[QGIS-Developer] Bummer... not working geopackage & QGIS

[Nyall Dawson]

On Mon, 15 Jul 2019 at 11:02, Paolo Cavallini <cavallini at faunalia.it> wrote:
>
> Hi Nyall, all,
> so you're implicitly suggesting refactoring this code entirely, right?

Yes, it needs a similar rework as has been done in the past with
symbology, vector iteration, print layouts, processing, layer
tree/legend etc. It's the oldest foundational code still in use in
QGIS today and is overdue for a rethink and rebuild.

> I imagine this would be a major effort. Does anybody have an idea of the order of magnitude of the work involved?

Well, Alessandro and I debated in the past whether or not to submit
this as a grant proposal, but in the end decided that the work
required was too large for available grant funds (and was too painful
to undertake on a partially funded/semi-volunteer basis). There's a
lot involved, because attribute table also includes forms and editor
widgets.

In my view the editor widget code itself is in decent shape, but the
form code has drifted far from its original design and had too many
extra parts bolted on, and now needs a (code) redesign. The attribute
table dialog/view and model need completely reworking, with
performance specifically in mind.

Nyall



> Regards.
>
> On 15 July 2019 01:58:53 CEST, Nyall Dawson <nyall.dawson at gmail.com> wrote:
>>
>> On Fri, 12 Jul 2019 at 00:24, Alessandro Pasotti <apasotti at gmail.com> wrote:
>>>
>>>
>>>
>>>
>>>  On Thu, Jul 11, 2019 at 3:03 PM Paolo Cavallini <cavallini at faunalia.it> wrote:
>>>>
>>>>
>>>>  Hi Richard,
>>>>
>>>>  On 10/07/19 20:03, Richard Duivenvoorde wrote:
>>>>>
>>>>>  Hi Devs,
>>>>>
>>>>>  may I draw ones attention to this (in my view pretty serious) issue:
>>>>>
>>>>>  https://github.com/qgis/QGIS/issues/28914
>>>>>
>>>>>  Showing off QGIS in combination with a geopackage failed today,
>>>>>  because 'it just does *NOT* work' :-)
>>>>>
>>>>>  Wanted to create an issue and then found the above one
>>>>>  (which is a duplicate of #29035, #29166, #29766  etc)
>>>>>  So added a test dataset to the issue.
>>>>>
>>>>>  The 'Move selection on top' in the attribute table is broken for already
>>>>>  some versions of QGIS. Not sure about others, but I often show this to
>>>>>  newbies, to find their selected features.
>>>>>
>>>>>  I think this kind of basic functionality should not be broken for such a
>>>>>  long time in our beloved QGIS is it?
>>>>
>>>>
>>>>  thanks a lot for the report. I'm also surprised, as I assumed this basic
>>>>  functionality was covered by tests. Should we invest more in adding
>>>>  tests in this area?
>>>
>>>
>>>
>>>
>>>  Test on the GUI are very very limited, and they are also quite hard to write and to maintain.
>>>
>>>  If you add that the attribute table code has been crying for a serious refactory since 2.x you know why it's broken (again).
>>>
>>>  But yes, we should invest more in the tests AND in the refactoring of the code, the problem IMO with the attribute table is that it would be a *huge* effort.
>>
>>
>> Also (it's not defence, but...) I seem to recall this particular bug
>> popping up in various forms since forever. Someone will post a
>> (partial?) fix, and then a month later the bug will be bug (maybe in
>> another form?). In any case, I very much doubt this is a 3.x
>> regression, rather just some piece of fragile code which was
>> sticky-taped together in the first place, and that sticky-tape keeps
>> coming undone.
>>
>> Unfortunately (and I hate to give bad news like this), but I'm with
>> Alessandro here. Attribute table is the most fragile code left in
>> modern QGIS versions. It's inefficient, difficult to debug, constantly
>> regressing, and the code is a nightmare mix of interdependent circular
>> signals. The code is honestly horrible to work with, which is another
>> reason I suspect attribute table bugs are "overlooked" during bug
>> fixing sprints...
>>
>> Nyall
>
>
> --
> Sorry for being short