Thursday, May 2, 2019

I reviewed end-user documentation (Besichtigungstour) and translations for Lino Presto.

Okay there are now 24 failures in the book test suite. Quite a few of them are caused by my changes. Hamza, I am glad if you can help me to do the dirty work of reviewing them, but be careful: don’t change the expected output of a doctest just to get rid of the failure ;-) You should adapt a failing doctest to make it pass only when you are sure you understand why the output has changed. Don’t hesitate to ask.

Lino (lino.mixins.ref.StructuredReferrable) now handles topics with an empty reference more elegantly. Use case lino_xl.lib.topics.Topic with empty ref field.

Side effect when sorting lazy choicelists

There is an interesting failure in docs/specs/ledger.rst of Lino Welfare:

File "docs/specs/ledger.rst", line 268, in ledger.rst
Failed example:
    rt.show(Journals, column_names="ref name voucher_type journal_group")
Expected:
    ========== ====================== ================================================ ====================
     Referenz   Bezeichnung            Belegart                                         Journalgruppe
    ---------- ---------------------- ------------------------------------------------ --------------------
     REG        Rechnungseingänge      Project invoices                                 Rechnungseingänge
     SREG       Sammelrechnungen       Rechnungen                                       Rechnungseingänge
     AAW        Ausgabeanweisungen     Ausgabeanweisungen                               Ausgabeanweisungen
     ZKBC       KBC Zahlungsaufträge   Zahlungsauftrag (finan.PaymentOrdersByJournal)   Zahlungsaufträge
    ========== ====================== ================================================ ====================
    <BLANKLINE>
Got:
    ========== ====================== ==================== ====================
     Referenz   Bezeichnung            Belegart             Journalgruppe
    ---------- ---------------------- -------------------- --------------------
     REG        Rechnungseingänge      Project invoices     Rechnungseingänge
     SREG       Sammelrechnungen       Rechnungen           Rechnungseingänge
     AAW        Ausgabeanweisungen     Ausgabeanweisungen   Ausgabeanweisungen
     ZKBC       KBC Zahlungsaufträge   Ausgabeanweisungen   Zahlungsaufträge
    ========== ====================== ==================== ====================
    <BLANKLINE>

Note: “Ausgabeanweisungen” is the German translation for “Disbursement orders”.

So the ZKBC journal has the wrong voucher type. Here is the code that creates this journal:

kw = dict()
kw.update(journal_group=JournalGroups.zau)
kw.update(dd.str2kw('name', _("KBC Payment Orders")))
kw.update(account=a5800, ref="ZKBC")
kw.update(dc=DEBIT)
jnl = PaymentOrder.create_journal(**kw)
yield jnl
yield MatchRule(journal=jnl, account=a4450)

This code looks perfect, and I cannot imagine which part of it would have introduced our error.

It took me some digging to find out that it has to do with my recent changes for sorting some choicelists in ledger:

def post_site_startup(self, site):
    super(Plugin, self).post_site_startup(site)
    site.models.ledger.CommonAccounts.sort()
    site.models.ledger.VoucherTypes.sort()

The problem disappears when I deactivate this method. It obviously also has to do with the fact that we use add_item_lazy

I was tempted to say “Let’s simply not sort them, if that causes problems”.

But –an example of where it is good to be obstinate– then I discovered the real cause. I mean the cause that caused sorting them to be a problem:

The add_item_lazy method creates a “temporary” function which it connects to the pre_analyze signal. And then it was using a class attribute _lazy_items to store the local function object somewhere. I had observed that “we must store the func somewhere because receiver only connects it to the signal, which is a weak reference.” But this hack was causing side effects now. The correct solution was to avoid that hack by using weak=False for the receiver. And imagine: this subtlety is even explained in the docs for Signal.connect. I just didn’t know it when I wrote add_item_lazy.

Who must pay for such work? The problem was “activated” when introducing the new feature for sorting a choicelist. Which was needed for Lino Presto where the orders plugin adds a new journal group “Orders”, which had to come first because otherwise their main menu wasn’t correctly sorted. Adding this new feature gave me the “good” idea to also use it for VoucherTypes and CommonAccounts in the ledger plugin. If I would manage to explain all this to the customer (which is already utopic), it is utopic to believe that they would accept to pay for this work. An example of work nobody wants to pay for but everybody wants it to be done. That’s why we have #881.

(Edit: above problem reappeared later (Monday, May 6, 2019), so it got a ticket on its own: #2987.)

Another case for #881 was a failure in docs/specs/voga/voga.rst. This is probably caused by the changes in invoicing (new concept of invoicing areas). There were no more sales invoices being generated. Because there was no invoicing area.

I do create a default set of invoice areas in the new demo fixture for lino_xl.lib.invoicing, but Lino Voga and Lino Tera did not yet define a pseudo-fixture which imports it. I reviewed the Plugin inheritance page which explains this mechanism.

TIL : when you create a new fixture in a plugin, then those who inherit your plugin and who just want to inherit your fixtures as well will now automatically get notified that you added a new fixture (for which they must create a wrapper if they want it).

I tried to avoid this pitfall by using an __all__ in the parent’s fixtures main package. Nope, this cannot work because Django discovers fixtures as files, not as Python modules. Django isn’t even aware that we use their fixtures for such a cool things as Python fixtures.

The failures in docs/specs/reception/index.rst and docs/specs/cal.rst (of Lino Welfare) were caused because states are now sorted, which causes the demo calendar entries to get generated a bit differently. So here it was okay to simply change the expected output.