Fastq record converter #1185

zyxue · 2016-09-28T05:51:09Z

There are still problems:

The description for convertRead does don't match with what it does (e.g. the return type). The method name is kind of vague, not sure what behavior is actually intended. The refactoring tries not to change its behavior.
I am also a bit confused for what default values should be used when creating a AlignmentRecord instance based on a Fastq entry. The original code set many fields to null. I thought leaving them as default would make more sense. e.g. ReadNegativeStrand is set to null while the default is false. ProperPair is set to true, which I don't think make sense since you can't really tell just based on Fastq, the default is also fasle. It would be helpful if someone can clarify what's the most sensible values for the following fields for both paired-end and single-end fastq entries.

.setReadPaired(readPaired)
.setReadInFragment(readInFragment)
.setReadNegativeStrand(null)
.setMateNegativeStrand(null)
.setPrimaryAlignment(null)
.setSecondaryAlignment(null)
.setSupplementaryAlignment(null)

Also, what to do when the read length and that of qualities don't match? There is a stringency parameter involved in convertRead. If it's STRICT, then qualities must exist (CANNOT be *) and also match the length of reads. When it's NOT STRICT, qualities will be padded with B if not exist or shorter than read length. If it's longer than read length, NotImplementedError will be thrown. Such behavior seems quite arbitrary and doesn't make much sense to me, and it doesn't apply to convertPair and convertFragment, in which the read length and qualities must match without consideration for stringency.

tested FastqRecordConverter.convertPair

when fastq lines number doesn't match

Refactored out the common part between FastqRecordConverter.convertPair & FastqRecordConverter.convertFragment

shared by convertPair & convertFragment

factoring out common code from convertRead

There is still quite a bit of problem with default values when creating a AlignmentRecord instance from a Fastq entry

bigdatagenomics#1172

AmplabJenkins · 2016-09-28T05:53:01Z

Can one of the admins verify this patch?

heuermh · 2016-09-28T11:57:54Z

Jenkins, test this please.

AmplabJenkins · 2016-09-28T12:00:43Z

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/ADAM-prb/1505/

Build result: FAILURE

GitHub pull request #1185 of commit d4c5ad6 automatically merged.Notifying endpoint 'HTTP:https://webhooks.gitter.im/e/ac8bb6e9f53357bc8aa8'[EnvInject] - Loading node environment variables.Building remotely on amp-jenkins-worker-05 (centos spark-test) in workspace /home/jenkins/workspace/ADAM-prb > /home/jenkins/git2/bin/git rev-parse --is-inside-work-tree # timeout=10Fetching changes from the remote Git repository > /home/jenkins/git2/bin/git config remote.origin.url https://github.com/bigdatagenomics/adam.git # timeout=10Fetching upstream changes from https://github.com/bigdatagenomics/adam.git > /home/jenkins/git2/bin/git --version # timeout=10 > /home/jenkins/git2/bin/git -c core.askpass=true fetch --tags --progress https://github.com/bigdatagenomics/adam.git +refs/pull/:refs/remotes/origin/pr/ # timeout=15 > /home/jenkins/git2/bin/git rev-parse origin/pr/1185/merge^{commit} # timeout=10 > /home/jenkins/git2/bin/git branch -a --contains f280ecc19296a5841548d95e94b1bc3b986b0012 # timeout=10 > /home/jenkins/git2/bin/git rev-parse remotes/origin/pr/1185/merge^{commit} # timeout=10Checking out Revision f280ecc19296a5841548d95e94b1bc3b986b0012 (origin/pr/1185/merge) > /home/jenkins/git2/bin/git config core.sparsecheckout # timeout=10 > /home/jenkins/git2/bin/git checkout -f f280ecc19296a5841548d95e94b1bc3b986b0012First time build. Skipping changelog.Triggering ADAM-prb ? 2.6.0,2.11,1.5.2,centosTriggering ADAM-prb ? 2.6.0,2.10,1.5.2,centosTouchstone configurations resulted in FAILURE, so aborting...Notifying endpoint 'HTTP:https://webhooks.gitter.im/e/ac8bb6e9f53357bc8aa8'
Test FAILed.

heuermh · 2016-09-28T12:07:50Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+
+  private def parseReadPairInFastq(input: String): (String, String, String, String, String, String) = {
+    val lines = input.toString.split('\n')
+    require(lines.length == 8,


I've mentioned this before; perhaps now is the time to fix it? FASTQ format allows for hard line wrapping, so there may be new line characters at any place in the record.

Can you give an example for hard line wrapping? What does it look like?

There are all kinds of correct and incorrect examples here https://github.com/biojava/biojava/tree/master/biojava-sequencing/src/test/resources/org/biojava/nbio/sequencing/io/fastq

I think we should make this work for the simple case first (what we have currently implemented --> fastq record is 4 lines, interleaved read pair is 8 lines). In a follow on, we can make the arbitrary wrapping case work. In my experience, "simply" formatted files are much more common than arbitrarily formatted files.

I tried to implement parsing for wrapped lines. Then I found that it would require exact match of sequence length and quality length. Otherwise, it's ambiguous to tell when the quality lines stop. This makes padding with B when length(qual line) < length(seq line), is that right? e.g. error_short_qual.fastq from biojava is an error.

heuermh · 2016-09-28T12:10:03Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      s"Input must have 4 lines (${lines.length.toString} found):\n${input}")
+
+    val readName = lines(0).drop(1)
+    if (readName.endsWith("/1") && setSecondOfPair)


I've seen files in the wild that use 1 and 2 (with space) instead of /1 and /2. Should we add that here?
See e.g. PairedEndFastqReader.java#L59

I have seen both, as well. I am not aware if there is a specification that lists all possibilities. I am thinking of using regex to account for all of them gradually.

There isn't a specification, only convention. See http://dx.doi.org/10.1093/nar/gkp1137

+1, I've also seen _1/2

addressed by regex like [/ +_]1$

heuermh · 2016-09-28T12:18:21Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      else {
+        if (readQualitiesRaw == "*") "B" * readSequence.length
+        else if (readQualitiesRaw.length < readSequence.length) readQualitiesRaw + ("B" * (readSequence.length - readQualitiesRaw.length))
+        else if (readQualitiesRaw.length > readSequence.length) throw new NotImplementedError("Not implemented")


NotImplementedError doesn't seem right, should be IllegalArgumentException. These length checks should also happen with strict stringency.

Also, what's the reason for padding B in case qualities information is incomplete?

IIRC, B is the code for "unknown" quality. CC @ryan-williams

https://en.wikipedia.org/wiki/FASTQ_format

Also, in Illumina runs using PhiX controls, the character 'B' was observed to represent an "unknown quality score". The error rate of 'B' reads was roughly 3 phred scores lower the mean observed score of a given run.

fixed NotImplementedError => IllegalArgumentException

heuermh · 2016-09-28T12:20:22Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      .setSequence(sequence)
+      .setQual(qual)
+      .setReadPaired(readPaired)
+      .setProperPair(null)


I'm not sure why these are explicitly set to null.

explicitly setting null is dropped in newer commits. If you rerun the tests, it should all pass.

heuermh · 2016-09-28T12:22:13Z

adam-core/src/test/scala/org/bdgenomics/adam/converters/FastqConverterSuite.scala

+import org.scalatest.FunSuite
+
+/**
+ * Created by zyxue on 2016-09-27.


Unit test suite for FastqRecordConverter.

What does this comment mean?

Sorry, am suggesting a doc comment change.

ok, removed the unnecessary comment.

heuermh · 2016-09-28T12:22:38Z

adam-core/src/test/scala/org/bdgenomics/adam/converters/FastqConverterSuite.scala

+/**
+ * Created by zyxue on 2016-09-27.
+ */
+class FastqConverterSuite extends FunSuite {


Should be FastqRecordConverterSuite.

Are you sure? I followed convention in FastaConverterSuite

OK, you're right. I realized it's

FastaConverter.scala & FastaConverterSuite

so it should be

FastqRecordConverter & FastqRecordConverterSuite

Yep, although I find the word "record" redundant almost everywhere, so we could possibly drop it here

I will stick to the original convention for now.

heuermh · 2016-09-28T12:27:55Z

Thanks @zyxue! I've left some review comments but they may not completely answer your questions. Hopefully others will chime in.

Note there are two unit test failures in AlignmentRecordRDDSuite, see https://amplab.cs.berkeley.edu/jenkins//job/ADAM-prb/1505/HADOOP_VERSION=2.6.0,SCALAVER=2.10,SPARK_VERSION=1.5.2,label=centos/testReport/

FastqConverterSuite.scala => FastqRecordConverterSuite.scala

zyxue · 2016-09-28T16:39:15Z

I have located the reason at least for the first test failure. As mentioned, it's due to inconsistent default values. In the original convertRead, it'ssetReadNegativeStrand(false), setting it to null, which is done in convertPair and convertFragment, would cause java.lang.NullPointerException when getReadNegativeStrand. I suggest dropping all the explicitly set null, which seems irrelevant to a fastq file, and just use default values instead, what do you think?

heuermh · 2016-09-30T14:46:16Z

@fnothaft I don't see any of your review comments. Were they lost or resolved?

fnothaft · 2016-09-29T21:39:41Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      s"Input must have 4 lines (${lines.length.toString} found):\n${input}")
+
+    val readName = lines(0).drop(1)
+    if (readName.endsWith("/1") && setSecondOfPair)


+1, I've also seen _1/2

fnothaft · 2016-09-29T21:41:41Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      else {
+        if (readQualitiesRaw == "*") "B" * readSequence.length
+        else if (readQualitiesRaw.length < readSequence.length) readQualitiesRaw + ("B" * (readSequence.length - readQualitiesRaw.length))
+        else if (readQualitiesRaw.length > readSequence.length) throw new NotImplementedError("Not implemented")


IIRC, B is the code for "unknown" quality. CC @ryan-williams

fnothaft · 2016-09-29T21:43:51Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+
+  private def parseReadPairInFastq(input: String): (String, String, String, String, String, String) = {
+    val lines = input.toString.split('\n')
+    require(lines.length == 8,


I think we should make this work for the simple case first (what we have currently implemented --> fastq record is 4 lines, interleaved read pair is 8 lines). In a follow on, we can make the arbitrary wrapping case work. In my experience, "simply" formatted files are much more common than arbitrarily formatted files.

fnothaft · 2016-09-29T21:44:16Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      secondReadName,
+      secondReadSequence,
+      secondReadQualities
+      ) = this.parseReadPairInFastq(element._2.toString)


You don't need the this. here or below.

fnothaft · 2016-09-29T21:44:26Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

-        .setSecondaryAlignment(null)
-        .setSupplementaryAlignment(null)
-        .build()
+      this.makeAlignmentRecord(firstReadName, firstReadSequence, firstReadQualities, 0),


Don't need the this..

fnothaft · 2016-09-29T21:44:35Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      secondReadName,
+      secondReadSequence,
+      secondReadQualities
+      ) = this.parseReadPairInFastq(element._2.toString)


Don't need the this..

fnothaft · 2016-09-29T21:44:49Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

      firstReadName == secondReadName,
      "Reads %s and %s in Fragment have different names.".format(
        firstReadName,
        secondReadName
      )
    )

+    val alignments = List(
+      this.makeAlignmentRecord(firstReadName, firstReadSequence, firstReadQualities, 0),


Don't need the this..

fnothaft · 2016-09-29T21:44:59Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

-    val readName = trimTrailingReadNumber(lines(0).drop(1))
-    val readSequence = lines(1)
+    val (readName, readSequence, readQualities) =
+      this.parseReadInFastq(element._2.toString, setFirstOfPair, setSecondOfPair, stringency)


Don't need the this..

fnothaft · 2016-09-29T21:45:38Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

-    recordGroupOpt.foreach(builder.setRecordGroupName)
-
-    builder.build()
+    this.makeAlignmentRecord(


Don't need the this..

I have removed all this., wondering when will this. ever be necessary?

this is only necessary if there is a collision between say a field and a local variable with the same name

zyxue · 2016-09-30T22:27:02Z

Where can I find more information on ValidationStringency.STRICT, what should the behavior be when it's STRICT or not, please?

throw IllegalArgumentException instead of NotImplementedError when quality length is longer than read length in FASTQ record under STRICT mode

zyxue · 2016-10-04T04:23:12Z

@heuermh, Can you test it again, please? If there is no further request, I think it's ready to be merged.

heuermh · 2016-10-04T14:49:38Z

Jenkins, test this please.

heuermh · 2016-10-04T15:00:24Z

Where can I find more information on ValidationStringency.STRICT, what should the behavior be when it's STRICT or not, please?

We're borrowing the enum and concept from htsjdk. Briefly, on errors strict should throw exceptions, lenient should log warnings, and silent should not complain.

AmplabJenkins · 2016-10-04T15:01:58Z

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/ADAM-prb/1516/

Build result: FAILURE

GitHub pull request #1185 of commit ce5e3a0 automatically merged.Notifying endpoint 'HTTP:https://webhooks.gitter.im/e/ac8bb6e9f53357bc8aa8'[EnvInject] - Loading node environment variables.Building remotely on amp-jenkins-worker-05 (centos spark-test) in workspace /home/jenkins/workspace/ADAM-prb > /home/jenkins/git2/bin/git rev-parse --is-inside-work-tree # timeout=10Fetching changes from the remote Git repository > /home/jenkins/git2/bin/git config remote.origin.url https://github.com/bigdatagenomics/adam.git # timeout=10Fetching upstream changes from https://github.com/bigdatagenomics/adam.git > /home/jenkins/git2/bin/git --version # timeout=10 > /home/jenkins/git2/bin/git -c core.askpass=true fetch --tags --progress https://github.com/bigdatagenomics/adam.git +refs/pull/:refs/remotes/origin/pr/ # timeout=15 > /home/jenkins/git2/bin/git rev-parse origin/pr/1185/merge^{commit} # timeout=10 > /home/jenkins/git2/bin/git branch -a --contains 5a913a0e1c5cb9874266ae2a108db3822c19c7b1 # timeout=10 > /home/jenkins/git2/bin/git rev-parse remotes/origin/pr/1185/merge^{commit} # timeout=10Checking out Revision 5a913a0e1c5cb9874266ae2a108db3822c19c7b1 (origin/pr/1185/merge) > /home/jenkins/git2/bin/git config core.sparsecheckout # timeout=10 > /home/jenkins/git2/bin/git checkout -f 5a913a0e1c5cb9874266ae2a108db3822c19c7b1First time build. Skipping changelog.Triggering ADAM-prb ? 2.6.0,2.11,1.5.2,centosTriggering ADAM-prb ? 2.6.0,2.10,1.5.2,centosTouchstone configurations resulted in FAILURE, so aborting...Notifying endpoint 'HTTP:https://webhooks.gitter.im/e/ac8bb6e9f53357bc8aa8'
Test FAILed.

heuermh · 2016-10-05T13:48:27Z

Jenkins, retest this please

AmplabJenkins · 2016-10-05T14:14:41Z

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/ADAM-prb/1520/
Test PASSed.

fnothaft · 2016-10-07T05:54:19Z

LGTM! Ping @heuermh for a review pass.
I'm thinking that we should merge this manually, since the commit history is pretty long. @heuermh let me know if/when it looks good to you, and I will merge it.

fnothaft · 2016-10-07T23:16:48Z

@heuermh is this still pending changes from your side?

heuermh · 2016-10-08T02:02:03Z

Will complete review on Monday

heuermh · 2016-10-13T12:30:20Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

@@ -41,6 +41,114 @@ import scala.collection.JavaConversions._
 private[adam] class FastqRecordConverter extends Serializable with Logging {

  /**
+   * Parse 4 lines at a time


This doc comment doesn't match the method.

Perhaps something like Return true if the read name suffix and flags match.

heuermh · 2016-10-13T12:32:17Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+    val match2 = secondReadSuffix.findAllIn(readName)
+
+    if (match1.nonEmpty && isSecondOfPair)
+      throw new IllegalArgumentException(


These exceptions are thrown without considering ValidationStringency. If the stringency is lenient or silent, is it possible to continue processing?

heuermh · 2016-10-13T12:33:53Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+    val readName = lines(0).drop(1)
+    if (setFirstOfPair || setSecondOfPair) readNameSuffixAndIndexOfPairMustMatch(readName, setFirstOfPair)
+
+    val suffix = """([/ +_]1$)|([/ +_]2$)""".r


I wonder if this regex and those above on line 50 and 51 might be combined as static private fields so that the two methods don't get out of sync

heuermh · 2016-10-13T12:36:14Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

-    val readName = trimTrailingReadNumber(lines(0).drop(1))
-    val readSequence = lines(1)
+    if (setFirstOfPair && setSecondOfPair)
+      throw new IllegalArgumentException("setFirstOfPair and setSecondOfPair cannot be true at the same time")


This exception is also thrown without considering ValidationStringency. If the stringency is lenient or silent, is it possible to continue processing?

heuermh · 2016-10-13T12:38:59Z

adam-core/src/main/scala/org/bdgenomics/adam/converters/FastqRecordConverter.scala

+      .setReadPaired(readPaired)
+      .setReadInFragment(readInFragment)
+
+    if (recordGroupOpt != None)


With opt.foreach you don't also need to check against None

fnothaft · 2016-10-13T22:27:39Z

Moved over to #1208.

zyxue added 19 commits September 27, 2016 18:16

added FastqConverterSuite.scala

8065a64

tested FastqRecordConverter.convertPair

refactored FastqRecordConverter.convertPair

ede941e

improved exception message

6597b63

when fastq lines number doesn't match

added tests for invalid fastq input

a944139

added parseReadPairInFastq private method

f2736e3

Refactored out the common part between FastqRecordConverter.convertPair & FastqRecordConverter.convertFragment

updated a testcase for FastqRecordConverter.convertPair

1b92e6e

removed assignment for tests with invalid input

eae9ca7

added 3 tests for FastqRecordConverter.convertFragment

0bff9dc

refactored FastqRecordConverter.convertFragment

790ea31

improved format of makeAlignmentRecord in convertPair

a931187

test convertFragment in more detail

914190f

factored out makeAlignmentRecord

07e6c78

shared by convertPair & convertFragment

use string Interpolation for exception message

8e7c029

refactored out parseReadInFastq

f604a9b

updated parseReadInFastq with param stringency

48c0f40

factoring out common code from convertRead

refactored convertRead

ec95050

There is still quite a bit of problem with default values when creating a AlignmentRecord instance from a Fastq entry

added tests for convertRead

63013b8

updated default values in tests

7782ab3

[ADAM-1172] formatting and run format-source

d4c5ad6

bigdatagenomics#1172

heuermh requested changes Sep 28, 2016

View reviewed changes

renaming a test suite

2c055a8

FastqConverterSuite.scala => FastqRecordConverterSuite.scala

zyxue added 4 commits September 28, 2016 09:48

drop explicitly set null in makeAlignmentRecord

f124fa5

changed default setFirstOfPair to false

bb3f26b

remove null value assertions in tests

2c4f7cb

test default getReadPaired in convertRead

6194ad4

fnothaft requested changes Sep 30, 2016

View reviewed changes

zyxue added 10 commits September 30, 2016 15:32

make methods private to package only

4d9b2f6

change exception raised

449a517

throw IllegalArgumentException instead of NotImplementedError when quality length is longer than read length in FASTQ record under STRICT mode

test read length > quality length lenient mode

698f015

replaced ValidationStringency.LENIENT with lenient

14c654a

added test case when no read quality is *

3ac85d6

enhanced read suffix removal

a797e26

added method readNameSuffixAndIndexOfPairMustMatch

d11edf8

refactored readNameSuffixAndIndexOfPairMustMatch

45d5150

removed unnecessary this.

e35648a

simplified requirement()

ce5e3a0

rerun ./scripts/format-source

272f637

fnothaft approved these changes Oct 7, 2016

View reviewed changes

heuermh requested changes Oct 13, 2016

View reviewed changes

heuermh modified the milestone: 0.20.0 Oct 13, 2016

fnothaft mentioned this pull request Oct 13, 2016

Fastq record converter #1208

Closed

fnothaft closed this Oct 13, 2016

Fastq record converter #1185

Fastq record converter #1185

Conversation

zyxue commented Sep 28, 2016

AmplabJenkins commented Sep 28, 2016

heuermh commented Sep 28, 2016

AmplabJenkins commented Sep 28, 2016

Build result: FAILURE

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

zyxue Oct 1, 2016 • edited Loading

Choose a reason for hiding this comment

heuermh Sep 28, 2016 • 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

zyxue Oct 1, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zyxue Sep 28, 2016 • 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

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

zyxue Sep 28, 2016 • 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

heuermh commented Sep 28, 2016 • edited Loading

zyxue commented Sep 28, 2016 • edited Loading

heuermh commented Sep 30, 2016

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

zyxue commented Sep 30, 2016

zyxue commented Oct 4, 2016

heuermh commented Oct 4, 2016

heuermh commented Oct 4, 2016

AmplabJenkins commented Oct 4, 2016

Build result: FAILURE

heuermh commented Oct 5, 2016

AmplabJenkins commented Oct 5, 2016

fnothaft commented Oct 7, 2016

fnothaft commented Oct 7, 2016

heuermh commented Oct 8, 2016

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

fnothaft commented Oct 13, 2016

zyxue Oct 1, 2016 •

edited

Loading

heuermh Sep 28, 2016 •

edited

Loading

zyxue Oct 1, 2016 •

edited

Loading

zyxue Sep 28, 2016 •

edited

Loading

zyxue Sep 28, 2016 •

edited

Loading

heuermh commented Sep 28, 2016 •

edited

Loading

zyxue commented Sep 28, 2016 •

edited

Loading