Add ability for ``__post_init__`` method to be defined. #111

tbeadle · 2016-11-11T04:20:11Z

If this method is defined for a class, it will get executed at the end
of __init__. Previously, there was no way to extend what was
done during __init__ when using @attr.s.

hynek · 2016-11-11T08:16:52Z

src/attr/_make.py

@@ -680,6 +680,10 @@ def fmt_setter_with_converter(attr_name, value_var):
                                                    a.name))
        names_for_globals[val_name] = a.validator
        names_for_globals[attr_name] = a
+    lines.extend([
+        "func = getattr(self, '__post_init__', None)",


This is approach means that attrs gets universally slower for everyone. attrs is explicitly static in its approach. We need to check for the existence of the method when assembling the class and call it unconditionally if it’s present.

codecov-io · 2016-11-11T12:29:47Z

Current coverage is 100% (diff: 100%)

Merging #111 into master will not change coverage

@@           master   #111   diff @@
====================================
  Files           8      8          
  Lines         466    468     +2   
  Methods         0      0          
  Messages        0      0          
  Branches      105    106     +1   
====================================
+ Hits          466    468     +2   
  Misses          0      0          
  Partials        0      0

Powered by Codecov. Last update cbfb229...7724d31

hynek · 2016-11-11T12:51:09Z

src/attr/_make.py

+    script, globs = _attrs_to_script(
+        attrs,
+        frozen,
+        post_init=hasattr(cls, '__post_init__')


No hasattr ever please. This lends itself to getattr(cls, "__post_init__", False).

hynek

I have added a bunch of comments many of which arguably should have been covered by the contribution guide so I’ve added it to my own todo list to fix that.

If you have problems with Hypothesis, offering candy to @Tinche always helped for me. ;)

Thank you for your contribution! And please remember to ping the ticket once you’re done.

hynek · 2016-11-13T08:06:15Z

src/attr/_make.py

@@ -540,7 +544,7 @@ def validate(inst):
            a.validator(inst, a, getattr(inst, a.name))


-def _attrs_to_script(attrs, frozen):
+def _attrs_to_script(attrs, frozen, post_init=False):


Any reason why this has a default value? I find them indicative of problems if used outside public APIs.

hynek · 2016-11-13T08:06:15Z

src/attr/_make.py

@@ -680,6 +684,8 @@ def fmt_setter_with_converter(attr_name, value_var):
                                                    a.name))
        names_for_globals[val_name] = a.validator
        names_for_globals[attr_name] = a
+    if post_init:
+        lines.append("self.__post_init__()")


My guts tell me the post init should be called before validating the object. But I’d love input from people who actually care about this feature. Summoning all the bikeshed artists from #68: @glyph @Insoleet @wearpants @Tinche @wbolster @mathieulongtin @ambv

Speak now or forever hold your peace.

I'd say post validation. I'd expect __post_init__ to be the very last thing init calls. Beside, we can call attr.validate.

I'm voting post validation/conversion. The name __post_init__ itself suggests it's called after the generated __init__, so I'd expect fields to be validated, converted and set. Having it called post-conversion lowers the amount of possible states we can expect the object to be in.

So be it.

We’ll have to add the pass-self-to-factories thing then because otherwise there’s no way to validate dependent attributes.

The other way to validate dependent attributes would be for __post_init__ to raise an exception, no?

Sure but I like the idea of separating the factory process from the validation process.

hynek · 2016-11-13T08:06:15Z

tests/test_make.py

@@ -302,6 +302,22 @@ class D(object):
        assert C.D.__name__ == "D"
        assert C.D.__qualname__ == C.__qualname__ + ".D"

+    def test_post_init(self):


While this test is fine as it is, we make sure that new features work well with the rest of attrs by making them part of our Hypothesis testing strategies: https://github.com/hynek/attrs/blob/master/tests/utils.py#L148-L171

And while squinting at this code I also realized that we should add post_init support to make_class so we have API parity! It should take a callable for argument and attach it as a method to the created class before applying attrs magic to it in https://github.com/hynek/attrs/blob/master/src/attr/_make.py#L816

make_class already supports it. A __post_init__ key can be defined with the dictionary passed in to make_class:

>>> import attr >>> def foo(self): ... print('in foo') ... >>> C1 = attr.make_class('C1', {'x': attr.ib(default=42), '__post_init__': foo}) >>> c = C1() in foo

Fair enough, please just add it to our testing strategy then.

oh wait you did never mind :) I’ll have another deeper look on the weekend

hynek · 2016-11-13T08:06:15Z

docs/examples.rst

@@ -551,6 +551,25 @@ You can still have power over the attributes if you pass a dictionary of name: `
   >>> i.y
   []

+Sometimes, you want to have your class's ``__init__`` method do more than just


Semantic new lines pls.

hynek · 2016-11-13T08:06:15Z

CHANGELOG.rst

@@ -13,6 +13,9 @@ Changes:

 - Don't overwrite ``__name__`` with ``__qualname__`` for ``attr.s(slots=True)`` classes.
  `#99 <https://github.com/hynek/attrs/issues/99>`_
+- Allow for a ``__post_init__`` method that, if defined, will get executed at


We use semantic new lines in prose. Simply make each sentence a new line please.

Also please add a link to this PR like in the entry before.

hynek · 2016-11-13T08:06:15Z

tests/test_make.py

+                self2.z = self2.x + self2.y
+
+        c = C(x=10, y=20)
+        assert hasattr(c, 'z')


hynek · 2016-11-13T08:06:15Z

tests/test_make.py

+
+        c = C(x=10, y=20)
+        assert hasattr(c, 'z')
+        assert c.z == 30


expected == actual please

If this method is defined for a class, it will get executed at the end of ``__init__``. Previously, there was no way to extend what was done during ``__init__`` when using ``@attr.s``.

wearpants · 2016-11-15T11:43:33Z

Late to the party, but yes absolutely post conversion

hynek · 2016-11-20T13:49:28Z

I’ve fixed it up so we don’t have ping-pong over minutiae anymore, thank you very much for your contribution!

I’ve decided to rename __post_init__ to __attrs_post_init__ for consistency with __attrs_attrs__ and to be explicit about it’s heritage.

If someone has good arguments against it, feel free to open a new issue before 16.3.0 is out.

hynek reviewed Nov 11, 2016

View changes

tbeadle force-pushed the tbeadle:post_init branch from 6106e0d to 1ac1cdd Nov 11, 2016

hynek reviewed Nov 11, 2016

View changes

src/attr/_make.py Outdated

script, globs = _attrs_to_script(

attrs,

frozen,

post_init=hasattr(cls, '__post_init__')

This comment has been minimized.

Sign in to view

hynek Nov 11, 2016
Member

No hasattr ever please. This lends itself to getattr(cls, "__post_init__", False).

tbeadle force-pushed the tbeadle:post_init branch from 1ac1cdd to 0f6636b Nov 11, 2016

hynek requested changes Nov 13, 2016

View changes

Add ability for ``__post_init__`` method to be defined.

Loading status checks…

7724d31

If this method is defined for a class, it will get executed at the end of ``__init__``. Previously, there was no way to extend what was done during ``__init__`` when using ``@attr.s``.

tbeadle force-pushed the tbeadle:post_init branch from 0f6636b to 7724d31 Nov 14, 2016

hynek added this to the 16.3.0 milestone Nov 19, 2016

hynek closed this in 1fbaa20 Nov 20, 2016

hynek mentioned this pull request Nov 20, 2016

init hooks #68

Closed

wsanchez mentioned this pull request Feb 17, 2017

frozen class is made immutable before __attrs_post_init__ gets a turn #147

Closed

python-attrs / attrs

Add ability for ``__post_init__`` method to be defined. #111

Add ability for ``__post_init__`` method to be defined. #111

tbeadle commented Nov 11, 2016

This comment has been minimized.

codecov-io commented Nov 11, 2016 •

edited

This comment has been minimized.

hynek left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

wearpants commented Nov 15, 2016

hynek commented Nov 20, 2016

python-attrs / attrs

Sponsor python-attrs/attrs

Add ability for ``__post_init__`` method to be defined. #111

Add ability for ``__post_init__`` method to be defined. #111

Conversation

tbeadle commented Nov 11, 2016

This comment has been minimized.

hynek Nov 11, 2016 Member

codecov-io commented Nov 11, 2016 • edited

Current coverage is 100% (diff: 100%)

This comment has been minimized.

hynek Nov 11, 2016 Member

hynek left a comment

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

mathieulongtin Nov 13, 2016

This comment has been minimized.

Tinche Nov 13, 2016 Member

This comment has been minimized.

hynek Nov 14, 2016 Member

This comment has been minimized.

glyph Nov 14, 2016 Contributor

This comment has been minimized.

hynek Nov 14, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

tbeadle Nov 14, 2016 Author Contributor

This comment has been minimized.

hynek Nov 17, 2016 Member

This comment has been minimized.

hynek Nov 17, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

This comment has been minimized.

hynek Nov 13, 2016 Member

wearpants commented Nov 15, 2016

hynek commented Nov 20, 2016

hynek Nov 11, 2016
Member

codecov-io commented Nov 11, 2016 •

edited

hynek Nov 11, 2016
Member

hynek Nov 13, 2016
Member

hynek Nov 13, 2016
Member

Tinche Nov 13, 2016
Member

hynek Nov 14, 2016
Member

glyph Nov 14, 2016
Contributor

hynek Nov 14, 2016
Member

hynek Nov 13, 2016
Member

tbeadle Nov 14, 2016
Author Contributor

hynek Nov 17, 2016
Member

hynek Nov 17, 2016
Member

hynek Nov 13, 2016
Member

hynek Nov 13, 2016
Member

hynek Nov 13, 2016
Member

hynek Nov 13, 2016
Member