Hi All, I'm working on adding support of Snappy files read ( https://issues.apache.org/jira/browse/FLINK-5944) and would like to get your opinion about one question
As you can see from the ticket description it's desired to support both Java Snappy (xerial) and Hadoop Snappy codecs which are incompatible. The thing is that desired codec for InputFormat is selected based on file extension (e.g. '.gzip' or '.snappy'). So the question is how we can distinguish whether the Hadoop Snappy codec or Java Snappy codec is needed. I can propose the following options: 1. Add new config option to flink-conf.yaml like fs.hadoop-snappy and select InputStreamFactory based on this option 2. Add flag parameter to API method readTextFile whether the file is Hadoop Snappy 3. Add separate API method for reading snappy-compressed files 4. Ask users to use '.snappy' extension for Java Snappy and some other extension like '.hsnappy' for Hadoop Snappy >From my point of view option 1. seems more natural. Please let me know your opinion Thanks, Mikhail
