add support for dataclasses #5010

Bogdanp · 2018-05-06T13:22:46Z

This is a work in progress for adding support for dataclasses as a plugin (heavily inspired by and reusing some parts of the attrs plugin). It's pretty close to complete, but I wanted to get some feedback before continuing as I'm new to the codebase. I'll take this over the finish line next weekend if it's deemed desirable, but here are my open questions:

Is the way I'm hooking the plugin in correct? fullname in get_class_decorator_hook doesn't seem to have the value I expect (i.e. the name is qualified to the module dataclass is imported in, so its value is some_package.some_module.dataclass rather than dataclasses.dataclass).
Should I be attempting to re-use more code between the two plugins? I only re-used the things that seemed to be fully generic between the two.

I tried it out on a real-world Python 3.6 project that makes heavy use of dataclasses and it seems to work great so far.

This would close #4792.

euresti · 2018-05-07T17:57:00Z

mypy/plugin.py

                auto_attribs_default=True
            )
+        # TODO: It's unclear if fullname should be a fully qualified


I believe this happens if the module is not in the typeshed.

Thanks! Should I add the module to the typeshed first?

I am fine with this solution. We can check later if this is fixed when the stub is in typeshed.

There is already an open typeshed PR to add a dataclasses stub: python/typeshed#1944. I found some issues in it, but the original author hasn't responded for a while. @Bogdanp if you're interested, you could take over that PR and I can help you get it merged into typeshed.

ilevkivskyi

Thanks for working on this! The PR looks good, I have made a first round of review. Please let me know if you are still interested in working on this.

ilevkivskyi · 2018-05-14T12:46:03Z

mypy/plugin.py

@@ -2,20 +2,19 @@

 from abc import abstractmethod
 from functools import partial
-from typing import Callable, List, Tuple, Optional, NamedTuple, TypeVar, Dict
+from typing import Callable, Dict, List, NamedTuple, Optional, Tuple, TypeVar


Please don't do automatic sort of imports. This only increases the size of the diff and makes it harder to review.

ilevkivskyi · 2018-05-14T12:47:50Z

mypy/plugin.py

-        if fullname in mypy.plugins.attrs.attr_class_makers:
-            return mypy.plugins.attrs.attr_class_maker_callback
-        elif fullname in mypy.plugins.attrs.attr_dataclass_makers:
+        from mypy.plugins import attrs


Why do you need the local import? If there is an import cycle, you should try breaking it. At mypy we try hard to not introduce import cycles, because they complicate the (already complex) code logic.

There had already been a circular dependency that was being resolved by this line. When I extracted mypy.plugins.common the same thing continued to work fine for the test suite, but the dynamically-generated test to ensure import mypy.plugins.common could be imported failed. I think the circular dependency can be fixed by extracting ClassDefContext into a separate module (like mypy.plugin_context). Let me know if you'd like me to do that!

I think the circular dependency can be fixed by extracting ClassDefContext into a separate module (like mypy.plugin_context). Let me know if you'd like me to do that!

I think moving out all the interfaces to a separate file mypy.api is a more permanent solution. But it is a big refactoring, so it is better to do this in a separate PR.

ilevkivskyi · 2018-05-14T14:06:43Z

mypy/plugins/common.py

+from mypy.types import CallableType, Overloaded
+
+
+def _get_decorator_bool_argument(


Do I need to review these three functions or you just copied these from attrs?

Yes, these are the same functions from the attrs module, unchanged.

ilevkivskyi · 2018-05-14T14:33:58Z

test-data/unit/fixtures/dataclasses.pyi

@@ -0,0 +1,27 @@
+# Builtins stub used to support @dataclass tests.


Do you really need a separate fixture? You can try tweaking an existing one.

ilevkivskyi · 2018-05-14T14:35:18Z

mypy/plugins/dataclasses.py

+
+def dataclass_class_maker_callback(ctx: ClassDefContext) -> None:
+    """Hooks into the class typechecking process to add support for dataclasses.
+    """


It is good that you added docstrings to all functions. We like the well documented code 👍

ilevkivskyi · 2018-05-14T15:58:14Z

test-data/unit/check-dataclasses.test

+
+app = Application.parse('')
+
+[builtins fixtures/list.pyi]


It looks like several tests are missing. Could you please add at least these:

Incremental tests, these are tests where you update something in the code and check that (de-)serialisation works correctly (you can look at the initial and subsequent attrs PR to get some ideas).

Test all error messages that you add.

Test all special feature in dataclasses, such as ClassVar and InitVar and special methods (you can look for some examples in the PEP)

Check situations where user defines some dunder methods manually

It looks like you didn't yet added incremental tests. They live in check-incremental.test (you can look for example in attrs PR).

Could you please add some tests for generic dataclasses, in particular type inference (again, you can look for examples in the original attrs PR)

ilevkivskyi · 2018-05-14T16:04:28Z

test-data/unit/lib-stub/dataclasses.pyi

+              frozen: bool = ...) -> Callable[[_C], _C]: ...
+
+
+def field(*,


It is good that you added this to lib-stub. Be sure to make updates here, if python/typeshed#1944 gets updated.

ilevkivskyi · 2018-05-14T16:06:03Z

test-data/unit/check-dataclasses.test

+@dataclass
+class Person:
+    name: str
+    age: int = field(default=0, init=False)


I would add more combinations of different values for init and present/absent defaults.

Have you added these tests?

ilevkivskyi · 2018-05-14T16:14:03Z

test-data/unit/check-dataclasses.test

+
+reveal_type(Person)  # E: Revealed type is 'def (name: builtins.str) -> __main__.Person'
+john = Person('John')
+john.age = 24


I would move this below the next line to check the type is inferred from class, not from this assignment.

ilevkivskyi · 2018-05-14T16:14:59Z

test-data/unit/check-dataclasses.test

+@dataclass
+class Application:
+    name: str = 'Unnamed'
+    rating: int = 0


I would add a test that an error is given if a field without a default comes after the one with a default.

ilevkivskyi · 2018-05-20T05:09:49Z

@Bogdanp are you still interested in working on this? If you don't have time or desire, then we can pick up the PR.

Bogdanp · 2018-05-20T05:19:22Z

@ilevkivskyi yup. I plan on working on it later today.

Bogdanp · 2018-05-20T09:40:42Z

@ilevkivskyi I believe I have addressed all the points you raised except for

the circular import (see this comment) and
adding support for InitVars.

I'm a bit out of my depth when it comes to adding support for InitVars. I could not find how to add a new generic alias type (which I think InitVar counts as). Help would be appreciated here or I would be happy to let someone more experienced implement that part.

JelleZijlstra · 2018-05-20T16:35:10Z

mypy/plugins/dataclasses.py

+            self._freeze(attributes)
+
+        info.metadata['dataclass'] = {
+            'attributes': {attr.name: attr.serialize() for attr in attributes},


This probably should be an OrderedDict in order to preserve order in pre-3.6 Python versions. I think that's what's causing the test failure on 3.4 and 3.5 in Travis.

Ahh, that's right! I'll make the change.

Change made. :)

Thanks! For the future though, it's generally easier on the reviewer to push new commits instead of amending your previous commit and force-pushing as you're doing now. That way, it's easier for the reviewer to see what changes were made in response to the review.

We squash the commits anyway before merging, so it's OK to have a lot of commits in your PR branch.

ilevkivskyi

Thanks for continuing to work on this!

I have several more comments and suggestions. Fox the next time, please don't rebase: I need to see your full commit history, otherwise it makes reviewing hard (especially for large PRs like this one).

ilevkivskyi · 2018-05-23T23:49:15Z

mypy/plugin.py

-        if fullname in mypy.plugins.attrs.attr_class_makers:
-            return mypy.plugins.attrs.attr_class_maker_callback
-        elif fullname in mypy.plugins.attrs.attr_dataclass_makers:
+        from mypy.plugins import attrs


I think the circular dependency can be fixed by extracting ClassDefContext into a separate module (like mypy.plugin_context). Let me know if you'd like me to do that!

I think moving out all the interfaces to a separate file mypy.api is a more permanent solution. But it is a big refactoring, so it is better to do this in a separate PR.

ilevkivskyi · 2018-05-23T23:50:32Z

mypy/plugin.py

                auto_attribs_default=True
            )
+        # TODO: Drop the or clause once dataclasses lands in typeshed.


I think you can actually overtake the typeshed PR python/typeshed#1944, the original author seems to not have time to finish it.

Sorry, but I don't think I'll have the time to carry that through at the moment.

ilevkivskyi · 2018-05-23T23:51:07Z

mypy/plugins/dataclasses.py

+from mypy.types import CallableType, NoneTyp, Type, TypeVarDef, TypeVarType
+from mypy.typevars import fill_typevars
+
+#: The set of decorators that generate dataclasses.


Why a colon after #?

Habit. :)

Sphinx picks these up as docstrings for the module vars, but I'll remove this one.

ilevkivskyi · 2018-05-23T23:52:52Z

mypy/plugins/dataclasses.py

+    def __init__(self, ctx: ClassDefContext) -> None:
+        self._ctx = ctx
+
+    def transform(self) -> None:


Add a docstring for this function.

ilevkivskyi · 2018-05-23T23:55:43Z

mypy/plugins/dataclasses.py

+                args=[attr.to_argument(info) for attr in attributes if attr.is_in_init],
+                return_type=NoneTyp(),
+            )
+            for stmt in self._ctx.cls.defs.body:


Why do you need this block? Add a comment about this.

ilevkivskyi · 2018-05-24T00:23:13Z

test-data/unit/check-dataclasses.test

+class A:
+    a: int
+
+


Please use at most one empty line in tests (even between tests).

ilevkivskyi · 2018-05-24T00:26:26Z

test-data/unit/check-dataclasses.test

+
+reveal_type(SpecializedApplication)  # E: Revealed type is 'def (id: Union[builtins.int, None], name: builtins.str, rating: builtins.int =) -> __main__.SpecializedApplication'
+
+[builtins fixtures/list.pyi]


Newline is missing at the end of file.

ilevkivskyi · 2018-05-24T00:32:57Z

test-data/unit/check-dataclasses.test

+
+app = Application.parse('')
+
+[builtins fixtures/list.pyi]


It looks like you didn't yet added incremental tests. They live in check-incremental.test (you can look for example in attrs PR).

ilevkivskyi · 2018-05-24T00:34:14Z

test-data/unit/check-dataclasses.test

+
+app = Application.parse('')
+
+[builtins fixtures/list.pyi]


Could you please add some tests for generic dataclasses, in particular type inference (again, you can look for examples in the original attrs PR)

ilevkivskyi · 2018-05-24T00:37:26Z

test-data/unit/lib-stub/dataclasses.pyi

+
+
+@overload
+def dataclass(_cls: _C,


You also need to add InitVar (here and on typeshed). You can define it as a normal generic class:

class InitVar(Generic[_T]): pass

Then in the plugin you can test for it e.g. like this if isinstance(some_type, Instance) and some_type.info.fullname() == 'dataclasses.InitVar'.

I had tried that and couldn't get it to work, but I must've messed something up last time because it works just fine now.

Bogdanp · 2018-05-27T11:55:34Z

@ilevkivskyi do you have any advice with regards to how to best remove InitVar attributes from a class? I've tried removing them from the SymbolTable, but that breaks inheritance. The best I could come up with was to remove them from the class' symbol table and add the types to the DataclassAttributes but then I wasn't sure how to best serialize/deserialize the types.

ilevkivskyi

Thanks! This looks almost ready, but there are few more comments.

Most importantly it looks like you still didn't add incremental tests. They live in check-incremental.test (you can look for examples in attrs tests). Please add at least these tests:

Basic functionality (creating boilerplate methods) still works after loading from cache
Base classes work after loading from cache (the base class and main class should be in different files).
Forward references work correctly. For example:

@dataclass
class C:
    x: 'Other'
class Other:
    ...

IIRC all these kinds of tests should be there for attrs.

ilevkivskyi · 2018-05-28T02:04:28Z

mypy/plugins/dataclasses.py

+            'has_default': self.has_default,
+            'line': self.line,
+            'column': self.column,
+        }


The best I could come up with was to remove them from the class' symbol table and add the types to the DataclassAttributes but then I wasn't sure how to best serialize/deserialize the types.

You can add list of InitVars here. Most types in mypy support (de-)serialization (those that are not supported like partial types, or forward reference types, should not appear in at the point where serialization happens). You can find signatures in mypy.types.Type.serialize and mypy.types.Type.deserialize.

I had tried that, but deserialization failed with De-serialization failure: TypeInfo not fixed. Reading the docstring for FakeInfo, I understand this is supposed to be done in two passes, but it looks like the fixup step requires a lot of information I don't necessarily have/that it would be ugly to pass around. Am I missing something obvious?

ilevkivskyi · 2018-05-28T02:16:33Z

mypy/plugins/dataclasses.py

+        info = self._ctx.cls.info
+        for attr in attributes:
+            try:
+                node = info.names[attr.name].node


I think this will look cleaner if refactored via info.names.get(attr.name) which cannot raise.

ilevkivskyi · 2018-05-28T02:17:37Z

setup.cfg

@@ -48,3 +48,6 @@ parallel = true

 [coverage:report]
 show_missing = true
+
+[isort]
+multi_line_output = 5


Why do you want this addition? Can it be made in a separate PR?

That was accidental, sorry! I'll remove it.

ilevkivskyi · 2018-05-28T02:19:06Z

mypy/plugins/dataclasses.py

+            try:
+                # parse_bool returns an optional bool, so we corece it
+                # to a bool here in order to appease the type checker.
+                is_in_init = bool(ctx.api.parse_bool(field_args['init']))


This will be definitely cleaner if refactored via filed_args.get('init') that can't raise. (Also the comment/coercion will be unnecessary after refactoring.)

I may be missing something but I don't think that's true. field_args is a dict from str -> Expression so I'll still need the parse_bool call, which means I'll also need the bool coercion since parse_bool returns an Optional[bool]. I'll refactor to drop the try-catch though (although I do think this approach is cleaner)!

ilevkivskyi · 2018-05-28T02:20:29Z

mypy/plugins/dataclasses.py

+
+            # Treat the assignment as an instance-level assignment
+            # even though it looks like a class-level assignment.
+            node.is_initialized_in_class = False


Could you please add a test that fails without this change? Or if you already added, could you please tell me the name?

It appears that wasn't necessary after all so I removed it.

ilevkivskyi · 2018-05-28T02:26:08Z

test-data/unit/check-dataclasses.test

+  def problem(self) -> T:
+    return self.z  # E: Incompatible return value type (got "List[T]", expected "T")
+
+reveal_type(A)  # E: Revealed type is 'def [T] (x: T`1, y: T`1, z: builtins.list[T`1]) -> __main__.A[T`1]'


This check is not very useful, I would rather called it with incompatible args to see an error, e.g. A('no', 'way', [1, 2]).

ilevkivskyi · 2018-05-28T02:27:09Z

test-data/unit/check-dataclasses.test

+reveal_type(a)  # E: Revealed type is '__main__.A[builtins.int*]'
+reveal_type(a.x)  # E: Revealed type is 'builtins.int*'
+reveal_type(a.y)  # E: Revealed type is 'builtins.int*'
+reveal_type(a.z)  # E: Revealed type is 'builtins.list[builtins.int*]'


I would add a test that a.foo() returns an int.

ilevkivskyi · 2018-05-28T02:27:34Z

test-data/unit/check-dataclasses.test

+
+[builtins fixtures/list.pyi]
+
+[case testDataclassGenerics]


Thanks for adding some tests!

Bogdanp · 2018-05-28T08:16:38Z

Thanks for the review @ilevkivskyi. I believe I've made all the requested changes apart from InitVar support which I'll try to tackle again towards the end of the week.

ilevkivskyi

I'll try to tackle again towards the end of the week.

We are close to the end of the week, so here is another (hopefully last) round of review.

Note that we will still need to add function plugin hooks to make asdict, astuple, replace etc. more precise (they can't be typed precisely in typeshed) in a separate PR. Are you interested in working on the second PR?

ilevkivskyi · 2018-06-02T21:10:31Z

mypy/plugin.py

                auto_attribs_default=True
            )
+        # TODO: Drop the or clause once dataclasses lands in typeshed.


The typeshed PR is landed. Is this still needed?

Looks like it still is, not sure why, but after updating the typeshed submodule to the current master, if I remove the or clause and run mypy against my project, the issue still happens.

ilevkivskyi · 2018-06-02T21:18:37Z

mypy/plugins/dataclasses.py

+            'has_default': self.has_default,
+            'line': self.line,
+            'column': self.column,
+        }


Am I missing something obvious?

No, you are right, this way it will not work. Here is another idea about InitVars that I think should work. You can just store names of InitVars in a special filed in .metadata, then later you can find the types in the type of __init__ method, since in mypy both arg types and and arg names are stored. This is a bit hacky, but should be robust and is much simpler than manually traversing types in fixup phase.

ilevkivskyi · 2018-06-02T21:21:41Z

test-data/unit/lib-stub/dataclasses.pyi

+              frozen: bool = ...) -> Callable[[_C], _C]: ...
+
+
+def field(*,


It looks like current definition in typeshed is a bit different, could you please sync this one?

This definition diverges from the one in typeshed in that it returns _T rather than Field[_T]. Returning Field[_T] breaks things like:

@dataclass class Foo: x: int = field(default=42)

Because int and Field[int] are incompatible.

ilevkivskyi · 2018-06-02T21:26:50Z

test-data/unit/check-incremental.test

+
+[builtins fixtures/list.pyi]
+[out1]
+[out2]


I think this test is not testing anything. If all files are unchanged like here, then mypy will at most load caches, but will not make any checks. What you should do is to put first part of the test in another module (e.g. [file b.py]). And then on the second run (i.e. in [file b.py.2]) just make a cosmetic change, like comment somewhere, so that mypy rechecks b.py while using cache for (unchanged) a.py.

ilevkivskyi · 2018-06-02T21:28:15Z

test-data/unit/check-incremental.test

+
+[builtins fixtures/list.pyi]
+[out1]
+main:7: error: Revealed type is 'def (x: builtins.int) -> __main__.B'


Any error (including a reveal) prevents writing the cache, so you should use reveal_type only on second run.

ilevkivskyi · 2018-06-02T21:31:25Z

test-data/unit/check-incremental.test

+
+[case testIncrementalDataclassesDunder]
+from a import A
+reveal_type(A)  # E: Revealed type is 'def (a: builtins.int) -> a.A'


Also here, please don't add errors on the first run.

ilevkivskyi · 2018-06-02T21:39:10Z

test-data/unit/check-incremental.test

+
+[builtins fixtures/list.pyi]
+[out1]
+[out2]


I like this test case, simple and to the point.

ilevkivskyi · 2018-06-02T21:40:29Z

test-data/unit/check-incremental.test

+
+[builtins fixtures/list.pyi]
+[out1]
+main:2: error: Argument 2 to "B" has incompatible type "str"; expected "int"


I like this one too. Although it has an error on first run, we should keep it, just to be sure, the error is cleared when error is fixed.

ilevkivskyi · 2018-06-02T21:41:22Z

test-data/unit/check-incremental.test

+
+[builtins fixtures/list.pyi]
+[out1]
+[out2]


I would introduce a simple change in one of a or b on the second run.

ilevkivskyi · 2018-06-02T21:43:14Z

test-data/unit/check-dataclasses.test

+A(b=B(42))
+A(b=42)  # E: Argument "b" to "A" has incompatible type "int"; expected "B"
+
+[builtins fixtures/list.pyi]


No newline at the end of file. We really should make this a lint error :-)

This reverts commit 8d2dd6d.

Bogdanp · 2018-06-04T17:41:27Z

Note that we will still need to add function plugin hooks to make asdict, astuple, replace etc. more precise (they can't be typed precisely in typeshed) in a separate PR. Are you interested in working on the second PR?

I don't think I'll have the free time to tackle that. In fact, I won't have time to work on this PR either in the next two or three weeks. Feel free to take over if you feel there are any other changes that need to be made.

ilevkivskyi

OK, I added some polish, this is now ready. I will merge once tests pass. Thanks for all your work!

I will continue working on this in subsequent PRs.

ilevkivskyi self-assigned this May 7, 2018

euresti reviewed May 7, 2018

View reviewed changes

ilevkivskyi reviewed May 14, 2018

View reviewed changes

Bogdanp changed the title ~~[wip] add basic support for dataclasses~~ add support for dataclasses May 20, 2018

JelleZijlstra reviewed May 20, 2018

View reviewed changes

add basic support for dataclasses

92b5347

ilevkivskyi reviewed May 24, 2018

View reviewed changes

Bogdanp added 6 commits May 27, 2018 13:32

clean up code comments

98becd6

avoid reusing types for comparison methods

3979713

drop extra newlines from tests

4ab42a1

add tests for multiple init arg combos

58ae577

add InitVar to lib stub

7f7415e

add generic dataclass test

addde6e

ilevkivskyi reviewed May 28, 2018

View reviewed changes

Bogdanp added 6 commits May 28, 2018 10:01

drop is_initialized_in_class block since it serves no purpose

3004fac

add more checks to generic dataclass test

f5f52a8

refactor KeyError blocks

74436ea

add incremental tests for dataclasses

2918ebd

fix failing type check

7c857cb

remove accidental modification to setup.cfg

37eaae8

Merge branch 'master' into dataclasses

afbeb78

ilevkivskyi reviewed Jun 2, 2018

View reviewed changes

Bogdanp added 3 commits June 4, 2018 19:12

drop now-unnecessary "or" clause on dataclass plugin init

8d2dd6d

implement InitVar support

9d08b7a

update dataclasses.field

8b20ceb

Bogdanp added 3 commits June 4, 2018 20:13

update typeshed

5b78ad7

Revert "drop now-unnecessary "or" clause on dataclass plugin init"

3b9c215

This reverts commit 8d2dd6d.

update incremental tests

55dfc7e

Some polish

988c6e8

ilevkivskyi mentioned this pull request Jun 5, 2018

Sync typeshed #5151

Merged

Ivan Levkivskyi added 2 commits June 5, 2018 17:20

Merge remote-tracking branch 'upstream/master' into add-dataclasses

f007a78

Fix newline

747473e

ilevkivskyi approved these changes Jun 5, 2018

View reviewed changes

ilevkivskyi merged commit df828b0 into python:master Jun 5, 2018

saulshanabrook mentioned this pull request Feb 21, 2019

Typescript Support foam-framework/foam2#1921

Open

Bogdanp mentioned this pull request Apr 5, 2020

Type Hinting? Bogdanp/dramatiq#302

Closed

		from mypy.types import CallableType, Overloaded


		def _get_decorator_bool_argument(

		@@ -0,0 +1,27 @@
		# Builtins stub used to support @dataclass tests.


		reveal_type(SpecializedApplication) # E: Revealed type is 'def (id: Union[builtins.int, None], name: builtins.str, rating: builtins.int =) -> __main__.SpecializedApplication'

		[builtins fixtures/list.pyi]

add support for dataclasses #5010

add support for dataclasses #5010

Conversation

Bogdanp commented May 6, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ilevkivskyi commented May 20, 2018

Bogdanp commented May 20, 2018

Bogdanp commented May 20, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Bogdanp commented May 27, 2018

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Bogdanp commented May 28, 2018

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Bogdanp commented Jun 4, 2018

Bogdanp commented May 6, 2018 •

edited

Loading

Bogdanp commented May 20, 2018 •

edited

Loading