[ https://issues.apache.org/jira/browse/SPARK-12045?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15046056#comment-15046056 ]
Jeff Zhang commented on SPARK-12045: ------------------------------------ bq. Our general policy for exceptions is that we return null for data dependent problems (i.e. a date string that doesn't parse) Should we provide an option for users whether to return null or throw exception ? And at least we should provide document to highlight that if return null. Or maybe use some built-in accumulator to be displayed as warning in client side, just like the counter in MR. > Use joda's DateTime to replace Calendar > --------------------------------------- > > Key: SPARK-12045 > URL: https://issues.apache.org/jira/browse/SPARK-12045 > Project: Spark > Issue Type: Improvement > Components: SQL > Affects Versions: 1.5.2 > Reporter: Jeff Zhang > > Currently spark use Calendar to build the Date when convert from string to > Date. But Calendar can not detect the invalid date format (e.g. 2011-02-29). > Although we can use Calendar.setLenient(false) to enable Calendar to detect > the invalid date format, but found the error message very confusing. So I > suggest to use joda's DateTime to replace Calendar. > Besides that, I found that there's already some format checking logic when > casting string to date. And if it is invalid format, it would return None. I > don't think it make sense to just return None without telling users. I think > by default should just throw exception, and user can set property to allow it > return None if invalid format. > {code} > if (i == 0 && j != 4) { > // year should have exact four digits > return None > } > {code} -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org