Re: [Qemu-devel] [PATCH 15/30] qmp-commands: move 'query-migrate-parameters' doc to schema

[Markus Armbruster <armbru@redhat.com>]

Marc-André Lureau <mlureau@redhat.com> writes:

> Hi
>
> ----- Original Message -----
>> Marc-André Lureau <mlureau@redhat.com> writes:
>> 
>> > Hi
>> >
>> > ----- Original Message -----
>> >> Marc-André Lureau <mlureau@redhat.com> writes:
>> >> 
>> >> > Hi
>> >> >
>> >> > ----- Original Message -----
>> >> >> On 09/13/2016 08:01 AM, Marc-André Lureau wrote:
>> >> >> > Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
>> >> >> > ---
>> >> >> >  docs/qmp-commands.txt | 29 -----------------------------
>> >> >> >  qapi-schema.json      | 13 +++++++++++++
>> >> >> >  2 files changed, 13 insertions(+), 29 deletions(-)
>> >> >> > 
>> >> >> 
>> >> >> > +++ b/qapi-schema.json
>> >> >> > @@ -1011,6 +1011,19 @@
>> >> >> >  # Returns: @MigrationParameters
>> >> >> >  #
>> >> >> >  # Since: 2.4
>> >> >> > +#
>> >> >> > +# Example:
>> >> >> > +#
>> >> >> > +# -> { "execute": "query-migrate-parameters" }
>> >> >> > +# <- { "return": {
>> >> >> > +#          "decompress-threads": 2,
>> >> >> > +#          "cpu-throttle-increment": 10,
>> >> >> > +#          "compress-threads": 8,
>> >> >> > +#          "compress-level": 1,
>> >> >> > +#          "cpu-throttle-initial": 20
>> >> >> > +#       }
>> >> >> > +#    }
>> >> >> > +#
>> >> >> >  ##
>> >> >> >  { 'command': 'query-migrate-parameters',
>> >> >> >    'returns': 'MigrationParameters' }
>> >> >> 
>> >> >> The example lacks 'cpu-throttle-increment', 'tls-creds', and
>> >> >> 'tls-hostname'; do we want to take this opportunity to touch it up?
>> >> >
>> >> > I suggest to put a [...] in the returned example, as this example could
>> >> > grow again, and there isn't much to learn from that query.
>> >> >  
>> >> >> Meanwhile, I have a series that touches this code, and will obviously
>> >> >> create a merge conflict for whoever gets in second:
>> >> >> https://lists.gnu.org/archive/html/qemu-devel/2016-09/msg01946.html
>> >> >
>> >> > Yes, the more we wait to review the series, the more conflicts we will
>> >> > get.
>> >> > There is still over 100 patches to go, I'll send the next 30.
>> >> 
>> >> We suggested restructuring the series, and you liked the idea with the
>> >> alternative step (3b), not (3a).  Would it make sense to repost the
>> >> beginning of the multi-part monster in that form before moving on to the
>> >> next part?
>> >
>> >  3. Merge qmp-commands.txt into QAPI schema comments, step by step
>> >
>> >    (b) If you delete qmp-commands.txt section as you cover them in the
>> >    QAPI schema, command documentation regresses temporarily.  Tolerable,
>> >    but needs to be explained in commit messages.  Your choice.
>> >
>> > Isn't that what this series is doing? it moves the remaining doc from
>> > qmp-commands.txt to the schema.
>> 
>> Misunderstanding?  Step (3b) is one step of a reordered series.  Let me
>> repeat the order I proposed:
>> 
>> 1. Fix existing issues in QAPI schema comments
>> 
>> 2. Generate documentation from it (not a replacement for
>>    qmp-commands.txt, yet)
>> 
>> 3. Merge qmp-commands.txt into QAPI schema comments, step by step
>> 
>>    (a) If you only update the QAPI schema comments, qmp-commands.txt
>>    stays intact throughout this work.
>> 
>>    (b) If you delete qmp-commands.txt section as you cover them in the
>>    QAPI schema, command documentation regresses temporarily.  Tolerable,
>>    but needs to be explained in commit messages.  Your choice.
>> 
>> 4. Generated documentation now contains everything qmp-commands.txt
>>    contains; delete qmp-commands.txt
>> 
>> This way, the first part contains everything that's really interesting:
>> step 1, 2 and some of 3a or 3b, depending on which alternative you pick.
>> The remaining parts are just more of 3a or 3b, plus the trivial step 4
>> in the last one.
>> 
>> I proposed this to get the interesting review of step 2 out of the way
>> early, and before we tire ourselves out on the not-so-interesting but
>> necessary review of step 3.
>> 
>
> It would have been easier to keep the discussion on the original thread. I disagreed with this plan:
>
>  Generating the documentation before the end of 3(b) will also lead to temporarily incomplete generated doc, and will conflict with existing qmp-commands.txt.

Having incomplete new documentation (getting less incomplete in each
commit) conflict with incomplete old documentation (getting more
incomplete) is no worse than having old documentation getting more
incomplete.  In both cases, we go through an intermediate state with
flawed documentation.

>  That's why I think the best solution is to go through 3(b) now, collect the move in a branch and push it in one go when qmp-commands.txt is empty and the doc is generated.

If you prefer to do it this way, I suggest to post everything at once,
because I won't bother reviewing any of step 3 before step 2 for the
reasons I explained above.  I don't want to see review of the step 2
invalidate all our review work on step 3.