Porting notes / deprecation docs

[Frederik Schwarzer]

Hi,

On 7/10/21 11:54 PM, Friedrich W. H. Kossebau wrote:
> Am Samstag, 10. Juli 2021, 22:47:58 CEST schrieb Frederik Schwarzer:
>> Hi,
>>
>> On 7/10/21 7:38 PM, Friedrich W. H. Kossebau wrote:
>>> Am Samstag, 10. Juli 2021, 18:00:13 CEST schrieb Frederik Schwarzer:
>>>> as mentioned earlier
>>>
>>> Any pointers? :)
>>
>> It was discussed in the weekly BBB meetings a few weeks ago.
> 
> I see. As contributor on occasions only myself, please refer to the respective
> meeting notes some thankfully write, so one can read up on more background,
> and such a planned task ideally would be backed up by a task board entry on
> phabricator, so people can coordinate and track things about it in an async
> manner.

https://mail.kde.org/pipermail/kde-frameworks-devel/2021-June/117653.html
Of course that out-of-context sentence at the end does not represent 
properly what has been said by people then. Some follow-up discussion 
lead to the "just grep it and put it somewhere first" approach.

What I take out of this now is that I need to be more phony about what I 
am planning on doing.


>>>> I would like to document classes/methods/etc that
>>>> are going to be deprecated during KF6 development.
>>>
>>> Can you help confused-on-first-read people by explaining what "deprecated
>>> during KF6 development" is referring to? Deprecated during KF5 development
>>> and no longer be available in KF6? Not yet deprecated due to no existing
>>> replacement, but with replacement planned in KF6?
>>
>> Everything that is marked deprecated when KF6 sees the light of day.
> 
> Okay. Not a good idea IMHO. There should be a single place of information, and
> we have that already with the current KF5 API docs. I hope no-one plans to
> just remove them from the website, though, Well, then there are also the
> offline docs in QCH format as backup generated during the builds and packaged
> by good distributions ;)
> 
>> The idea is to have the APIs that are being deprecated now documented
>> when those APIs (and with it the API docs) are removed.
>> The audience is everyone who is starting the porting work when KF6 is
>> already there for some time.
> 
> Ideally that audience should get the recommendation to first port away from
> deprecated API using the last released version of KF5 and Qt6. That way they
> are able to do a big chunk of the work while being able to maintain a fully
> working state of their software, without serious regressions. Once that
> checkpoint is reached, then go for porting all those things which disappeared/
> changed in KF6 & Qt6 without any preparations in KF5 & Qt5.
> 
> Remember that this is not just KF 5 -> 6, but also Qt 5 -> 6. And perhaps even
> C++11 -> C++17. IMHO only those would recommend to port directly from one set
> of APIs to an other one without any intermediate checkppints for the working
> sate of the software who want to secure their job for a while, because it will
> take ages to fix all the regressions introduced during the port. Unless the
> company/community goes down in the meantime, because the ported software does
> not get done.
> 
> BTW, even the Qt Company recommends that step-by-step approach, and they
> surely do want to have their customers be successful in a short time ;) ->
> https://doc.qt.io/qt-6/portingguide.html
> 
> That is also why some of us invested so much of our time into properly tagging
> API with deprecations warning macros and visibility guards, so porting can be
> done step by step away from the old AP assisted by the compiler, always having
> a working software. Because we have been through some porting in KDE and
> learned our lessons, haven't we... ;)
> 
>> Yes it is manual work. However, since the documentation does not remove
>> stuff that has been removed from the API, it's a thing of adding newer
>> deprecation markers, which seems manageable.
> 
> While perhaps it might be a nice thing to have a shortcut list of API that is
> deprecated in KF5 times, as a manifest to look-up things, ideally we find ways
> to auto-generate that from the existing API markup.
> 
> After all KDE is a software developing community, so we should be able to
> automate that, no? ;)
> 
> So, I can only really ask to keep documentation of KF5's deprecated API in one
> place, and do it properly there, with nice examples, already now useful to
> those who port away when they can. And that place should be the current KF5
> API docs.
> Even more as people come and go, and having yet another place which needs to
> be kept even more manually up-tod-ate does not improve things, it adds more
> risk to have outdated unmaintained information. As you could see in review, it
> already now needs poking in every second review to have proper documentation.
> And then also do that in some separate content?
> 
> What would be very good to have though IMHO, are preparations of the porting
> documentation of that API which is not deprecated in KF5, but will be replaced
> by something else in KF6 (because it cannot be done earlier for reasons). The
> KF6 task board should have some data about such plans.
> Such documentation will need a place and a structure, so also need someone to
> work on and prepare it, so developers can put in the data once those
> replacements are created during KF6 init phase.

Those are all very good points that I wish I had learned about weeks ago. ;)

About automating the extraction process:
Sometimes it is better for a (kind-of) one-shot action to do it manually 
instead of writing a tool that maybe is harder to do than anticipated 
and in the end there is an 80% finished tool that's not suitable for the 
job (yet) and the actual work is not done because everyone is waiting 
for the perfect solution. I suspect, I am not the only one who went down 
that road one time or another. :D

Regarding single source of truth:
There was the argument that the deprecation macros would be removed at 
some point and then 3rd-party developers would be left alone with just 
removed API. In that scenario the extracted information would have 
become that single source of truth.
That argument does not hold, of course, if we (as you suggest) promote 
the porting steps as they are currently suggested for Qt porting. I need 
to dig into that a bit to better understand the process.

I would like to hear more opinions about this, even if they just agree 
with points already made.

Thanks so far.

Cheers,
Frederik