losipiuk commented on a change in pull request #2040:
URL: https://github.com/apache/hive/pull/2040#discussion_r589363045
##########
File path:
ql/src/java/org/apache/hadoop/hive/ql/io/avro/AvroGenericRecordReader.java
##########
@@ -88,22 +88,38 @@ public AvroGenericRecordReader(JobConf job, FileSplit
split, Reporter reporter)
gdr.setExpected(latest);
}
- if (split.getLength() == 0) {
- this.isEmptyInput = true;
- this.start = 0;
- this.reader = null;
+ org.apache.avro.file.FileReader<GenericRecord> nonFinalReader = null;
+ try {
+ if (split.getLength() == 0) {
+ this.isEmptyInput = true;
+ this.start = 0;
+ this.reader = null;
+ } else {
+ this.isEmptyInput = false;
+ nonFinalReader = new DataFileReader<GenericRecord>(new
FsInput(split.getPath(), job), gdr);
+ this.reader = nonFinalReader;
+ this.reader.sync(split.getStart());
+ this.start = reader.tell();
Review comment:
We could drop `nonFinalReader` if we just wrapped two `extract*` calls
with `try..catch` like this:
```java
if (split.getLength() == 0) {
this.isEmptyInput = true;
this.start = 0;
this.reader = null;
} else {
this.isEmptyInput = false;
this.reader = new DataFileReader<GenericRecord>(new
FsInput(split.getPath(), job), gdr);
this.reader.sync(split.getStart());
this.start = reader.tell();
}
this.stop = split.getStart() + split.getLength();
this.recordReaderID = new UID();
try {
this.writerTimezone = extractWriterTimezoneFromMetadata(job, split,
gdr);
this.writerProleptic = extractWriterProlepticFromMetadata(job, split,
gdr);
} catch (Exception e) {
if (this.reader != null) {
try {
this.reader.close();
} catch (Exception closeException) {
if (closeException != e) {
e.addSuppressed(closeException);
}
}
}
throw e;
}
```
But then we can still have a `close`-leak if `this.reader.sync()` or
`this.reader.tell()` throws an exception.
And If I want to wrap everything with `try...catch` I need extra variable.
Specifically this (use `this.reader` instead `nonFinalReader`):
```java
try {
if (split.getLength() == 0) {
this.isEmptyInput = true;
this.start = 0;
this.reader = null;
} else {
this.isEmptyInput = false;
this.reader = new DataFileReader<GenericRecord>(new
FsInput(split.getPath(), job), gdr);
this.reader.sync(split.getStart());
this.start = reader.tell();
}
this.stop = split.getStart() + split.getLength();
this.recordReaderID = new UID();
this.writerTimezone = extractWriterTimezoneFromMetadata(job, split,
gdr);
this.writerProleptic = extractWriterProlepticFromMetadata(job, split,
gdr);
} catch (Exception e) {
if (this.reader != null) {
try {
this.reader.close();
} catch (Exception closeException) {
if (closeException != e) {
e.addSuppressed(closeException);
}
}
}
throw e;
}
}
```
does not work because compiler complains that `this.reader` in `catch`
clause can be not initialized
And this (initialize `this.reader` to null before `if`):
```java
this.reader = null;
try {
if (split.getLength() == 0) {
this.isEmptyInput = true;
this.start = 0;
this.reader = null;
} else {
this.isEmptyInput = false;
this.reader = new DataFileReader<GenericRecord>(new
FsInput(split.getPath(), job), gdr);
this.reader.sync(split.getStart());
this.start = reader.tell();
}
this.stop = split.getStart() + split.getLength();
this.recordReaderID = new UID();
this.writerTimezone = extractWriterTimezoneFromMetadata(job, split,
gdr);
this.writerProleptic = extractWriterProlepticFromMetadata(job, split,
gdr);
} catch (Exception e) {
if (this.reader != null) {
try {
this.reader.close();
} catch (Exception closeException) {
if (closeException != e) {
e.addSuppressed(closeException);
}
}
}
throw e;
}
}
```
does not work because `this.reader` is assigned more than once.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
