ENH: Convert read_gbq() function to use google-cloud-python #25

jasonqng · 2017-04-08T02:20:04Z

Description

I've rewritten the current read_gbq() function using google-cloud-python, which handles the naming of structs and arrays out of the box. For more discussion about this, see: #23.

However, because of the fact that google-cloud-python potentially uses different authentication flows and may break existing behavior, I've left the existing read_gbq() function and and named this new function from_gbq(). If in the future we are able to reconcile the authentication flows and/or decide to deprecate flows that are not supported in google-cloud-python, we can rename this to read_gbq().

UPDATE: As requested in comment by @jreback (https://github.com/pydata/pandas-gbq/pull/25/files/a763cf071813c836b7e00ae40ccf14e93e8fd72b#r110518161), I deleted old read_gbq() and named my new function read_gbq(), deleting all legacy functions and code.

Added in a few lines to requirements file, but I'll leave it to you @jreback to deal with conda dependency issues which you mentioned in Issue 23.

Let know if any questions or if any tests need to be written. You can confirm that it works by running the following:

q = """
select ROW_NUMBER() over () row_num, struct(a,b) col, c, d, c*d c_times_d, e
from
(select * from
    (SELECT 1 a, 2 b, null c, 0 d, 100 e)
    UNION ALL
    (SELECT 5 a, 6 b, 0 c, null d, 200 e)
    UNION ALL
    (SELECT 8 a, 9 b, 10.0 c, 10 d, 300 e)
)
"""
df = gbq.read_gbq(q, dialect='standard')
df

row_num	col	c	d	c_times_d	e
2	{u'a': 5, u'b': 6}	0.0	NaN	NaN	200
1	{u'a': 1, u'b': 2}	NaN	0.0	NaN	100
3	{u'a': 8, u'b': 9}	10.0	10.0	100.0	300

q = """
select array_agg(a) mylist
from
(select "1" a UNION ALL select "2" a)
"""
df = gbq.read_gbq(q, dialect='standard')
df

mylist
[1, 2]

q = """
select array_agg(struct(a,b)) col, f
from
(select * from
    (SELECT 1 a, 2 b, null c, 0 d, 100 e, "hello" f)
    UNION ALL
    (SELECT 5 a, 6 b, 0 c, null d, 200 e, "ok" f)
    UNION ALL
    (SELECT 8 a, 9 b, 10.0 c, 10 d, 300 e, "ok" f)
)
group by f
"""
df = gbq.read_gbq(q, dialect='standard')
df

col	f
[{u'a': 5, u'b': 6}, {u'a': 8, u'b': 9}]	ok
[{u'a': 1, u'b': 2}]	hello

Confirmed that col_order and index_col still work ~~(feel free to pull that out into a separate function since there's now redundant code with read_gbq())~~, and I removed the type conversion lines which appear to be unnecessary (google-cloud-python and/or pandas appears to do the necessary type conversion automatically, even if there are nulls; can confirm by examining the datatypes in the resulting dataframes).

codecov-io · 2017-04-08T04:53:59Z

Codecov Report

Merging #25 into master will decrease coverage by 43.58%.
The diff coverage is 7.84%.

@@             Coverage Diff             @@
##           master      #25       +/-   ##
===========================================
- Coverage   72.56%   28.97%   -43.59%     
===========================================
  Files           4        4               
  Lines        1578     1491       -87     
===========================================
- Hits         1145      432      -713     
- Misses        433     1059      +626

Impacted Files	Coverage Δ
pandas_gbq/tests/test_gbq.py	`27.98% <30%> (-54.36%)`	⬇️
pandas_gbq/gbq.py	`20.55% <6.29%> (-53.6%)`	⬇️
pandas_gbq/_version.py	`44.4% <0%> (+1.8%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cd76dde...26d6431. Read the comment docs.

jreback · 2017-04-08T13:13:53Z

pandas_gbq/gbq.py

@@ -767,6 +770,116 @@ def read_gbq(query, project_id=None, index_col=None, col_order=None,

    return final_df

+def from_gbq(query, project_id=None, index_col=None, col_order=None,


this should simply replace read_gbq, changing the top-level API is a non-starter.

Sure, I can rename mine to read_gbq and add in the query length/bytes processed info. Want me to delete the old read_gbq and related code or just rename it?

jreback · 2017-04-09T02:59:03Z

this need to pass all of the original test
it's simply an implementation change

jreback

can you show the output of running the test suite.

jreback · 2017-04-09T19:50:46Z

pandas_gbq/gbq.py


    Parameters
    ----------
    query : str
        SQL-Like Query to return data values
-    project_id : str
+    project_id : str (optional)


how is this optional?

project (str) – the project which the client acts on behalf of. Will be passed when creating a dataset / job. If not passed, falls back to the default inferred from the environment.

https://googlecloudplatform.github.io/google-cloud-python/stable/bigquery-client.html#google.cloud.bigquery.client.Client

Hm, doc still says that the project will be inferred from environment (https://googlecloudplatform.github.io/google-cloud-python/stable/bigquery/client), but I don't think it does anymore in the latest version. Thus, project_id is now required again, which is incredibly annoying as most of the time, you're probably ok with the query being associated with the same project. Thoughts on whether we should allow the user to specify a default project env variable or other method? (~/.bigqueryrc looks like it can hold your default project_id, but I don't know if that's deprecated and/or a command-line only implementation: https://cloud.google.com/bigquery/bq-command-line-tool)

@MaximilianR @jreback

jreback · 2017-04-09T19:51:12Z

pandas_gbq/gbq.py

        Google BigQuery Account project ID.
    index_col : str (optional)
        Name of result column to use for index in results DataFrame
    col_order : list(str) (optional)
        List of BigQuery column names in the desired order for results
        DataFrame
-    reauth : boolean (default False)


is this now not needed? if so, then simply mark it as deprecated (and raise a warning if its passed)

Yeah, not sure how to implement or if such behavior can be replicated in new api (or is even desired). If any folks have thought, let know. Otherwise, if I can't, will do as you suggest and raise a warning if reauth is passed.

jreback · 2017-04-09T19:51:51Z

pandas_gbq/gbq.py

    dialect : {'legacy', 'standard'}, default 'legacy'
        'legacy' : Use BigQuery's legacy SQL dialect.
        'standard' : Use BigQuery's standard SQL (beta), which is
        compliant with the SQL 2011 standard. For more information
        see `BigQuery SQL Reference
        <https://cloud.google.com/bigquery/sql-reference/>`__
-


why the change here?

From what I can gather from the discussion here (googleapis/google-cloud-python#2765) passing an arbitrary JSON of configuration settings isn't supported in the way it was in the previous python api. As such, we might as well make the passing of configuration settings a little easier with a dict like so, but happy to consider alternatives.

FYI: this is one reason I'd prefer to use google-auth library directly. #26

@tswast @jasonqng
You can get a Credentials object from the json with google.oauth2.service_account.Credentials.from_service_account_info(json.loads(key))

Forgive me for not PRing this sort of thing in directly - I'm totally jammed at the moment (forgive me @jreback too...). But happy to help with any questions on this stuff - we've run through a lot of it over here

@MaximilianR solved the auth with a key in latest commit. is this other concern about the config settings still an open issue to be resolved in this PR?

I think this is solved now, thank you @jasonqng

jreback · 2017-04-09T19:52:56Z

@jasonqng we have to be careful about back-compat here.

jreback · 2017-04-09T19:54:20Z

you need to update the ci/requirements-*.pip files, removing the old requirements and adding the new

jasonqng · 2017-04-10T02:36:23Z

@jreback Yeah, the back compatibility issues with the authentication is partly why I suggested writing it as a new function, but hopefully we can replicate a form of the pop-up authentication->refresh token with the new api (https://googlecloudplatform.github.io/google-cloud-python/stable/google-cloud-auth.html#user-accounts-3-legged-oauth-2-0-with-a-refresh-token). I might need some help with that if others are more familiar with it. Almost everything else should carry over, so I'm not too concerned with compatibility otherwise.

thejens · 2017-04-10T15:14:17Z

pandas_gbq/gbq.py

-    if dialect not in ('legacy', 'standard'):
-        raise ValueError("'{0}' is not valid for dialect".format(dialect))
+    if private_key:
+        os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = private_key


Not sure if it is nice to set environ vars, can't you just include a custom reference to the private_key, if it exists, when creating the Client on row 540?

Good suggestion, will do.

@thejens Fixed and verified it works with my service account json key.

thejens · 2017-04-10T15:15:07Z

requirements.txt

@@ -2,3 +2,5 @@ pandas
 httplib2
 google-api-python-client
 oauth2client
+google-cloud


google-cloud is not the most stable package (I've noticed), I'd require a specific version

that's fine, you can pip specifically

```google-cloud=0.24.0`` (or whatever)

You can also be more specific and only require the bigquery library with google-cloud-bigquery: https://github.com/GoogleCloudPlatform/google-cloud-python/tree/master/bigquery

+1 to using google-cloud-bigquery.

Also, the google-cloud-bigquery package is still beta, meaning there will likely be breaking changes, so we'd have to pin a very specific version number to maintain stability.

Fixed and pinned to google-cloud-bigquery to 0.25.0 since 0.26.0 breaks things

parthea · 2017-04-12T03:09:04Z

pandas_gbq/gbq.py

@@ -603,106 +458,56 @@ def delete_and_recreate_table(self, dataset_id, table_id, table_schema):
        table.create(table_id, table_schema)
        sleep(delay)

-
-def _parse_data(schema, rows):


In test_gbq.py, there are several tests that have gbq._parse_data(...). Could you update test_gbq.py as well? See the tests with prefix test_should_return_bigquery*(...).

Yup, will tackle this weekend hopefully. Thanks!

tswast · 2017-04-21T18:20:29Z

I recommend closing this PR.

The google-cloud-bigquery library is not yet 1.0. Breaking changes are very likely as we (Google) get the libraries in good shape for a 1.0 release (no specific timeline on that yet). There are also a few things yet-to-be-implemented in google-cloud-bigquery such as retry logic, which are in the "API client library".

We should revisit this after the google-cloud-bigquery library goes 1.0.

jasonqng · 2017-04-21T20:24:57Z

@jreback In light of @tswast's comment, should we just close this or should we go back to building this as a separate function (e.g. from_gbq()) as I had it before? We could clearly mark it as experimental. For my own selfish sake, I'd vote for latter just so I can start using it in the mean time before google-cloud API goes 1.0, but obviously defer to others here.

jreback · 2017-04-21T22:34:27Z

we can easily just pin to a specific version of api stability is a concern but in general i don't see this as a big deal

no reason to wait for a 1.0

Pandas itself is not even 1.0 and lots of people use / depend on it

tswast · 2017-04-27T00:02:41Z

pandas_gbq/gbq.py

-      By default "application default credentials" are used.
-
-      If default application credentials are not found or are restrictive,
-      user account credentials are used. In this case, you will be asked to


Are you sure you want to remove the user account credentials method? I believe it can be made to work with the google-cloud-biguery library. I'm working on adding a sample to the Google docs that does the user-auth flow.

That would be awesome. Do feel free to point to the doc when it's added. Sorry, been a bit distracted with work but hope to address all comments next week (and agree with @jreback's comment to move forward with this). Thanks.

GoogleCloudPlatform/python-docs-samples#925

That pull request adds a sample which uses https://pypi.python.org/pypi/google-auth-oauthlib to run a query with user account credentials.

I'll be writing some tests and docs around it, but the code should be easy enough to follow.

@tswast Latest commit uses existing GbqConnector to generate appropriate credentials. Let know if anything inappropriate with the way I've implemented.

tswast · 2017-05-23T17:38:23Z

@jasonqng You'll probably want to rebase this change after #39 gets merged. The google-auth library is used by both google-api-python-client and google-cloud-python, so you should be able to reuse my changes for user and service account authentication.

jreback · 2017-07-07T10:03:59Z

is this PR relevant after #39 ?

@tswast

parthea · 2017-07-07T10:11:40Z

We still have #23 as an open issue. It would be great to move this forward to address that. I recently added a conda recipe for google-cloud-bigquery.
https://github.com/conda-forge/google-cloud-bigquery-feedstock

jreback · 2017-07-07T11:32:22Z

@parthea ok is it worth waiting on this for 0.2.0 to avoid even more changes?

parthea · 2017-07-07T12:34:14Z

My initial thought is that the milestone for this PR should be 0.3.0 as thorough testing is required. I think ultimately it depends on how soon we can begin testing this PR and whether we are in a hurry to release 0.2.0.

@jasonqng Please could you rebase ?

tswast · 2017-07-08T21:06:42Z

Yeah, this PR is still relevant. #39 moves pandas-gbq to use google-auth but still use the Google API client libraries. This PR is to move it to the Google Cloud client libraries (which incidentally also uses google-auth, so rebasing would be really helpful in completing the work required to get this working properly)

jasonqng · 2017-07-10T01:42:10Z

@tswast @parthea @jreback Sorry, been swamped these past few months. Hope to scratch out some time this week to incorporate all comments (and also get it to working with queries with large results, which it currently fails with). Just checking, any particular reason to rebase vs a merge? Happy to do former, just haven't done rebase on any collaborative projects so this would be first time. (Haha, worst case scenario, I mess up my branch and I just rewrite and open a new PR branched off new master.)

parthea · 2017-07-11T04:11:55Z

@jasonqng Thanks for taking care of this! Rebase is preferred because it will allow you to add commits on top of the latest master which is much nicer to look at during code review.

Also says which libraries are no longer required, for easier upgrades.

tswast · 2017-11-30T22:12:25Z

@jreback Could you take a look at the docs changes I made? I've documented both the dependencies for 0.3.0 and what they were before (with notes on how they've changed)

tswast

This change LGTM, but since I made some contributions to this one I'd like one of the other maintainers to also review before we merge it.

max-sixty · 2017-12-06T20:34:05Z

pandas_gbq/gbq.py

@@ -781,6 +656,14 @@ def verify_schema(self, dataset_id, table_id, schema):
                               key=lambda x: x['name'])
        fields_local = sorted(schema['fields'], key=lambda x: x['name'])

+        # Ignore mode when comparing schemas.
+        for field in fields_local:
+            if 'mode' in field:


not worth changing, but this could be marginally simpler as field.pop('mode', None)

max-sixty · 2017-12-06T20:35:56Z

pandas_gbq/gbq.py

+        dtype_map.get(field['type'].upper(), object)
+        for field in fields
+    ]
+    print(fields)


Do we want to print all the fields here?

If we do: elsewhere, this uses self._print, which can can transition to logging at some point.

Oops. Good catch. I added that to debug the "ignore mode when comparing schemas" logic. Removed.

max-sixty · 2017-12-06T20:40:56Z

pandas_gbq/gbq.py

-                        tableId=table_id,
-                        body=body).execute()
-                except HttpError as ex:
+                    self.client.load_table_from_file(


This is so much better than the existing method

Yeah, it's technically a change in behavior (kicks off a load job instead of using the streaming API), but I think the change is small enough to be worth it. Load jobs should be much more reliable for the use case of this library.

max-sixty · 2017-12-06T20:43:22Z

pandas_gbq/gbq.py

-            rows.append(row_dict)
+            row_json = row.to_json(
+                force_ascii=False, date_unit='s', date_format='iso')
+            rows.append(row_json)


This isn't worse than the last version, but this would be much faster if .to_json was called on the whole table rather than each row, iterating in python

CSV might be even faster given the reduced space (and pandas can't use nesting or structs anyway). But potentially wait until parquet is GA to make the jump

I'd prefer to keep the current behavior for now and do a subsequent PR for any changes like this for performance improvements. I've filed #96 to track the work for speeding up encoding for the to_gbq() method.

100% re doing that separately

max-sixty · 2017-12-06T20:46:54Z

pandas_gbq/gbq.py

-                                       field_type)
+    for row_num, entries in enumerate(rows):
+        for col_num in range(len(col_types)):
+            field_value = entries[col_num]


I don't think I'd realized we were looping over all the values in python before. This explains a lot of why exporting a query to a file on GCS and then reading from that file is an order of magnitude faster.

If we could pass rows directly into DataFrame that would be much faster, but I'm not sure if that's possible

I've filed #97 to track improving the performance in the read case.

max-sixty · 2017-12-06T20:47:29Z

pandas_gbq/gbq.py

-                                       field_type)
+    for row_num, entries in enumerate(rows):
+        for col_num in range(len(col_types)):
+            field_value = entries[col_num]


Because this is being called so many times, you may even get a small speed up from eliminating assigning to field_value

(but these are all things that are either the same or better than the existing version)

max-sixty · 2017-12-06T20:51:27Z

pandas_gbq/gbq.py

                self._print('Standard price: ${:,.2f} USD\n'.format(
-                    bytes_processed * self.query_price_for_TB))
+                    bytes_billed * self.query_price_for_TB))

            self._print('Retrieving results...')


Presumably this is never going to be relevant because the prior part is blocking?

Yes and no. I think it indeed is less relevant, but actually fetching the rows and constructing the dataframe will take non-zero time, especially for larger result sets.

max-sixty · 2017-12-06T21:13:11Z

I took a proper read through, though needs someone like @jreback to approve

I think this strictly dominates the existing version. There are a couple of extremely small tweaks that we can do in a follow-up if not now.

There are also some areas for huge speed-ups - IIUC the code is currently running through each value in python atm.

In line with that: we've built a function for exporting to a file to GCS and loading that in, which works much better for > 1-2m rows. We can do a PR for that if people are interested, in addition to speeding up the current path.

max-sixty · 2017-12-12T20:17:08Z

@jreback I know there's lots going on in pandas, but would be super if you could take a glance at this. A few follow-ups are dependent on this merging.

Thanks v much

jreback · 2017-12-12T20:18:00Z

sure will look

jreback

lgtm. some small doc comments that I would do. better to be over explanatory in the whatsnew.

jreback · 2017-12-14T12:21:26Z

docs/source/changelog.rst

+0.3.0 / 2017-??-??
+------------------
+
+- Use the `google-cloud-bigquery <https://googlecloudplatform.github.io/google-cloud-python/latest/bigquery/usage.html>`__ library for API calls instead of ``google-api-client`` and ``httplib2``. (:issue:`93`)


needs a more comprehensive note here. show what you used to import / depend on, and what it should be now.

you can link to install.rst as well

jreback · 2017-12-14T12:27:26Z

pandas_gbq/gbq.py

@@ -181,14 +158,6 @@ class QueryTimeout(ValueError):
    pass


-class StreamingInsertError(ValueError):


mention that this is eliminated in whatsnew

max-sixty · 2017-12-20T22:44:13Z

Congrats @jasonqng & @tswast !

tswast · 2017-12-20T22:48:45Z

Thanks! A couple things to clean up before we make a release. I'd like to add a couple tests for some of the other issues we think this PR might address.

Plus I'm not sure it makes sense to do a release right before the holidays.

jasonqng · 2017-12-21T05:34:40Z

TY @tswast for carrying this across the finish line!!! I learned so much watching you and @jreback mold this. Can't wait to stop having to rely on my janky fork and install the new version!

jasonqng force-pushed the read-gbq-google-cloud-library branch from 1381dd7 to a763cf0 Compare April 8, 2017 05:33

jreback reviewed Apr 8, 2017

View reviewed changes

jasonqng force-pushed the read-gbq-google-cloud-library branch from 94065f3 to 2f93f7b Compare April 9, 2017 14:00

jreback suggested changes Apr 9, 2017

View reviewed changes

jasonqng changed the title ~~Add new from_gbq() function using google-cloud-python~~ Convert read_gbq() function to use google-cloud-python Apr 10, 2017

thejens reviewed Apr 10, 2017

View reviewed changes

parthea reviewed Apr 12, 2017

View reviewed changes

jreback mentioned this pull request Apr 19, 2017

Use google-auth library and requests instead of google-api-client #26

Closed

parthea added this to the 0.2.0 milestone Apr 22, 2017

tswast reviewed Apr 27, 2017

View reviewed changes

parthea changed the title ~~Convert read_gbq() function to use google-cloud-python~~ ENH: Convert read_gbq() function to use google-cloud-python Jun 13, 2017

jreback removed this from the 0.2.0 milestone Jul 7, 2017

parthea mentioned this pull request Jul 11, 2017

ENH: Supporting REPEATED schema for list types #60

Closed

Document dependencies for previous verions.

e959571

Also says which libraries are no longer required, for easier upgrades.

tswast mentioned this pull request Nov 30, 2017

Update pandas.read_gbq docs to point to pandas-gbq pandas-dev/pandas#18548

Merged

1 task

tswast approved these changes Dec 5, 2017

View reviewed changes

max-sixty reviewed Dec 6, 2017

View reviewed changes

tswast mentioned this pull request Dec 8, 2017

Do not manually loop over all rows when encoding a dataframe as JSON #96

Closed

Remove print statement used for debugging.

6448abb

This was referenced Dec 8, 2017

Do not manually loop over all rows when reading a dataframe #97

Closed

StreamingInsertError occurs when uploading to a table with a new schema #75

Closed

jreback approved these changes Dec 14, 2017

View reviewed changes

Add deps and StreamingInsertError to changelog.

26d6431

tswast merged commit afdbcfa into googleapis:master Dec 20, 2017

This was referenced Dec 20, 2017

Structs lack proper names as dicts and arrays get turned into array of dicts #23

Closed

Add support for DATETIME BigQuery type #69

Closed

This was referenced Dec 21, 2017

Bulk upload #7

Closed

DOC: link to issue 7 for to_gbq changes #100

Merged

This was referenced Jan 3, 2018

Set project_id (and other settings) once for all subsequent queries so you don't have to pass every time #103

Closed

Use google-cloud-bigquery instead of google-api-client #93

Closed

		@@ -767,6 +770,116 @@ def read_gbq(query, project_id=None, index_col=None, col_order=None,

		return final_df

		def from_gbq(query, project_id=None, index_col=None, col_order=None,

		@@ -181,14 +158,6 @@ class QueryTimeout(ValueError):
		pass


		class StreamingInsertError(ValueError):

ENH: Convert read_gbq() function to use google-cloud-python #25

ENH: Convert read_gbq() function to use google-cloud-python #25

Conversation

jasonqng commented Apr 8, 2017 • edited Loading

Description

codecov-io commented Apr 8, 2017 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback commented Apr 9, 2017

jreback 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

jreback commented Apr 9, 2017

jreback commented Apr 9, 2017

jasonqng commented Apr 10, 2017

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

tswast commented Apr 21, 2017

jasonqng commented Apr 21, 2017

jreback commented Apr 21, 2017

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

tswast commented May 23, 2017

jreback commented Jul 7, 2017

parthea commented Jul 7, 2017

jreback commented Jul 7, 2017

parthea commented Jul 7, 2017

tswast commented Jul 8, 2017

jasonqng commented Jul 10, 2017

parthea commented Jul 11, 2017

tswast commented Nov 30, 2017

tswast 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

max-sixty Dec 6, 2017 • 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

max-sixty Dec 6, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

max-sixty commented Dec 6, 2017 • edited Loading

max-sixty commented Dec 12, 2017

jreback commented Dec 12, 2017

jreback 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

max-sixty commented Dec 20, 2017

tswast commented Dec 20, 2017

jasonqng commented Dec 21, 2017

jasonqng commented Apr 8, 2017 •

edited

Loading

codecov-io commented Apr 8, 2017 •

edited

Loading

max-sixty Dec 6, 2017 •

edited

Loading

max-sixty Dec 6, 2017 •

edited

Loading

max-sixty commented Dec 6, 2017 •

edited

Loading