Re: [PATCH] qapi: Fix code generation with Python 3.5

[Markus Armbruster]

John Snow <address@hidden> writes:

> On 1/17/20 2:07 AM, Markus Armbruster wrote:
>> John Snow <address@hidden> writes:
>> 
>>> On 1/16/20 3:25 PM, Markus Armbruster wrote:
>>>> Recent commit 3e7fb5811b "qapi: Fix code generation for empty modules"
>>>> modules" switched QAPISchema.visit() from
>>>>
>>>>     for entity in self._entity_list:
>>>>
>>>> effectively to
>>>>
>>>>     for mod in self._module_dict.values():
>>>>         for entity in mod._entity_list:
>>>>
>>>> Visits in the same order as long as .values() is in insertion order.
>>>> That's the case only for Python 3.6 and later.  Before, it's in some
>>>> arbitrary order, which results in broken generated code.
>>>>
>>>> Fix by making self._module_dict an OrderedDict rather than a dict.
>>>>
>>>> Fixes: 3e7fb5811baab213dcc7149c3aa69442d683c26c
>>>> Signed-off-by: Markus Armbruster <address@hidden>
>>>> ---
>>>>  scripts/qapi/schema.py | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/scripts/qapi/schema.py b/scripts/qapi/schema.py
>>>> index 0bfc5256fb..5100110fa2 100644
>>>> --- a/scripts/qapi/schema.py
>>>> +++ b/scripts/qapi/schema.py
>>>> @@ -795,7 +795,7 @@ class QAPISchema(object):
>>>>          self.docs = parser.docs
>>>>          self._entity_list = []
>>>>          self._entity_dict = {}
>>>> -        self._module_dict = {}
>>>> +        self._module_dict = OrderedDict()
>>>>          self._schema_dir = os.path.dirname(fname)
>>>>          self._make_module(None) # built-ins
>>>>          self._make_module(fname)
>>>>
>>>
>>> This problem has bitten me *many* times. I'm wondering if there's a
>>> prescription that isn't just "Wait until we can stipulate 3.6+".
>> 
>> No clue.
>> 
>> 3.5 EOL is scheduled for 2020-09-13.
>> https://devguide.python.org/#status-of-python-branches
>> 
>> We support 3.5 because we support Debian 9.
>> 
>> We'd normally drop support for Debian 9 two years after Debian 10,
>> i.e. July 2021.  Assuming Debian supports it that far.  Whether they can
>> truly support Python 3.5 after uptstream EOL seems doubtful.
>> 
>
> We should decide whether we consider Debian LTS to be adequately
> supported, yes-or-no.
>
> We should use a rule of "two years after successor, or End-of-Support,
> whichever comes first."

Yes.

> For Debian, is end of support three years after it comes out, or is it
> when the LTS is EOL?

We need to define end-of-support for Debian: is it Debian proper or is
it Debian LTS?

<https://wiki.debian.org/DebianOldStable>:

    Q) How long will security updates be provided?

    The security team tries to support a stable distribution for about
    one year after the next stable distribution has been released,
    except when another stable distribution is released within this
    year.  It is not possible to support three distributions; supporting
    two simultaneously is already difficult enough.

<https://wiki.debian.org/LTS>:

    Debian Long Term Support (LTS) is a project to extend the lifetime
    of all Debian stable releases to (at least) 5 years.  Debian LTS is
    not handled by the Debian security team, but by a separate group of
    volunteers and companies interested in making it a success.

    Thus the Debian LTS team takes over security maintenance of the
    various releases once the Debian Security team stops its work.

Debian 10 "Buster" was released in July 2019.  Debian 9 "Stretch" will
receive security updates from Debian until mid 2020, i.e. just about
when Python 3.5 reaches EOL.  LTS will attempt to support it until June
2022.

I think we should give ourselves a bit more flexibility than the
categorical "Support for the previous major version will be dropped 2
years after the new major version is released."  At some point, the cost
of supporting old hosts exceeds the utility.  We should face this
tradeoff.

> In this specific case, do we trust that Debian 9 LTS will continue to
> patch Python3.5 all the way up until July 2021?

June 2022 even.

We use Python at compile time with trusted input.  The need for security
maintenance is relatively low there.  I'm not ready to vouch for "and we
don't use Python for anything else".