Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
Yes, but I think we would pretty much have to do that. I don't think we can stop doing 2.11 releases. > On 8. Oct 2018, at 15:37, Chesnay Schepler wrote: > > The infrastructure would only be required if we opt for releasing 2.11 and > 2.12 builds simultaneously, correct? > > On 08.10.2018 15:04, Aljoscha Krettek wrote: >> Breaking the API (or not breaking it but requiring explicit types when using >> Scala 2.12) and the Maven infrastructure to actually build a 2.12 release. >> >>> On 8. Oct 2018, at 13:00, Chesnay Schepler wrote: >>> >>> And the remaining parts would only be about breaking the API? >>> >>> On 08.10.2018 12:24, Aljoscha Krettek wrote: I have an open PR that does everything we can do for preparing the code base for Scala 2.12 without breaking the API: https://github.com/apache/flink/pull/6784 > On 8. Oct 2018, at 09:56, Chesnay Schepler wrote: > > I'd rather not maintain 2 master branches. Beyond the maintenance > overhead I'm > wondering about the benefit, as the API break still has to happen at some > point. > > @Aljoscha how much work for supporting scala 2.12 can be merged without > breaking the API? > If this is the only blocker I suggest to make the breaking change in 1.8. > > On 05.10.2018 10:31, Till Rohrmann wrote: >> Thanks Aljoscha for starting this discussion. The described problem >> brings >> us indeed a bit into a pickle. Even with option 1) I think it is somewhat >> API breaking because everyone who used lambdas without types needs to add >> them now. Consequently, I only see two real options out of the ones >> you've >> proposed: >> >> 1) Disambiguate the API (either by removing >> reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) >> 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely >> out >> >> Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit >> problematic because then all Scala API users who have implemented a >> GroupReduceFunction need to convert it into a Scala lambda. Moreover, I >> think it will be problematic with RichGroupReduceFunction which you need >> to >> get access to the RuntimeContext. >> >> Maintaining two master branches puts a lot of burden onto the developers >> to >> always keep the two branches in sync. Ideally I would like to avoid this. >> >> I also played a little bit around with implicit conversions to add the >> lambda methods in Scala 2.11 on demand, but I was not able to get it work >> smoothly. >> >> I'm cross posting this thread to user as well to get some more user >> feedback. >> >> Cheers, >> Till >> >> On Thu, Oct 4, 2018 at 7:36 PM Elias Levy >> wrote: >> >>> The second alternative, with the addition of methods that take functions >>> with Scala types, seems the most sensible. I wonder if there is a need >>> then to maintain the *J Java parameter methods, or whether users could >>> just >>> access the functionality by converting the Scala DataStreams to Java via >>> .javaStream and whatever the equivalent is for DataSets. >>> >>> On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek >>> wrote: >>> Hi, I'm currently working on >>> https://issues.apache.org/jira/browse/FLINK-7811, with the goal of adding support for Scala 2.12. There is a bit of a >>> hurdle and I have to explain some context first. With Scala 2.12, lambdas are implemented using the lambda mechanism of Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This means that the following two method definitions can both take a lambda: def map[R](mapper: MapFunction[T, R]): DataSet[R] def map[R](fun: T => R): DataSet[R] The Scala compiler gives precedence to the lambda version when you call map() with a lambda in simple cases, so it works here. You could still >>> call map() with a lambda if the lambda version of the method weren't here because they are now considered the same. For Scala 2.11 we need both signatures, though, to allow calling with a lambda and with a >>> MapFunction. The problem is with more complicated method signatures, like: def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): DataSet[R] def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] (for reference, GroupReduceFunction is a SAM with void reduce(java.lang.Iterable values, Collector out)) These two signatures are not the same but similar enough for the Scala 2.12 compiler to "get confused". In Scala 2.11, I could call >>>
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
The infrastructure would only be required if we opt for releasing 2.11 and 2.12 builds simultaneously, correct? On 08.10.2018 15:04, Aljoscha Krettek wrote: Breaking the API (or not breaking it but requiring explicit types when using Scala 2.12) and the Maven infrastructure to actually build a 2.12 release. On 8. Oct 2018, at 13:00, Chesnay Schepler wrote: And the remaining parts would only be about breaking the API? On 08.10.2018 12:24, Aljoscha Krettek wrote: I have an open PR that does everything we can do for preparing the code base for Scala 2.12 without breaking the API: https://github.com/apache/flink/pull/6784 On 8. Oct 2018, at 09:56, Chesnay Schepler wrote: I'd rather not maintain 2 master branches. Beyond the maintenance overhead I'm wondering about the benefit, as the API break still has to happen at some point. @Aljoscha how much work for supporting scala 2.12 can be merged without breaking the API? If this is the only blocker I suggest to make the breaking change in 1.8. On 05.10.2018 10:31, Till Rohrmann wrote: Thanks Aljoscha for starting this discussion. The described problem brings us indeed a bit into a pickle. Even with option 1) I think it is somewhat API breaking because everyone who used lambdas without types needs to add them now. Consequently, I only see two real options out of the ones you've proposed: 1) Disambiguate the API (either by removing reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit problematic because then all Scala API users who have implemented a GroupReduceFunction need to convert it into a Scala lambda. Moreover, I think it will be problematic with RichGroupReduceFunction which you need to get access to the RuntimeContext. Maintaining two master branches puts a lot of burden onto the developers to always keep the two branches in sync. Ideally I would like to avoid this. I also played a little bit around with implicit conversions to add the lambda methods in Scala 2.11 on demand, but I was not able to get it work smoothly. I'm cross posting this thread to user as well to get some more user feedback. Cheers, Till On Thu, Oct 4, 2018 at 7:36 PM Elias Levy wrote: The second alternative, with the addition of methods that take functions with Scala types, seems the most sensible. I wonder if there is a need then to maintain the *J Java parameter methods, or whether users could just access the functionality by converting the Scala DataStreams to Java via .javaStream and whatever the equivalent is for DataSets. On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek wrote: Hi, I'm currently working on https://issues.apache.org/jira/browse/FLINK-7811, with the goal of adding support for Scala 2.12. There is a bit of a hurdle and I have to explain some context first. With Scala 2.12, lambdas are implemented using the lambda mechanism of Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This means that the following two method definitions can both take a lambda: def map[R](mapper: MapFunction[T, R]): DataSet[R] def map[R](fun: T => R): DataSet[R] The Scala compiler gives precedence to the lambda version when you call map() with a lambda in simple cases, so it works here. You could still call map() with a lambda if the lambda version of the method weren't here because they are now considered the same. For Scala 2.11 we need both signatures, though, to allow calling with a lambda and with a MapFunction. The problem is with more complicated method signatures, like: def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): DataSet[R] def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] (for reference, GroupReduceFunction is a SAM with void reduce(java.lang.Iterable values, Collector out)) These two signatures are not the same but similar enough for the Scala 2.12 compiler to "get confused". In Scala 2.11, I could call reduceGroup() with a lambda that doesn't have parameter type definitions and things would be fine. With Scala 2.12 I can't do that because the compiler can't figure out which method to call and requires explicit type definitions on the lambda parameters. I see some solutions for this: 1. Keep the methods as is, this would force people to always explicitly specify parameter types on their lambdas. 2. Rename the second method to reduceGroupJ() to signal that it takes a user function that takes Java-style interfaces (the first parameter is java.lang.Iterable while the Scala lambda takes a scala.Iterator). This disambiguates the code, users can use lambdas without specifying explicit parameter types but breaks the API. One effect of 2. would be that we can add a reduceGroup() method that takes a api.scala.GroupReduceFunction that takes proper Scala types, thus it would allow people to implement user functions without
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
Breaking the API (or not breaking it but requiring explicit types when using Scala 2.12) and the Maven infrastructure to actually build a 2.12 release. > On 8. Oct 2018, at 13:00, Chesnay Schepler wrote: > > And the remaining parts would only be about breaking the API? > > On 08.10.2018 12:24, Aljoscha Krettek wrote: >> I have an open PR that does everything we can do for preparing the code base >> for Scala 2.12 without breaking the API: >> https://github.com/apache/flink/pull/6784 >> >>> On 8. Oct 2018, at 09:56, Chesnay Schepler wrote: >>> >>> I'd rather not maintain 2 master branches. Beyond the maintenance overhead >>> I'm >>> wondering about the benefit, as the API break still has to happen at some >>> point. >>> >>> @Aljoscha how much work for supporting scala 2.12 can be merged without >>> breaking the API? >>> If this is the only blocker I suggest to make the breaking change in 1.8. >>> >>> On 05.10.2018 10:31, Till Rohrmann wrote: Thanks Aljoscha for starting this discussion. The described problem brings us indeed a bit into a pickle. Even with option 1) I think it is somewhat API breaking because everyone who used lambdas without types needs to add them now. Consequently, I only see two real options out of the ones you've proposed: 1) Disambiguate the API (either by removing reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit problematic because then all Scala API users who have implemented a GroupReduceFunction need to convert it into a Scala lambda. Moreover, I think it will be problematic with RichGroupReduceFunction which you need to get access to the RuntimeContext. Maintaining two master branches puts a lot of burden onto the developers to always keep the two branches in sync. Ideally I would like to avoid this. I also played a little bit around with implicit conversions to add the lambda methods in Scala 2.11 on demand, but I was not able to get it work smoothly. I'm cross posting this thread to user as well to get some more user feedback. Cheers, Till On Thu, Oct 4, 2018 at 7:36 PM Elias Levy wrote: > The second alternative, with the addition of methods that take functions > with Scala types, seems the most sensible. I wonder if there is a need > then to maintain the *J Java parameter methods, or whether users could > just > access the functionality by converting the Scala DataStreams to Java via > .javaStream and whatever the equivalent is for DataSets. > > On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek > wrote: > >> Hi, >> >> I'm currently working on > https://issues.apache.org/jira/browse/FLINK-7811, >> with the goal of adding support for Scala 2.12. There is a bit of a > hurdle >> and I have to explain some context first. >> >> With Scala 2.12, lambdas are implemented using the lambda mechanism of >> Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This >> means that the following two method definitions can both take a lambda: >> >> def map[R](mapper: MapFunction[T, R]): DataSet[R] >> def map[R](fun: T => R): DataSet[R] >> >> The Scala compiler gives precedence to the lambda version when you call >> map() with a lambda in simple cases, so it works here. You could still > call >> map() with a lambda if the lambda version of the method weren't here >> because they are now considered the same. For Scala 2.11 we need both >> signatures, though, to allow calling with a lambda and with a > MapFunction. >> The problem is with more complicated method signatures, like: >> >> def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): >> DataSet[R] >> >> def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] >> >> (for reference, GroupReduceFunction is a SAM with void >> reduce(java.lang.Iterable values, Collector out)) >> >> These two signatures are not the same but similar enough for the Scala >> 2.12 compiler to "get confused". In Scala 2.11, I could call > reduceGroup() >> with a lambda that doesn't have parameter type definitions and things > would >> be fine. With Scala 2.12 I can't do that because the compiler can't > figure >> out which method to call and requires explicit type definitions on the >> lambda parameters. >> >> I see some solutions for this: >> >> 1. Keep the methods as is, this would force people to always explicitly >> specify parameter types on their lambdas. >> >> 2. Rename the second method to reduceGroupJ() to signal that it takes a >>
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
And the remaining parts would only be about breaking the API? On 08.10.2018 12:24, Aljoscha Krettek wrote: I have an open PR that does everything we can do for preparing the code base for Scala 2.12 without breaking the API: https://github.com/apache/flink/pull/6784 On 8. Oct 2018, at 09:56, Chesnay Schepler wrote: I'd rather not maintain 2 master branches. Beyond the maintenance overhead I'm wondering about the benefit, as the API break still has to happen at some point. @Aljoscha how much work for supporting scala 2.12 can be merged without breaking the API? If this is the only blocker I suggest to make the breaking change in 1.8. On 05.10.2018 10:31, Till Rohrmann wrote: Thanks Aljoscha for starting this discussion. The described problem brings us indeed a bit into a pickle. Even with option 1) I think it is somewhat API breaking because everyone who used lambdas without types needs to add them now. Consequently, I only see two real options out of the ones you've proposed: 1) Disambiguate the API (either by removing reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit problematic because then all Scala API users who have implemented a GroupReduceFunction need to convert it into a Scala lambda. Moreover, I think it will be problematic with RichGroupReduceFunction which you need to get access to the RuntimeContext. Maintaining two master branches puts a lot of burden onto the developers to always keep the two branches in sync. Ideally I would like to avoid this. I also played a little bit around with implicit conversions to add the lambda methods in Scala 2.11 on demand, but I was not able to get it work smoothly. I'm cross posting this thread to user as well to get some more user feedback. Cheers, Till On Thu, Oct 4, 2018 at 7:36 PM Elias Levy wrote: The second alternative, with the addition of methods that take functions with Scala types, seems the most sensible. I wonder if there is a need then to maintain the *J Java parameter methods, or whether users could just access the functionality by converting the Scala DataStreams to Java via .javaStream and whatever the equivalent is for DataSets. On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek wrote: Hi, I'm currently working on https://issues.apache.org/jira/browse/FLINK-7811, with the goal of adding support for Scala 2.12. There is a bit of a hurdle and I have to explain some context first. With Scala 2.12, lambdas are implemented using the lambda mechanism of Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This means that the following two method definitions can both take a lambda: def map[R](mapper: MapFunction[T, R]): DataSet[R] def map[R](fun: T => R): DataSet[R] The Scala compiler gives precedence to the lambda version when you call map() with a lambda in simple cases, so it works here. You could still call map() with a lambda if the lambda version of the method weren't here because they are now considered the same. For Scala 2.11 we need both signatures, though, to allow calling with a lambda and with a MapFunction. The problem is with more complicated method signatures, like: def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): DataSet[R] def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] (for reference, GroupReduceFunction is a SAM with void reduce(java.lang.Iterable values, Collector out)) These two signatures are not the same but similar enough for the Scala 2.12 compiler to "get confused". In Scala 2.11, I could call reduceGroup() with a lambda that doesn't have parameter type definitions and things would be fine. With Scala 2.12 I can't do that because the compiler can't figure out which method to call and requires explicit type definitions on the lambda parameters. I see some solutions for this: 1. Keep the methods as is, this would force people to always explicitly specify parameter types on their lambdas. 2. Rename the second method to reduceGroupJ() to signal that it takes a user function that takes Java-style interfaces (the first parameter is java.lang.Iterable while the Scala lambda takes a scala.Iterator). This disambiguates the code, users can use lambdas without specifying explicit parameter types but breaks the API. One effect of 2. would be that we can add a reduceGroup() method that takes a api.scala.GroupReduceFunction that takes proper Scala types, thus it would allow people to implement user functions without having to cast the various Iterator/Iterable parameters. Either way, people would have to adapt their code when moving to Scala 2.12 in some way, depending on what style of methods they use. There is also solution 2.5: 2.5 Rename the methods only in the Scala 2.12 build of Flink and keep the old method names for Scala 2.11. This would require some infrastructure
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
I have an open PR that does everything we can do for preparing the code base for Scala 2.12 without breaking the API: https://github.com/apache/flink/pull/6784 > On 8. Oct 2018, at 09:56, Chesnay Schepler wrote: > > I'd rather not maintain 2 master branches. Beyond the maintenance overhead I'm > wondering about the benefit, as the API break still has to happen at some > point. > > @Aljoscha how much work for supporting scala 2.12 can be merged without > breaking the API? > If this is the only blocker I suggest to make the breaking change in 1.8. > > On 05.10.2018 10:31, Till Rohrmann wrote: >> Thanks Aljoscha for starting this discussion. The described problem brings >> us indeed a bit into a pickle. Even with option 1) I think it is somewhat >> API breaking because everyone who used lambdas without types needs to add >> them now. Consequently, I only see two real options out of the ones you've >> proposed: >> >> 1) Disambiguate the API (either by removing >> reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) >> 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out >> >> Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit >> problematic because then all Scala API users who have implemented a >> GroupReduceFunction need to convert it into a Scala lambda. Moreover, I >> think it will be problematic with RichGroupReduceFunction which you need to >> get access to the RuntimeContext. >> >> Maintaining two master branches puts a lot of burden onto the developers to >> always keep the two branches in sync. Ideally I would like to avoid this. >> >> I also played a little bit around with implicit conversions to add the >> lambda methods in Scala 2.11 on demand, but I was not able to get it work >> smoothly. >> >> I'm cross posting this thread to user as well to get some more user >> feedback. >> >> Cheers, >> Till >> >> On Thu, Oct 4, 2018 at 7:36 PM Elias Levy >> wrote: >> >>> The second alternative, with the addition of methods that take functions >>> with Scala types, seems the most sensible. I wonder if there is a need >>> then to maintain the *J Java parameter methods, or whether users could just >>> access the functionality by converting the Scala DataStreams to Java via >>> .javaStream and whatever the equivalent is for DataSets. >>> >>> On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek >>> wrote: >>> Hi, I'm currently working on >>> https://issues.apache.org/jira/browse/FLINK-7811, with the goal of adding support for Scala 2.12. There is a bit of a >>> hurdle and I have to explain some context first. With Scala 2.12, lambdas are implemented using the lambda mechanism of Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This means that the following two method definitions can both take a lambda: def map[R](mapper: MapFunction[T, R]): DataSet[R] def map[R](fun: T => R): DataSet[R] The Scala compiler gives precedence to the lambda version when you call map() with a lambda in simple cases, so it works here. You could still >>> call map() with a lambda if the lambda version of the method weren't here because they are now considered the same. For Scala 2.11 we need both signatures, though, to allow calling with a lambda and with a >>> MapFunction. The problem is with more complicated method signatures, like: def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): DataSet[R] def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] (for reference, GroupReduceFunction is a SAM with void reduce(java.lang.Iterable values, Collector out)) These two signatures are not the same but similar enough for the Scala 2.12 compiler to "get confused". In Scala 2.11, I could call >>> reduceGroup() with a lambda that doesn't have parameter type definitions and things >>> would be fine. With Scala 2.12 I can't do that because the compiler can't >>> figure out which method to call and requires explicit type definitions on the lambda parameters. I see some solutions for this: 1. Keep the methods as is, this would force people to always explicitly specify parameter types on their lambdas. 2. Rename the second method to reduceGroupJ() to signal that it takes a user function that takes Java-style interfaces (the first parameter is java.lang.Iterable while the Scala lambda takes a scala.Iterator). This disambiguates the code, users can use lambdas without specifying explicit parameter types but breaks the API. One effect of 2. would be that we can add a reduceGroup() method that takes a api.scala.GroupReduceFunction that takes proper Scala types, thus it would allow people to implement user functions without having to cast the various Iterator/Iterable parameters.
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
I'd rather not maintain 2 master branches. Beyond the maintenance overhead I'm wondering about the benefit, as the API break still has to happen at some point. @Aljoscha how much work for supporting scala 2.12 can be merged without breaking the API? If this is the only blocker I suggest to make the breaking change in 1.8. On 05.10.2018 10:31, Till Rohrmann wrote: Thanks Aljoscha for starting this discussion. The described problem brings us indeed a bit into a pickle. Even with option 1) I think it is somewhat API breaking because everyone who used lambdas without types needs to add them now. Consequently, I only see two real options out of the ones you've proposed: 1) Disambiguate the API (either by removing reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit problematic because then all Scala API users who have implemented a GroupReduceFunction need to convert it into a Scala lambda. Moreover, I think it will be problematic with RichGroupReduceFunction which you need to get access to the RuntimeContext. Maintaining two master branches puts a lot of burden onto the developers to always keep the two branches in sync. Ideally I would like to avoid this. I also played a little bit around with implicit conversions to add the lambda methods in Scala 2.11 on demand, but I was not able to get it work smoothly. I'm cross posting this thread to user as well to get some more user feedback. Cheers, Till On Thu, Oct 4, 2018 at 7:36 PM Elias Levy wrote: The second alternative, with the addition of methods that take functions with Scala types, seems the most sensible. I wonder if there is a need then to maintain the *J Java parameter methods, or whether users could just access the functionality by converting the Scala DataStreams to Java via .javaStream and whatever the equivalent is for DataSets. On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek wrote: Hi, I'm currently working on https://issues.apache.org/jira/browse/FLINK-7811, with the goal of adding support for Scala 2.12. There is a bit of a hurdle and I have to explain some context first. With Scala 2.12, lambdas are implemented using the lambda mechanism of Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This means that the following two method definitions can both take a lambda: def map[R](mapper: MapFunction[T, R]): DataSet[R] def map[R](fun: T => R): DataSet[R] The Scala compiler gives precedence to the lambda version when you call map() with a lambda in simple cases, so it works here. You could still call map() with a lambda if the lambda version of the method weren't here because they are now considered the same. For Scala 2.11 we need both signatures, though, to allow calling with a lambda and with a MapFunction. The problem is with more complicated method signatures, like: def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): DataSet[R] def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] (for reference, GroupReduceFunction is a SAM with void reduce(java.lang.Iterable values, Collector out)) These two signatures are not the same but similar enough for the Scala 2.12 compiler to "get confused". In Scala 2.11, I could call reduceGroup() with a lambda that doesn't have parameter type definitions and things would be fine. With Scala 2.12 I can't do that because the compiler can't figure out which method to call and requires explicit type definitions on the lambda parameters. I see some solutions for this: 1. Keep the methods as is, this would force people to always explicitly specify parameter types on their lambdas. 2. Rename the second method to reduceGroupJ() to signal that it takes a user function that takes Java-style interfaces (the first parameter is java.lang.Iterable while the Scala lambda takes a scala.Iterator). This disambiguates the code, users can use lambdas without specifying explicit parameter types but breaks the API. One effect of 2. would be that we can add a reduceGroup() method that takes a api.scala.GroupReduceFunction that takes proper Scala types, thus it would allow people to implement user functions without having to cast the various Iterator/Iterable parameters. Either way, people would have to adapt their code when moving to Scala 2.12 in some way, depending on what style of methods they use. There is also solution 2.5: 2.5 Rename the methods only in the Scala 2.12 build of Flink and keep the old method names for Scala 2.11. This would require some infrastructure and I don't yet know how it can be done in a sane way. What do you think? I personally would be in favour of 2. but it breaks the existing API. Best, Aljoscha
Re: [DISCUSS] Breaking the Scala API for Scala 2.12 Support
Thanks Aljoscha for starting this discussion. The described problem brings us indeed a bit into a pickle. Even with option 1) I think it is somewhat API breaking because everyone who used lambdas without types needs to add them now. Consequently, I only see two real options out of the ones you've proposed: 1) Disambiguate the API (either by removing reduceGroup(GroupReduceFunction) or by renaming it to reduceGroupJ) 2) Maintain a 2.11 and 2.12 master branch until we phase 2.11 completely out Removing the reduceGroup(GroupReduceFunction) in option 1 is a bit problematic because then all Scala API users who have implemented a GroupReduceFunction need to convert it into a Scala lambda. Moreover, I think it will be problematic with RichGroupReduceFunction which you need to get access to the RuntimeContext. Maintaining two master branches puts a lot of burden onto the developers to always keep the two branches in sync. Ideally I would like to avoid this. I also played a little bit around with implicit conversions to add the lambda methods in Scala 2.11 on demand, but I was not able to get it work smoothly. I'm cross posting this thread to user as well to get some more user feedback. Cheers, Till On Thu, Oct 4, 2018 at 7:36 PM Elias Levy wrote: > The second alternative, with the addition of methods that take functions > with Scala types, seems the most sensible. I wonder if there is a need > then to maintain the *J Java parameter methods, or whether users could just > access the functionality by converting the Scala DataStreams to Java via > .javaStream and whatever the equivalent is for DataSets. > > On Thu, Oct 4, 2018 at 8:10 AM Aljoscha Krettek > wrote: > > > Hi, > > > > I'm currently working on > https://issues.apache.org/jira/browse/FLINK-7811, > > with the goal of adding support for Scala 2.12. There is a bit of a > hurdle > > and I have to explain some context first. > > > > With Scala 2.12, lambdas are implemented using the lambda mechanism of > > Java 8, i.e. Scala lambdas are now SAMs (Single Abstract Method). This > > means that the following two method definitions can both take a lambda: > > > > def map[R](mapper: MapFunction[T, R]): DataSet[R] > > def map[R](fun: T => R): DataSet[R] > > > > The Scala compiler gives precedence to the lambda version when you call > > map() with a lambda in simple cases, so it works here. You could still > call > > map() with a lambda if the lambda version of the method weren't here > > because they are now considered the same. For Scala 2.11 we need both > > signatures, though, to allow calling with a lambda and with a > MapFunction. > > > > The problem is with more complicated method signatures, like: > > > > def reduceGroup[R](fun: (scala.Iterator[T], Collector[R]) => Unit): > > DataSet[R] > > > > def reduceGroup[R](reducer: GroupReduceFunction[T, R]): DataSet[R] > > > > (for reference, GroupReduceFunction is a SAM with void > > reduce(java.lang.Iterable values, Collector out)) > > > > These two signatures are not the same but similar enough for the Scala > > 2.12 compiler to "get confused". In Scala 2.11, I could call > reduceGroup() > > with a lambda that doesn't have parameter type definitions and things > would > > be fine. With Scala 2.12 I can't do that because the compiler can't > figure > > out which method to call and requires explicit type definitions on the > > lambda parameters. > > > > I see some solutions for this: > > > > 1. Keep the methods as is, this would force people to always explicitly > > specify parameter types on their lambdas. > > > > 2. Rename the second method to reduceGroupJ() to signal that it takes a > > user function that takes Java-style interfaces (the first parameter is > > java.lang.Iterable while the Scala lambda takes a scala.Iterator). This > > disambiguates the code, users can use lambdas without specifying explicit > > parameter types but breaks the API. > > > > One effect of 2. would be that we can add a reduceGroup() method that > > takes a api.scala.GroupReduceFunction that takes proper Scala types, thus > > it would allow people to implement user functions without having to cast > > the various Iterator/Iterable parameters. > > > > Either way, people would have to adapt their code when moving to Scala > > 2.12 in some way, depending on what style of methods they use. > > > > There is also solution 2.5: > > > > 2.5 Rename the methods only in the Scala 2.12 build of Flink and keep the > > old method names for Scala 2.11. This would require some infrastructure > and > > I don't yet know how it can be done in a sane way. > > > > What do you think? I personally would be in favour of 2. but it breaks > the > > existing API. > > > > Best, > > Aljoscha > > > > > > > > >
Re: Scala 2.12 Support
Hi Aaron, we just released Flink 1.6 and the discussion for the roadmap of 1.7 will begin soon. I guess the Jira issue will also updated then. I would recommend to watch it for now. Regards, Timo Am 16.08.18 um 17:08 schrieb Aaron Levin: Hi Piotr, Thanks for the update. Glad to hear it's high on the priority list! I'm looking forward to the 1.7 update! It may be worth having someone more official from the Flink team give an update on that ticket. It wasn't clear if the 1.7 comment from that user was just a reference to the fact that 1.6 had come out (or where they got that information). I know a few people have cited the ticket and concluded "not clear what's going on with Scala 2.12 support." If you have the bandwidth, a note from you or anyone else would be helpful! Thanks again! Best, Aaron Levin On Thu, Aug 16, 2018 at 6:04 AM, Piotr Nowojski mailto:pi...@data-artisans.com>> wrote: Hi, Scala 2.12 support is high on our priority list and we hope to have it included for the 1.7 release (as you can see in the ticket itself), which should happen later this year. Piotrek On 15 Aug 2018, at 17:59, Aaron Levin mailto:aaronle...@stripe.com>> wrote: Hello! I'm wondering if there is anywhere I can see Flink's roadmap for Scala 2.12 support. The last email I can find on the list for this was back in January, and the FLINK-7811[0], the ticket asking for Scala 2.12 support, hasn't been updated in a few months. Recently Spark fixed the ClosureCleaner code to support Scala 2.12[1], and from what I can gather this was one of the main barrier for Flink supporting Scala 2.12. Given this has been fixed, is there work in progress to support Scala 2.12? Any updates on FLINK-7811? Thanks for your help! [0] https://issues.apache.org/jira/browse/FLINK-7811 <https://issues.apache.org/jira/browse/FLINK-7811> [1] https://issues.apache.org/jira/browse/SPARK-14540 <https://issues.apache.org/jira/browse/SPARK-14540> Best, Aaron Levin
Re: Scala 2.12 Support
Hi Piotr, Thanks for the update. Glad to hear it's high on the priority list! I'm looking forward to the 1.7 update! It may be worth having someone more official from the Flink team give an update on that ticket. It wasn't clear if the 1.7 comment from that user was just a reference to the fact that 1.6 had come out (or where they got that information). I know a few people have cited the ticket and concluded "not clear what's going on with Scala 2.12 support." If you have the bandwidth, a note from you or anyone else would be helpful! Thanks again! Best, Aaron Levin On Thu, Aug 16, 2018 at 6:04 AM, Piotr Nowojski wrote: > Hi, > > Scala 2.12 support is high on our priority list and we hope to have it > included for the 1.7 release (as you can see in the ticket itself), which > should happen later this year. > > Piotrek > > > On 15 Aug 2018, at 17:59, Aaron Levin wrote: > > Hello! > > I'm wondering if there is anywhere I can see Flink's roadmap for Scala > 2.12 support. The last email I can find on the list for this was back in > January, and the FLINK-7811[0], the ticket asking for Scala 2.12 support, > hasn't been updated in a few months. > > Recently Spark fixed the ClosureCleaner code to support Scala 2.12[1], and > from what I can gather this was one of the main barrier for Flink > supporting Scala 2.12. Given this has been fixed, is there work in progress > to support Scala 2.12? Any updates on FLINK-7811? > > Thanks for your help! > > [0] https://issues.apache.org/jira/browse/FLINK-7811 > [1] https://issues.apache.org/jira/browse/SPARK-14540 > > Best, > > Aaron Levin > > >
Re: Scala 2.12 Support
Hi, Scala 2.12 support is high on our priority list and we hope to have it included for the 1.7 release (as you can see in the ticket itself), which should happen later this year. Piotrek > On 15 Aug 2018, at 17:59, Aaron Levin wrote: > > Hello! > > I'm wondering if there is anywhere I can see Flink's roadmap for Scala 2.12 > support. The last email I can find on the list for this was back in January, > and the FLINK-7811[0], the ticket asking for Scala 2.12 support, hasn't been > updated in a few months. > > Recently Spark fixed the ClosureCleaner code to support Scala 2.12[1], and > from what I can gather this was one of the main barrier for Flink supporting > Scala 2.12. Given this has been fixed, is there work in progress to support > Scala 2.12? Any updates on FLINK-7811? > > Thanks for your help! > > [0] https://issues.apache.org/jira/browse/FLINK-7811 > <https://issues.apache.org/jira/browse/FLINK-7811> > [1] https://issues.apache.org/jira/browse/SPARK-14540 > <https://issues.apache.org/jira/browse/SPARK-14540> > > Best, > > Aaron Levin
Scala 2.12 Support
Hello! I'm wondering if there is anywhere I can see Flink's roadmap for Scala 2.12 support. The last email I can find on the list for this was back in January, and the FLINK-7811[0], the ticket asking for Scala 2.12 support, hasn't been updated in a few months. Recently Spark fixed the ClosureCleaner code to support Scala 2.12[1], and from what I can gather this was one of the main barrier for Flink supporting Scala 2.12. Given this has been fixed, is there work in progress to support Scala 2.12? Any updates on FLINK-7811? Thanks for your help! [0] https://issues.apache.org/jira/browse/FLINK-7811 [1] https://issues.apache.org/jira/browse/SPARK-14540 Best, Aaron Levin
Re: scala 2.12 support/cross-compile
Thanks Stephan and Alhoscha for the info! On Wed, Jan 3, 2018 at 2:41 AM Aljoscha Krettek <aljos...@apache.org> wrote: > Hi, > > This is the umbrella issue for Scala 2.12 support. As Stephan pointed out, > the ClosureCleaner and SAMs are currently the main problems. The first is > also a problem for Spark, which track their respective progress here: > https://issues.apache.org/jira/browse/SPARK-14540 > <https://issues.apache.org/jira/browse/SPARK-14540>. > > Best, > Aljoscha > > > On 3. Jan 2018, at 10:39, Stephan Ewen <se...@apache.org> wrote: > > Hi Hao Sun! > > This is work in progress, but Scala 2.12 is a bit tricky. I think the > Scala folks have messed this version up a bit, to be honest. > > The main blockers is that Scala 2.12 breaks some classes through its > addition of SAM interface lambdas (similar to Java). Many of the DataStream > API classes have two method variants (one with a Scala Function, one with a > Java SAM interface) which now become ambiguously overloaded methods in > Scala 2.12. > > In addition, Scala 2.12 also needs a different closure cleaner, because > Scala 2.12 compiles differently. > > I am adding Aljoscha, who has started working on this... > > Best, > Stephan > > > On Wed, Jan 3, 2018 at 4:13 AM, Hao Sun <ha...@zendesk.com> wrote: > >> Hi team, I am wondering if there is a schedule to support scala 2.12? >> If I need flink 1.3+ with scala 2.12, do I just have to cross compile >> myself? Is there anything blocking us from using scala 2.12? >> >> Thanks >> > > >
Re: scala 2.12 support/cross-compile
Hi, This is the umbrella issue for Scala 2.12 support. As Stephan pointed out, the ClosureCleaner and SAMs are currently the main problems. The first is also a problem for Spark, which track their respective progress here: https://issues.apache.org/jira/browse/SPARK-14540 <https://issues.apache.org/jira/browse/SPARK-14540>. Best, Aljoscha > On 3. Jan 2018, at 10:39, Stephan Ewen <se...@apache.org> wrote: > > Hi Hao Sun! > > This is work in progress, but Scala 2.12 is a bit tricky. I think the Scala > folks have messed this version up a bit, to be honest. > > The main blockers is that Scala 2.12 breaks some classes through its addition > of SAM interface lambdas (similar to Java). Many of the DataStream API > classes have two method variants (one with a Scala Function, one with a Java > SAM interface) which now become ambiguously overloaded methods in Scala 2.12. > > In addition, Scala 2.12 also needs a different closure cleaner, because Scala > 2.12 compiles differently. > > I am adding Aljoscha, who has started working on this... > > Best, > Stephan > > > On Wed, Jan 3, 2018 at 4:13 AM, Hao Sun <ha...@zendesk.com > <mailto:ha...@zendesk.com>> wrote: > Hi team, I am wondering if there is a schedule to support scala 2.12? > If I need flink 1.3+ with scala 2.12, do I just have to cross compile myself? > Is there anything blocking us from using scala 2.12? > > Thanks >
Re: scala 2.12 support/cross-compile
Hi Hao Sun! This is work in progress, but Scala 2.12 is a bit tricky. I think the Scala folks have messed this version up a bit, to be honest. The main blockers is that Scala 2.12 breaks some classes through its addition of SAM interface lambdas (similar to Java). Many of the DataStream API classes have two method variants (one with a Scala Function, one with a Java SAM interface) which now become ambiguously overloaded methods in Scala 2.12. In addition, Scala 2.12 also needs a different closure cleaner, because Scala 2.12 compiles differently. I am adding Aljoscha, who has started working on this... Best, Stephan On Wed, Jan 3, 2018 at 4:13 AM, Hao Sunwrote: > Hi team, I am wondering if there is a schedule to support scala 2.12? > If I need flink 1.3+ with scala 2.12, do I just have to cross compile > myself? Is there anything blocking us from using scala 2.12? > > Thanks >