Saturday, December 9, 2017

Lino no longer uses UNION querysets

I found the explanation and a solution for #2190 (IntracomPurchases is empty on test client but not on devserver).

IntracomSales and IntracomPurchases are currently the only tables with an abstract model. For such tables I was using Django’s UNION feature. The problem came because of Django ticket #28281. Django’s UNION feature is a pseudo feature because count() doesn’t work and because it doesn’t work on mysql.

So now I don’t use that feature anymore, and lino.core.dbtables.Table.get_request_queryset() now returns an iteration over the different models which implement the abstract model. The disadvantage is of course that they are ordered only within their model. But that’s not a problem for now, since IntracomSales and IntracomPurchases de facto are usually in a single model.

Changed the API for defining merge actions

Until now each application had to explicitly specify the models that would have a merge action. Now it is the opposite: every model has a merge action unless the application developer sets lino.core.model.Model.allow_merge_action to False.

TODO: adapt welfare test suite. adapt documentation. should Lino automatically set it to False on models with MTI children?

One problem is that now we have dozens of new dialog actions in Welfare or Tera, one for each model.

Lino Book passing on Travis

The last build failure on Travis CI was because building the The Lino Book requires graphviz to be installed on the system:

sudo apt-get install -y graphviz

I now added this to the .travis.yml file.

And yess! Build #557 passed. It is the first one since #468 3 months ago. And at that moment the build didn’t include inv bd. So actually it is the first time in the history of Lino that the full test suite passes.

Nothing is perfect, and there is at least one problem: the coverage is reported to be only 7%. There is obviously something missing. I guess that it has to do with the warnings at line 6006: warning: Module lino was never imported. (module-not-imported)

At line 3679 it says:

Running `which invoke` prep test clean --batch bd in book within coverage...

After looking at the atelier.invlib.tasks source and the Coverage docs), I tried a few things:

  • Maybe the explanation is that I should not call coverage combine when I have only one coverage run? –> No, coverage combine seems to be necessary. Without it, we get an error “no data to combine”.

  • Maybe I should use --parallel-mode? –> No, at least not immediately.

As a summary: The problem is that we launch a single Python process under coverage:

`which invoke` prep test clean --batch bd

And that process runs subprocesses. And it seems that the collected coverage data is getting lost or overwritten.

Also the Lino book continues to be too big for