[SPARK-18118][SQL] fix a compilation error due to nested JavaBeans

[kiszk]

What changes were proposed in this pull request?

This PR avoids a compilation error due to more than 64KB Java byte code size. This error occur since generated java code SpecificSafeProjection.apply() for nested JavaBeans is too big. This PR avoids this compilation error by splitting a big code chunk into multiple methods by calling CodegenContext.splitExpression at InitializeJavaBean.doGenCode
An object reference for JavaBean is stored to an instance variable javaBean.... Then, the instance variable will be referenced in the split methods.

Generated code with this PR

/* 22098 */   private void apply130_0(InternalRow i) {
...
/* 22125 */     boolean isNull238 = i.isNullAt(2);
/* 22126 */     InternalRow value238 = isNull238 ? null : (i.getStruct(2, 3));
/* 22127 */     boolean isNull236 = false;
/* 22128 */     test.org.apache.spark.sql.JavaDatasetSuite$Nesting1 value236 = null;
/* 22129 */     if (!false && isNull238) {
/* 22130 */
/* 22131 */       final test.org.apache.spark.sql.JavaDatasetSuite$Nesting1 value239 = null;
/* 22132 */       isNull236 = true;
/* 22133 */       value236 = value239;
/* 22134 */     } else {
/* 22135 */
/* 22136 */       final test.org.apache.spark.sql.JavaDatasetSuite$Nesting1 value241 = false ? null : new test.org.apache.spark.sql.JavaDatasetSuite$Nesting1();
/* 22137 */       this.javaBean14 = value241;
/* 22138 */       if (!false) {
/* 22139 */         apply25_0(i);
/* 22140 */         apply25_1(i);
/* 22141 */         apply25_2(i);
/* 22142 */       }
/* 22143 */       isNull236 = false;
/* 22144 */       value236 = value241;
/* 22145 */     }
/* 22146 */     this.javaBean.setField2(value236);
/* 22147 */
/* 22148 */   }
...
/* 22928 */   public java.lang.Object apply(java.lang.Object _i) {
/* 22929 */     InternalRow i = (InternalRow) _i;
/* 22930 */
/* 22931 */     final test.org.apache.spark.sql.JavaDatasetSuite$NestedComplicatedJavaBean value1 = false ? null : new test.org.apache.spark.sql.JavaDatasetSuite$NestedComplicatedJavaBean();
/* 22932 */     this.javaBean = value1;
/* 22933 */     if (!false) {
/* 22934 */       apply130_0(i);
/* 22935 */       apply130_1(i);
/* 22936 */       apply130_2(i);
/* 22937 */       apply130_3(i);
/* 22938 */       apply130_4(i);
/* 22939 */     }
/* 22940 */     if (false) {
/* 22941 */       mutableRow.setNullAt(0);
/* 22942 */     } else {
/* 22943 */
/* 22944 */       mutableRow.update(0, value1);
/* 22945 */     }
/* 22946 */
/* 22947 */     return mutableRow;
/* 22948 */   }

How was this patch tested?

added a test suite into JavaDatasetSuite.java

[SparkQA]

Test build #69240 has finished for PR 16032 at commit 5debc84.

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

[hvanhovell]

LGTM - merging to master/2.1/2.0. Thanks!

[cloud-fan]

I think it's safer to avoid using this in generated code, as the code snippet may be used in an inner class, and we already guarantee the variable name is unique globally.

[hvanhovell]

Ok, I'll touch that up.

[hvanhovell]

Done.