hat Inc.
+# Copyright (C) 2015-2016 Red Hat Inc.
+# Copyright (C) 2012 IBM Corp.
+# Authors:
+#  John Snow <jsnow@redhat.com>
+#  Fam Zheng <fam@euphon.net>
+# This work is licensed under the terms of the GNU GPL, version 2.  See
+# the COPYING file in the top-level directory.
+from .accel import kvm_available, list_accel, tcg_available
+from .machine import QEMUMachine
+from .qtest import QEMUQtestMachine, QEMUQtestProtocol
+__all__ = (
+    'list_accel',
+    'kvm_available',
+    'tcg_available',
+    'QEMUMachine',
+    'QEMUQtestProtocol',
+    'QEMUQtestMachine',

How far would this approach go?  I mean, should all future APIs be developed
inside module under "qemu/machine", say "qemu/machine/foo.py" and their main
functionality imported here?  Or do you see this as a temporary state?

I assume any new functionality can be directly imported. Having this here
doesn't prevent using the fully qualified module names.

i.e. you can still do "from qemu.machine.machine import QEMUMachine".

Offering root-level aliases seems like a very prevalent paradigm in popular
third-party packages like 'requests'. Not justification in and of itself,
but there's precedent.

There's always someone doing something someway... I think the point
here is to try to deliver something that makes the most sense to most
users of the API.  Not an easy task, but it's our task.

Just gotta follow the patterns I see in prominent examples that are often held up as good citizens. In my experience, commonly used aliases for the purposes of ease-of-use is a common pattern.

I don't think this *hurts* anything, but I suppose there's deeper questions about what we want the imports to look like. OK, but I think this discussion about the use of __all__ winds up being unrelated to the actual matter at hand.

IMO, this is hard to maintain, and is a very easy path to future

Maybe. The alternative is to do what I did to QMP and stick it back inside
of __init__.

What I don't like is:

from qemu.machine.machine import QEMUMachine

I prefer:

from qemu.machine import QEMUMachine

Of course, I also like help(qemu.machine) to show something useful, which is
going to generally involve updating the docstring to explain what the major
APIs here are anyway, so I think actually you cannot escape your fate of
having to keep a list of APIs updated in a multi-module package.

Maybe you could make an argument that keeping both the docstring and the
__all__ attribute in sync is an undue maintenance burden, but I'd need some
more convincing.

diff --git a/python/qemu/accel.py b/python/qemu/machine/accel.py
similarity index 100%
rename from python/qemu/accel.py
rename to python/qemu/machine/accel.py
diff --git a/python/qemu/console_socket.py 
similarity index 100%
rename from python/qemu/console_socket.py
rename to python/qemu/machine/console_socket.py
diff --git a/python/qemu/machine.py b/python/qemu/machine/machine.py
similarity index 98%
rename from python/qemu/machine.py
rename to python/qemu/machine/machine.py
index 6420f01bed..a5dc305539 100644
--- a/python/qemu/machine.py
+++ b/python/qemu/machine/machine.py
@@ -38,8 +38,14 @@
-from . import console_socket, qmp
-from .qmp import QMPMessage, QMPReturnValue, SocketAddrT
+from qemu.qmp import (
+    QEMUMonitorProtocol,
+    QMPMessage,
+    QMPReturnValue,
+    SocketAddrT,
+from . import console_socket
   LOG = logging.getLogger(__name__)
@@ -139,7 +145,7 @@ def __init__(self,
           self._events: List[QMPMessage] = []
           self._iolog: Optional[str] = None
           self._qmp_set = True   # Enable QMP monitor by default.
-        self._qmp_connection: Optional[qmp.QEMUMonitorProtocol] = None
+        self._qmp_connection: Optional[QEMUMonitorProtocol] = None
           self._qemu_full_args: Tuple[str, ...] = ()
           self._temp_dir: Optional[str] = None
           self._launched = False
@@ -314,7 +320,7 @@ def _pre_launch(self) -> None:
               if self._remove_monitor_sockfile:
                   assert isinstance(self._monitor_address, str)
-            self._qmp_connection = qmp.QEMUMonitorProtocol(
+            self._qmp_connection = QEMUMonitorProtocol(
@@ -534,7 +540,7 @@ def set_qmp_monitor(self, enabled: bool = True) -> None:
           self._qmp_set = enabled
-    def _qmp(self) -> qmp.QEMUMonitorProtocol:
+    def _qmp(self) -> QEMUMonitorProtocol:
           if self._qmp_connection is None:
               raise QEMUMachineError("Attempt to access QMP with no 
           return self._qmp_connection
diff --git a/python/qemu/pylintrc b/python/qemu/machine/pylintrc
similarity index 100%
rename from python/qemu/pylintrc
rename to python/qemu/machine/pylintrc
diff --git a/python/qemu/qtest.py b/python/qemu/machine/qtest.py
similarity index 99%
rename from python/qemu/qtest.py
rename to python/qemu/machine/qtest.py
index 39a0cf62fe..53926e434a 100644
--- a/python/qemu/qtest.py
+++ b/python/qemu/machine/qtest.py
@@ -26,8 +26,9 @@
+from qemu.qmp import SocketAddrT
   from .machine import QEMUMachine
-from .qmp import SocketAddrT
   class QEMUQtestProtocol:
diff --git a/python/qemu/qmp.py b/python/qemu/qmp/__init__.py
similarity index 96%
rename from python/qemu/qmp.py
rename to python/qemu/qmp/__init__.py
index 2cd4d43036..9606248a3d 100644
--- a/python/qemu/qmp.py
+++ b/python/qemu/qmp/__init__.py
@@ -1,4 +1,14 @@
-""" QEMU Monitor Protocol Python class """
+QEMU Monitor Protocol (QMP) development library & tooling.
+This package provides a fairly low-level class for communicating to QMP
+protocol servers, as implemented by QEMU, the QEMU Guest Agent, and the
+QEMU Storage Daemon. This library is not intended for production use.
+`QEMUMonitorProtocol` is the primary class of interest, and all errors
+raised derive from `QMPError`.
   # Copyright (C) 2009, 2010 Red Hat Inc.
   # Authors:
diff --git a/tests/acceptance/boot_linux.py b/tests/acceptance/boot_linux.py
index 0055dc7cee..54f999e647 100644
--- a/tests/acceptance/boot_linux.py
+++ b/tests/acceptance/boot_linux.py
@@ -12,8 +12,7 @@
   from avocado_qemu import Test, BUILD_DIR
-from qemu.accel import kvm_available
-from qemu.accel import tcg_available
+from qemu.machine import kvm_available, tcg_available

I think this reveals a slight downgrade of the API caused by the usage
of the symbols from "qemu.machine" namespace.

My point is that there are valid use cases for using "kvm_available"
without using a "QEMUMachine".  This presuposes that "qemu.machine" is
named like that because "QEMUMachine" are the main actors of this

No, the accel functions literally start a QEMU process and probe its output
to determine what they support. This belongs directly alongside the machine
code in my opinion.

accel.kvm_available() may not even run QEMU at all... and when it
does, it runs it just like your most mundane command line application,
"grepping" its stdout.

(I also don't know why'd you call these functions unless you intended to
create a machine based on what you found.

*If* this is an API that you intend others to leverage, you should
reverse the question: "Is there any reason for not making this utility
available and as obvious and as easily available as possible to be
used by people whom I can't actually tell how they'd use it?".

For instance, I may want to write Python code that skips over a number
of tests if I detect that KVM is not available.  This code of mine,
may be completely separate from actual code creating the QEMUMachines,
or may even be compiled C code that I don't want to call because I
know that KVM is not available.

And that's not hypothetical.  I personally have a use case for that.

All of the usages in tests/acceptance pass a binary in. The vm/ tests are the ones that don't.

With tcg_available necessitating providing a binary, it's weird that kvm_available has an overloaded invocation where it avoids doing so. The generic detection for host KVM support likely ought to be in a different function.

(And that one, you are right, would have only the most tenuous relationship to a qemu.machine package.)

I see your point that it is not a method of the Machine class itself nor an
extension thereof; but I think for the purpose of three small helper
functions it's not worth creating an entirely new package to contain.

Not only that, but it's not even used by QEMUMachine itself.

But it's not wholly unrelated: I maintain the most obvious use for these is to probe a binary and adjust flags accordingly. I would go so far as to say that shipping qemu.machine without such a feature would create a deficit in that package!

Nor is it worth watering down the "machine" name to something less specific
to accommodate three tiny helper functions.

If I am considering "machine" the "package that knows how to start and
interact with QEMU", then accel belongs here.

That's pretty broad and generic.  Do you mean that "machine" should
know how to interact with, say, "info qtree"?  Would a utility that
outputs and parse that deserve to live under "machine"?

It is broad and generic. I am arguing explicitly against creating an entire package to house a single module which contains three utility functions. What I am really pressing on you to answer is: "What are they utility functions *for*?"

Answering that might give us good names. Without names, I can't respin the patches to your liking. :)

But, yes, that means I am considering "machine" as something very broad and generic. I believe that's correct until such time as we have justification to create something more targeted and focused. If you have a better name I'm open to it, but I did intent to intuit that the primary purpose of this package was to provide QEMUMachine and friends, because it is.

That's also why I have adopted Andrea's suggestion to start with a 0.x version: I am treating this domain as an incubation space, and not stable APIs -- precisely because this type of argument is not one I am sure needed to be solved up front. We can graduate packages out of this incubator.

What I had initially:

qemu.core, containing *everything*

What I have now:

qemu.qmp     - QMP library and tools that rely on the QMP library
qemu.machine - machine.py, console_socket.py, qtest.py, accel helpers

What you want, I believe:

qemu.qmp     - QMP library and misc QMP-using tools
qemu.machine - machine.py, console_socket.py, qtest.py
qemu.util    - accel helpers

What I'm protesting against is creating a "utilities" package at all, because with no central thesis to its existence, I do not know how to version or maintain such a package.

It's my belief that keeping the subpackages tightly focused (qmp, machine) it will help guide what kind of API we want to develop for each.

Making a "util" package to house random bits and pieces seems like the sort of thing that will be very hard to maintain in a meaningful way; and it's something I'd like to avoid supporting for external usage.

Please understand that these are the awkward, speculative and hard
questions that we have to ask *if* the intention is to make an API
available for general use.

I did invite bikeshedding! Now's the time to do it, if ever there was one.

Naming is hard, (so don't assume I believe "utils" below is the best
or even a good name), but I'd be more confortable using an API along
the lines:

     from qemu.utils.accel import kvm_available


     from qemu.utils import accel

utils/tools I am reserving as a package for various debugging scripts and
tools in the future that I don't intend to upload to PyPI, e.g. many of the
"thematically loose" scripts in ./scripts.

The *actual* name, if already "taken in advance", is not the core issue.

Didn't mean to imply the name was an issue, it was the semantics. I do intend to create a type of "misc" subpackage, but it's one I won't support; functionally it's something for internal dev usage on the QEMU tree was my intent.

If you want accel functions to be supported and available outside the QEMU source tree, I am saying that I don't want to support a "util" package to do that.

This would also remove the need to keep importing symbols into the
current "qemu.machine" namespace.

It would, but it's three functions. It doesn't warrant an entire package for
a single module with three standalone functions.

That calculus may change in the future if we accumulate lots more
miscellaneous thingies to make that more worthwhile!

   from avocado.utils import cloudinit
   from avocado.utils import network
diff --git a/tests/qemu-iotests/297 b/tests/qemu-iotests/297
index 5c5420712b..8236875222 100755
--- a/tests/qemu-iotests/297
+++ b/tests/qemu-iotests/297
@@ -36,7 +36,7 @@ MYPYPATH=../../python/ mypy --warn-unused-configs 
--disallow-subclassing-any \
       --disallow-any-generics --disallow-incomplete-defs \
       --disallow-untyped-decorators --no-implicit-optional \
       --warn-redundant-casts --warn-unused-ignores \
-    --no-implicit-reexport iotests.py
+    --no-implicit-reexport --namespace-packages iotests.py
   # success, all done
   echo "*** done"
diff --git a/tests/qemu-iotests/300 b/tests/qemu-iotests/300
index 5b75121b84..a41fac4e1d 100755
--- a/tests/qemu-iotests/300
+++ b/tests/qemu-iotests/300
@@ -23,7 +23,7 @@ import random
   import re
   from typing import Dict, List, Optional, Union
   import iotests
-import qemu
+from qemu.machine import machine

I believe this gives my previous comment some extra strength.  I think
this shows that "qemu.machine" containing "machine" really means that
the first "machine" naming component doesn't really mean that.

Yeah, I don't like the duplication of 'machine' here either. I believe this
test needs access to specific error classes which I didn't believe were
important enough to receive an alias.

Might be that the correct thing to do is to stick machine.py in
machine/__init__.py to avoid that duplication, just like I did with qmp.py.

I agree this is a better approach wrt *machine.py* itself, from the
API perspective.

In a broad sense (excluding qmp) you're proposing that current usage of:

  "from qemu import myutility"


  "from qemu.machine import myutility"

And "everything" (take it with a grain of salt), will live under
"qemu.machine".  I still think, as you put it, that the future will
show that other modules not connected to "machine", but generally
useful, will need a place to live.

I mean, the only things we *have* that aren't directly connected to Machine are these three accel functions, and even those launch QEMU binaries to probe them for features for the purpose of .. passing flags to QEMUMachine.

This may be an opportunity to *keep* a place where contributors can
put generally useful bits.  But, if you want to recognize that current
"from qemu import myutility" was an unintended naming side effect, and
that namespace is really about *machine*, and broaden the meaning of
"machine", OK.

You're right, it is the "removal" of a place where we can put random bits and pieces. That's a conscious decision, yes.

We can create that space (and I intend to, as intuited above), but in so doing I also intend to preclude it from any kind of API stability guarantees, visibility in our sphinx manuals, that sort of thing. I think you actually do want to keep using accel and want to support people using it, so it needs a place to live.

Reintroduction of a "core" package that we intend to support externally could work; but without a strong idea of what will go in there, I'd rather not make one pre-emptively.

I guess put another way:

I'd prefer accel helpers to be in the machine package, since they provide an obvious utility for helping to launch machines (though they may provide additional utility in further contexts), instead of creating a completely un-themed miscellaneous package to house just a few things, especially for callers who are *almost guaranteed* to be using the other subpackages already.

- Cleber.

All that said, I'm not truly married to the exact names I chose. Feel free to recommend some, but this is the best I could come up with. I think "qemu.machine" in particular could have a better name, but what to name it? I don't know.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]