[SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON #17956

HyukjinKwon · 2017-05-12T02:47:03Z

What changes were proposed in this pull request?

This PR is based on #16199 and extracts the valid change from #9759 to resolve SPARK-18772

This avoids additional conversion try with toFloat and toDouble.

For avoiding additional conversions, please refer the codes below:

Before

scala> import org.apache.spark.sql.types._
import org.apache.spark.sql.types._

scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show()
17/05/12 11:30:41 ERROR Executor: Exception in task 0.0 in stage 2.0 (TID 2)
java.lang.NumberFormatException: For input string: "nan"
...

After

scala> import org.apache.spark.sql.types._
import org.apache.spark.sql.types._

scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show()
17/05/12 11:44:30 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.RuntimeException: Cannot parse nan as DoubleType.
...

How was this patch tested?

Unit tests added in JsonSuite.

Closes #16199

HyukjinKwon · 2017-05-12T02:48:30Z

cc @NathanHowell, @cloud-fan and @viirya.

(I just want to note this will not change any input/output but just the exception type and avoid additional conversion try.)

HyukjinKwon · 2017-05-12T02:55:17Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JacksonParser.scala

#9759 (comment)

"infinity".toDouble, "inf".toDouble are not legal. These non-numeric numbers are case-sensitive, both for Jackson and Scala.

cloud-fan · 2017-05-12T03:13:00Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JacksonParser.scala

how about

parser.getText match { case "NaN" => Float.NaN case "+INF" | "INF" | "+Infinity" | "Infinity" => Float.PositiveInfinity case "-INF" | "-Infinity" => Float.NegativeInfinity case other => throw new RuntimeException(s"Cannot parse $other as FloatType.") }

cloud-fan · 2017-05-12T03:46:59Z

LGTM

SparkQA · 2017-05-12T05:05:47Z

Test build #76841 has finished for PR 17956 at commit 660a284.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

viirya · 2017-05-12T05:17:00Z

sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/json/JsonSuite.scala

Shall we also test other negative cases?

viirya · 2017-05-12T05:17:19Z

LGTM except for a minor comment.

SparkQA · 2017-05-12T06:04:02Z

Test build #76846 has finished for PR 17956 at commit aa7c658.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

gatorsmile · 2017-05-12T06:21:58Z

sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/json/JsonSuite.scala

What is difference between line 1995 and line 1997?

Do we want to test +INF?

gatorsmile · 2017-05-12T06:24:26Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JacksonParser.scala

This sounds a bug fix, because toFloat is case sensitive?

For input/output, it is not a bug. Please see #17956 (comment) and the PR description.

Not sure whether we should follow it. How about our CSV parsers? How about Pandas? Are they case sensitive?

Is JSON format itself related with CSV and Pandas? Please see the discussion in #9759 (comment).

Also, this does not change the existing behaviour.

If possible, we want to be consistent with the others. Could you check Pandas?

Probably let me check and open JIRA/PR if there are some cases we should handle when I have some time. Let's leave out that here. It sounds that does not block this PR and I don't want extra changes hold off this PR like #17901.

This PR changes the behavior, right? Does your test case pass without the code changes?

Okay, this adds some cases. I am not still convinced that we should take care of every other cases. Let's add options nanValue, positiveInf and negativeInf in a separate PR like our CSV datasource.

Without this PR, +INF, INF and -INF" do not work.

Yea, I added some cases here. I will open another PR/JIRA to add those options into JSON related functionalities.

SparkQA · 2017-05-12T06:27:36Z

Test build #76851 has started for PR 17956 at commit e858789.

gatorsmile · 2017-05-12T06:53:42Z

This PR description is misleading. This PR is actually a bug fix, I think

gatorsmile · 2017-05-12T07:02:10Z

sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/json/JsonSuite.scala

I do not understand which standards we follow. The above seven values are accepted by toFloat and toDouble?

This follows Jackson + Scala. #9759 (comment)

Below is from JsonParser.Feature ALLOW_NON_NUMERIC_NUMBERS

"INF" (for positive infinity), as well as alias of "Infinity"
"-INF" (for negative infinity), alias "-Infinity"
"NaN" (for other not-a-numbers, like result of division by zero)

Then, this PR also adds +INF and +Infinity.

What is the Scala standard?

Just let us assume we want to follow JsonParser.Feature ALLOW_NON_NUMERIC_NUMBERS, does our JSON option flag work? allowNonNumericNumbers.

If we turn it off, these test cases should fail, right?

I think we also can address https://issues.apache.org/jira/browse/SPARK-11753 in this PR, right? Simply accepting all these non-numeric numbers without respecting our flag allowNonNumericNumbers does not sound right to me.

HyukjinKwon · 2017-05-12T07:06:06Z

retest this please

HyukjinKwon · 2017-05-12T07:36:19Z

@gatorsmile, I added some more cases I could identify based on http://docs.oracle.com/javase/7/docs/api/java/lang/Double.html#valueOf(java.lang.String) (this looks calling java.lang.Double.parseDouble which toDouble calls).

And also, for sure, I made this falls back to toDouble or toFloat and throw the same exceptions.

… tests

gatorsmile · 2017-05-12T07:42:37Z

Could we update the JIRA, PR title, and PR description?

Also check the JSON flag allowNonNumericNumbers when doing the conversion?

cloud-fan · 2017-05-12T07:50:59Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JacksonParser.scala

-            value.toFloat
-          } else {
-            throw new RuntimeException(s"Cannot parse $value as FloatType.")
+          parser.getText match {


one last question is, should we be case sensitive here? can we check other systems like pandas?

Up to my knowledge, at least Python supports some case-insensitive cases. I would rather leave this working with options (discussed in #17956 (comment)) treating this as a variant. At least, I think we can say here it follows Scala + Jackson's standard.

It seems to me that we don't really support case insensitive before. Although we check the lower case of input string, but we actually call toFloat on the original string. And "nan".toFloat will get java.lang.NumberFormatException: For input string: "nan".

Seems to me NaN and Infinity, INF are special and case insensitive are not really making sense.

Actually, here, we are not trying to educate users/customers what are right or not. We are trying to improve the usability. If the other popular platforms accept some formats, we need to follow them even if they are strange to us.

I think we should not add all the variants in the default behaviour. We could add the options for those cases. We all already have different ideas. Is it worth adding those cases right now in this PR? This is not the end of fixing this code path.

HyukjinKwon · 2017-05-12T07:58:22Z

We should upgrade Jackson accordingly to support that option correctly. This looks reverted due to dependency problem - #9759 (comment). It sounds a bigger problem to me. (Also, strictly, it looks rather related with parsing itself). We already support the special floating cases and it might be fine just to improve this case.

I updated PR title and description already. I will fix JIRA as well (I was hesitated because that was not open by me).

gatorsmile · 2017-05-12T08:03:54Z

If we do not upgrade Jackson, is that possible we can do it by checking the flag allowNonNumericNumbers before doing the conversion?

gatorsmile · 2017-05-12T08:06:08Z

uh... That is different... This is String. That is Float. nvm. They are different issues.

HyukjinKwon · 2017-05-12T08:06:19Z

Yes, thanks.

gatorsmile · 2017-05-12T08:11:18Z

sql/core/src/test/scala/org/apache/spark/sql/execution/datasources/json/JsonSuite.scala

+      """{"a": "-NaN"}""",
+      """{"a": "Infinity"}""",
+      """{"a": "+Infinity"}""",
+      """{"a": "-Infinity"}""")


I did not check it, but does """{"a": -Infinity}""" fail?

It looks -Infinity succeeding.

scala> "-Infinity".toDouble res0: Double = -Infinity scala> "-Infinity".toFloat res1: Float = -Infinity

It sounds like Jackson only support the following four cases.

val jsons = Seq( """{"a": NaN}""", """{"a": Infinity}""", """{"a": +Infinity}""", """{"a": -Infinity}""")

Could you add a separate test case for them? Also add the negative cases too.

See #17956 (comment)

gatorsmile · 2017-05-12T08:13:35Z

Do we need to follow allowNonNumericNumbers when parsing """{"a": "-Infinity"}"""? cc @cloud-fan @viirya

HyukjinKwon · 2017-05-12T08:17:52Z

I didn't understand what ^ means. Could you elaborate please? allowNonNumericNumbers option here for now is incomplete up to my knowledge.

viirya · 2017-05-12T08:36:49Z

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/json/JacksonParser.scala

-          } else {
-            throw new RuntimeException(s"Cannot parse $value as FloatType.")
+          parser.getText match {
+            case "NaN" | "+NaN" | "-NaN" => Float.NaN


I remember that Jackson's JsonParser.Feature.ALLOW_NON_NUMERIC_NUMBERS doesn't support "+NaN" and "-NaN".

Yea, but it is supported in Scala. These were added to address #17956 (comment) to say we support Scala + Jackson standard.

scala> "+NaN".toDouble res2: Double = NaN scala> "-NaN".toDouble res3: Double = NaN

viirya · 2017-05-12T08:43:47Z

Do we need to follow allowNonNumericNumbers when parsing """{"a": "-Infinity"}"""? cc @cloud-fan @viirya

allowNonNumericNumbers supports it and Scala supports it too. It seems to me that we should follow them.

Btw we support it too before this change.

SparkQA · 2017-05-12T09:52:01Z

Test build #76857 has finished for PR 17956 at commit d581467.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2017-05-12T10:03:25Z

Test build #76858 has finished for PR 17956 at commit fa63ff4.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

gatorsmile · 2017-05-12T17:41:53Z

Before this PR, the option allowNonNumericNumbers affects the results of the following four cases:

{"a": NaN}
{"a": Infinity}
{"a": +Infinity}
{"a": -Infinity}

We should also enable this flag for the following cases too.

{"a": "+INF"}
{"a": "INF"}
{"a": "-INF"}
{"a": "NaN"}
{"a": "+NaN"}
{"a": "-NaN"}
{"a": "Infinity"}
{"a": "+Infinity"}
{"a": "-Infinity"}

HyukjinKwon · 2017-05-12T17:54:10Z

allowNonNumericNumbers option is undocumented and incomplete for now - #9724 (comment)

allowNonNumericNumbers is undocumented for now, since I can't figure out how it works.

gatorsmile · 2017-05-12T18:10:41Z

This PR also introduces the following behavior changes.

    def floatRecords: Dataset[String] =
      spark.createDataset(spark.sparkContext.parallelize(
        """{"f": "18.00"}""" ::
          """{"f": "18.30"}""" ::
          """{"f": "20.00"}""" :: Nil))(Encoders.STRING)

    val jsonDF = spark.read.schema("f float").json(floatRecords)

Before this PR, the output is like

+----+
|   f|
+----+
|null|
|null|
|null|
+----+

After this PR, the output is like

+----+
|   f|
+----+
|18.00|
|18.30|
|20.00|
+----+

HyukjinKwon · 2017-05-12T18:14:21Z

Yea, I am aware of it because we support special floating cases in string and I think we'd expect other doubles could be in string?

HyukjinKwon · 2017-05-12T18:16:05Z

Hey @gatorsmile, how about picking up the commits here and opening a PR?

gatorsmile · 2017-05-12T18:16:09Z

We really need to be careful when making the code changes. If we support such a feature, we need to support all the data types. This is unexpected to users.

HyukjinKwon · 2017-05-12T18:25:41Z

WDYT? I think it'd be faster to get this merged.

gatorsmile · 2017-05-12T18:43:38Z

Let me share my understanding.

First, we do not have a deadline for any non-urgent fixes. This PR is an example. How many JSON writers write a float "INF" as a string?

Being a committer needs to make a judgement to see how critical a PR is. My personal judgement is that we do not need to rush to merge it. I even want to wait for understanding how the other platforms handle such a string-represented "INF" in a float-type column. I believe we are not the first platform to face such an issue. That is why @cloud-fan and I asked the question above.

Second, we have to be careful to avoid introducing bugs when fixing the old issue. Spark is an infrastructure open-source project. We have to be very careful when doing the code reviews and code changes. If we accidentally introduced some supports, it is painful to remove the support and thus we have to keep it maybe forever. @viirya just found it very recently. Thus, please be careful when you do the code review and code changes.

If the users really need such a fix, they also can fix it and make a build by themselves. I believe that is why people like open source projects.

At the end, I want to share you a slides about bug costs. You might already saw similar figure before.

HyukjinKwon · 2017-05-13T00:24:35Z

It is not good to hold off PRs. If a PR looks good and coherent, I think we could merge.
"How many JSON writers write a float "INF" as a string?", if it is your only worry, I will take out the cases here. This is why I am thinking we should add such cases via options.
My question was to ask taking over this. Please answer to this question.
Other information looks not related with the topic in this PR.

gatorsmile · 2017-05-13T06:07:32Z

Based on my previous work experience, I am very conservative to introduce any behavior change unless we are sure this is the right thing to do. I do not think this is a high priority PR. I can put it in my to-do list. When I am free, I can work on it.

To prove the significance of this change, we need to check whether the other projects/products have the same/similar behavior? We do not want to introduce anything new or different. This is my concern.

Below is the list we need to check.

{"a": "+INF"}
{"a": "INF"}
{"a": "-INF"}
{"a": "NaN"}
{"a": "+NaN"}
{"a": "-NaN"}
{"a": "Infinity"}
{"a": "+Infinity"}
{"a": "-Infinity"}

Note, the above is different from the following JSON strings.

{"a": NaN}
{"a": Infinity}
{"a": +Infinity}
{"a": -Infinity}

In my previous comment, I just want to share something for the people who want to be a committer in Spark. It is not related to this PR.

gatorsmile · 2017-05-13T06:13:13Z

BTW, if we want to support the JSON strings like {"a": "-Infinity"} as the FLOAT type, I think we also should support the float data in a string. Below is an example.

    def floatRecords: Dataset[String] =
      spark.createDataset(spark.sparkContext.parallelize(
        """{"f": "18.00"}""" ::
          """{"f": "18.30"}""" ::
          """{"f": "20.00"}""" :: Nil))(Encoders.STRING)
    val jsonDF = spark.read.schema("f float").json(floatRecords)

If we want to support parsing float data in a string, we should also support the other data types. Thus, this is another to-do issue, we should do an investigation.

HyukjinKwon · 2017-05-13T06:48:57Z

Yea, Let's focus on the topic. For the cases below:

{"a": NaN}
{"a": Infinity}
{"a": +Infinity}
{"a": -Infinity}

They are related with allowNonNumericNumbers and related tests are already being ignored.

Okay. So the point is behaviour change, right? Let me give a shot to remove the behaviour changes here.

HyukjinKwon · 2017-05-13T07:09:45Z

Given the existing logic below:

if (lowerCaseValue.equals("nan") ||
  lowerCaseValue.equals("infinity") ||
  lowerCaseValue.equals("-infinity") ||
  lowerCaseValue.equals("inf") ||
  lowerCaseValue.equals("-inf")) {
    value.toFloat

It looks we have try all the combinations (lowercase and uppcass) of "nan", "infinity", "-infinity", "inf", "-inf" sofar.

import itertools
items = ["nan", "infinity", "-infinity", "inf", "-inf"]
comb = []
for item in items:
   comb.extend(map(''.join, itertools.product(*zip(item.upper(), item.lower()))))

print("Set(%s)" % ", ".join(map(lambda c: '"%s"' % c, comb)))

This code is to generate the combinations. (I feel convenient for dealing with strings in Python).

val s = Set("NAN", "NAn", ... "-inf")
s.foreach { w =>
  try {
    w.toDouble // or w.toFloat
    println(w)
  } catch {
    case _ => // pass
  }
}

It looks we have supported the cases below so far:

NaN
-Infinity
Infinity

Let me update this.

gatorsmile · 2017-05-13T07:29:05Z

We are handling the Json data sources here. Are the following inputs widely used?

{"a": "+INF"}
{"a": "INF"}
{"a": "-INF"}
{"a": "NaN"}
{"a": "+NaN"}
{"a": "-NaN"}
{"a": "Infinity"}
{"a": "+Infinity"}
{"a": "-Infinity"}

gatorsmile · 2017-05-13T07:30:12Z

Note, we do not support the following cases

    def floatRecords: Dataset[String] =
      spark.createDataset(spark.sparkContext.parallelize(
        """{"f": "18.00"}""" ::
          """{"f": "18.30"}""" ::
          """{"f": "20.00"}""" :: Nil))(Encoders.STRING)
    val jsonDF = spark.read.schema("f float").json(floatRecords)

Basically, we do not assume users are use string in double columns for JSON data sources.

HyukjinKwon · 2017-05-13T07:32:07Z

Unfortunately, we already support

"NaN"
"-Infinity"
"Infinity"

Now, this PR targets primarily to avoid unnecessary conversion try primarily.

HyukjinKwon · 2017-05-13T08:39:05Z

@gatorsmile, @cloud-fan and @viirya, could you take another look please? I tried to get rid of all the behaviour changes existing in both previous PRs but only leave the change to avoid the conversion.

I tried to generate combinations of the existing supports - #17956 (comment) and extracted three cases - "NaN", "-Infinity" and "Infinity" which I believe we have supported so far.

SparkQA · 2017-05-13T09:43:24Z

Test build #76894 has finished for PR 17956 at commit 90330bc.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

…s in JSON ## What changes were proposed in this pull request? This PR is based on #16199 and extracts the valid change from #9759 to resolve SPARK-18772 This avoids additional conversion try with `toFloat` and `toDouble`. For avoiding additional conversions, please refer the codes below: **Before** ```scala scala> import org.apache.spark.sql.types._ import org.apache.spark.sql.types._ scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show() 17/05/12 11:30:41 ERROR Executor: Exception in task 0.0 in stage 2.0 (TID 2) java.lang.NumberFormatException: For input string: "nan" ... ``` **After** ```scala scala> import org.apache.spark.sql.types._ import org.apache.spark.sql.types._ scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show() 17/05/12 11:44:30 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0) java.lang.RuntimeException: Cannot parse nan as DoubleType. ... ``` ## How was this patch tested? Unit tests added in `JsonSuite`. Closes #16199 Author: hyukjinkwon <[email protected]> Author: Nathan Howell <[email protected]> Closes #17956 from HyukjinKwon/SPARK-18772. (cherry picked from commit 3f98375) Signed-off-by: Wenchen Fan <[email protected]>

cloud-fan · 2017-05-13T12:59:31Z

thanks, merging to master/2.2!

I think this change is pretty safe, we can discuss 2 things later:

if we want to support more special strings like Inf
if we want to make it case-insensitive

HyukjinKwon · 2017-05-13T13:09:08Z

Thank you everybody sincerely.

…s in JSON ## What changes were proposed in this pull request? This PR is based on apache#16199 and extracts the valid change from apache#9759 to resolve SPARK-18772 This avoids additional conversion try with `toFloat` and `toDouble`. For avoiding additional conversions, please refer the codes below: **Before** ```scala scala> import org.apache.spark.sql.types._ import org.apache.spark.sql.types._ scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show() 17/05/12 11:30:41 ERROR Executor: Exception in task 0.0 in stage 2.0 (TID 2) java.lang.NumberFormatException: For input string: "nan" ... ``` **After** ```scala scala> import org.apache.spark.sql.types._ import org.apache.spark.sql.types._ scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show() 17/05/12 11:44:30 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0) java.lang.RuntimeException: Cannot parse nan as DoubleType. ... ``` ## How was this patch tested? Unit tests added in `JsonSuite`. Closes apache#16199 Author: hyukjinkwon <[email protected]> Author: Nathan Howell <[email protected]> Closes apache#17956 from HyukjinKwon/SPARK-18772.

[SPARK-18772][SQL] NaN/Infinite float parsing in JSON is inconsistent

f83eec7

HyukjinKwon commented May 12, 2017

View reviewed changes

cloud-fan reviewed May 12, 2017

View reviewed changes

viirya reviewed May 12, 2017

View reviewed changes

gatorsmile reviewed May 12, 2017

View reviewed changes

HyukjinKwon changed the title ~~[SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON and add related tests~~ [SPARK-18772][SQL] Avoid unnecessary conversion try and add missing cases for special floats in JSON May 12, 2017

Avoid unnecessary cast try for special floats in JSON and add related…

fa63ff4

… tests

HyukjinKwon force-pushed the SPARK-18772 branch from d581467 to fa63ff4 Compare May 12, 2017 07:39

cloud-fan reviewed May 12, 2017

View reviewed changes

gatorsmile reviewed May 12, 2017

View reviewed changes

viirya reviewed May 12, 2017

View reviewed changes

Address comments

90330bc

HyukjinKwon changed the title ~~[SPARK-18772][SQL] Avoid unnecessary conversion try and add missing cases for special floats in JSON~~ [SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON May 13, 2017

asfgit closed this in 3f98375 May 13, 2017

HyukjinKwon deleted the SPARK-18772 branch January 2, 2018 03:42

[SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON #17956

[SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON #17956

Uh oh!

Conversation

HyukjinKwon commented May 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

HyukjinKwon commented May 12, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cloud-fan commented May 12, 2017

Uh oh!

SparkQA commented May 12, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

viirya commented May 12, 2017

Uh oh!

SparkQA commented May 12, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gatorsmile May 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HyukjinKwon May 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented May 12, 2017

Uh oh!

gatorsmile commented May 12, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HyukjinKwon commented May 12, 2017

Uh oh!

HyukjinKwon commented May 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gatorsmile commented May 12, 2017

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HyukjinKwon May 12, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

HyukjinKwon commented May 12, 2017 •

edited

Loading

gatorsmile May 12, 2017 •

edited

Loading

HyukjinKwon May 12, 2017 •

edited

Loading

HyukjinKwon commented May 12, 2017 •

edited

Loading

HyukjinKwon May 12, 2017 •

edited

Loading