Themes for 1.22
Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're almost done here. There's really only one remaining thing to do: * Build components and different ways (e.g. profiling) in parallel. We could build both profiling and non-profiling versions in parallel. We could also build e.g. all test suites in parallel. The key challenge here is to coordinate all parallel jobs so we don't spawn too many. ## Support large projects This is also a carry-over from the 1.20 goals. We still don't have a good story for large projects. Sandboxes are too annoying to use if there are 100 add-source deps. We need more automation and more opinionated defaults (e.g. scan the sub-directories from in which cabal was run to find source packages.) What we need most of all here is a design. Perhaps we could try to get together at some Hackathon/ICFP and discuss. ## Issue tracker spring cleaning and test suite improvements The issue tracker is out-of-hand. It's too unwieldy to use for planning our work and get an overview of the most important issues. We should try to close down bugs that haven't had updates in years with extreme prejudice. If the issue is important it will pop up again. We're also severely lacking in the testing department. There are two problems: * There aren't enough tests: the cabal user facing surface is quite larger (lots of features and flags) and many of them are not tested at all, which will lead to regressions as we keep fixing bugs and adding features. * The tests take too long to run: we have too many end-to-end style tests (i.e. build a whole package) and not enough unit style tests. We need to add more of the latter kind. Cheers, Johan ___ cabal-devel mailing list cabal-devel@haskell.org http://www.haskell.org/mailman/listinfo/cabal-devel
Themes for 1.22
Hi Johan, Not sure if this mail is a request for comments, but on the story for large projects one thing that I would like to see is the ability to add packages that aren't in hackage to the depends list. I agree that adding some scanning and auto add ability is definitely sorely needed, but this information goes into add-source-timestamps in the sandbox folder, and if you blow away the sandbox, it's gone. Or you can try and version a file that contains ever changing timestamps. Right now I have a shell script that maintains a list of added sources and I keep that versioned in the project. It would be nice to have git dependencies in there as well, and then disallow a hackage upload for any cabal file with a non-hackage dependency listed. I would be happy to contribute time to both design and implementation. Ben On Thu Apr 24 2014 at 10:53:59, Johan Tibell johan.tib...@gmail.com wrote: Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're almost done here. There's really only one remaining thing to do: * Build components and different ways (e.g. profiling) in parallel. We could build both profiling and non-profiling versions in parallel. We could also build e.g. all test suites in parallel. The key challenge here is to coordinate all parallel jobs so we don't spawn too many. ## Support large projects This is also a carry-over from the 1.20 goals. We still don't have a good story for large projects. Sandboxes are too annoying to use if there are 100 add-source deps. We need more automation and more opinionated defaults (e.g. scan the sub-directories from in which cabal was run to find source packages.) What we need most of all here is a design. Perhaps we could try to get together at some Hackathon/ICFP and discuss. ## Issue tracker spring cleaning and test suite improvements The issue tracker is out-of-hand. It's too unwieldy to use for planning our work and get an overview of the most important issues. We should try to close down bugs that haven't had updates in years with extreme prejudice. If the issue is important it will pop up again. We're also severely lacking in the testing department. There are two problems: * There aren't enough tests: the cabal user facing surface is quite larger (lots of features and flags) and many of them are not tested at all, which will lead to regressions as we keep fixing bugs and adding features. * The tests take too long to run: we have too many end-to-end style tests (i.e. build a whole
Re: Themes for 1.22
You can handle this scenario pretty easily by setting up a private hackage. Then add a second remote-repo to your cabal config, either globally or in the sandbox. We do this, and it works pretty well. Regards, Erik On Thu, Apr 24, 2014 at 1:33 PM, Benjamin Edwards edwards.b...@gmail.com wrote: Hi Johan, Not sure if this mail is a request for comments, but on the story for large projects one thing that I would like to see is the ability to add packages that aren't in hackage to the depends list. I agree that adding some scanning and auto add ability is definitely sorely needed, but this information goes into add-source-timestamps in the sandbox folder, and if you blow away the sandbox, it's gone. Or you can try and version a file that contains ever changing timestamps. Right now I have a shell script that maintains a list of added sources and I keep that versioned in the project. It would be nice to have git dependencies in there as well, and then disallow a hackage upload for any cabal file with a non-hackage dependency listed. I would be happy to contribute time to both design and implementation. Ben On Thu Apr 24 2014 at 10:53:59, Johan Tibell johan.tib...@gmail.com wrote: Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're almost done here. There's really only one remaining thing to do: * Build components and different ways (e.g. profiling) in parallel. We could build both profiling and non-profiling versions in parallel. We could also build e.g. all test suites in parallel. The key challenge here is to coordinate all parallel jobs so we don't spawn too many. ## Support large projects This is also a carry-over from the 1.20 goals. We still don't have a good story for large projects. Sandboxes are too annoying to use if there are 100 add-source deps. We need more automation and more opinionated defaults (e.g. scan the sub-directories from in which cabal was run to find source packages.) What we need most of all here is a design. Perhaps we could try to get together at some Hackathon/ICFP and discuss. ## Issue tracker spring cleaning and test suite improvements The issue tracker is out-of-hand. It's too unwieldy to use for planning our work and get an overview of the most important issues. We should try to close down bugs that haven't had updates in years with extreme prejudice. If the issue is important it will pop up again. We're also severely lacking in the testing department. There are two problems: * There
Re: Themes for 1.22
I will definitely look into that! Thank you. The one your that does force you down though is the path only developing against releases, and even those are light-weight internal patch releases to your own hackage it still means a longer compile-test cycle. Maybe if that's a problem it indicates that my package boundaries are too fine, but I find add-source to be invaluable, I would find it even nicer if it was a little bit more first class. If you look at how source dependencies are covered in VS when programming in .Net, that's somewhat how I would like to see it managed in cabal. It's trivial to have common assemblies in a source tree shared between projects. Ben On Thu Apr 24 2014 at 12:50:45, Erik Hesselink hessel...@gmail.com wrote: You can handle this scenario pretty easily by setting up a private hackage. Then add a second remote-repo to your cabal config, either globally or in the sandbox. We do this, and it works pretty well. Regards, Erik On Thu, Apr 24, 2014 at 1:33 PM, Benjamin Edwards edwards.b...@gmail.com wrote: Hi Johan, Not sure if this mail is a request for comments, but on the story for large projects one thing that I would like to see is the ability to add packages that aren't in hackage to the depends list. I agree that adding some scanning and auto add ability is definitely sorely needed, but this information goes into add-source-timestamps in the sandbox folder, and if you blow away the sandbox, it's gone. Or you can try and version a file that contains ever changing timestamps. Right now I have a shell script that maintains a list of added sources and I keep that versioned in the project. It would be nice to have git dependencies in there as well, and then disallow a hackage upload for any cabal file with a non-hackage dependency listed. I would be happy to contribute time to both design and implementation. Ben On Thu Apr 24 2014 at 10:53:59, Johan Tibell johan.tib...@gmail.com wrote: Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're almost done here. There's really only one remaining thing to do: * Build components and different ways (e.g. profiling) in parallel. We could build both profiling and non-profiling versions in parallel. We could also build e.g. all test suites in parallel. The key challenge here is to coordinate all parallel jobs so we don't spawn too many. ## Support large projects
Re: Themes for 1.22
Having to share code via releases is indeed the problem with an internal Hackage. I don't see that scaling well as an organization gets bigger. You really need to be able to check out the organization's source repo and have that contain a bunch of (non-released) packages and build them together. On Thu, Apr 24, 2014 at 2:10 PM, Benjamin Edwards edwards.b...@gmail.comwrote: I will definitely look into that! Thank you. The one your that does force you down though is the path only developing against releases, and even those are light-weight internal patch releases to your own hackage it still means a longer compile-test cycle. Maybe if that's a problem it indicates that my package boundaries are too fine, but I find add-source to be invaluable, I would find it even nicer if it was a little bit more first class. If you look at how source dependencies are covered in VS when programming in .Net, that's somewhat how I would like to see it managed in cabal. It's trivial to have common assemblies in a source tree shared between projects. Ben On Thu Apr 24 2014 at 12:50:45, Erik Hesselink hessel...@gmail.com wrote: You can handle this scenario pretty easily by setting up a private hackage. Then add a second remote-repo to your cabal config, either globally or in the sandbox. We do this, and it works pretty well. Regards, Erik On Thu, Apr 24, 2014 at 1:33 PM, Benjamin Edwards edwards.b...@gmail.com wrote: Hi Johan, Not sure if this mail is a request for comments, but on the story for large projects one thing that I would like to see is the ability to add packages that aren't in hackage to the depends list. I agree that adding some scanning and auto add ability is definitely sorely needed, but this information goes into add-source-timestamps in the sandbox folder, and if you blow away the sandbox, it's gone. Or you can try and version a file that contains ever changing timestamps. Right now I have a shell script that maintains a list of added sources and I keep that versioned in the project. It would be nice to have git dependencies in there as well, and then disallow a hackage upload for any cabal file with a non-hackage dependency listed. I would be happy to contribute time to both design and implementation. Ben On Thu Apr 24 2014 at 10:53:59, Johan Tibell johan.tib...@gmail.com wrote: Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're
Re: Themes for 1.22
I'd like to add to the list: change cabal-install so that most of its code is exported to hackage as a library. In the past I've occasionally wanted to do some of the same things the cabal binary does from code, and it would be much more convenient to link the relevant code in than to exec the binary all the time. On Thu, Apr 24, 2014 at 11:53 AM, Johan Tibell johan.tib...@gmail.comwrote: Hi all, While I'm sure we still have a bugfix release or two to make on the 1.20 branch, I thought it'd be worth looking at what we want to accomplish for 1.22. Here are my thoughts on what we should focus on: ## A dependency solver that always works As Hackage has grown so have the requirements of the dependency solver. There are three distinct problems I'm seeing now that we should tackle: * Treat each sections (i.e. library, test suite, benchmark, and executable) in the .cabal file separately for the purpose of dependency resolution. Today all the sections' dependencies are merged and used as the constraints of the package as a whole. This is troublesome for all packages that are dependencies of QC, HUnit, test-framework, and criterion, as there's a dependency cycle if you treat e.g. the containers package and its test suite as one unit. The solution here is to treat each unit as a mini package for the purpose of dependency resolution. This would also allow you to have e.g. several executables with conflicting dependencies. * Improve performance. Some packages (e.g. yesod) can take over 10 seconds to run over. This problem will get worse as Hackage grows and we build bigger applications on top of it, so we need to tackle this now before it becomes a real problem. * Fix Hackage package blacklisting. Users can blacklist packages on Hackage e.g. if they know them to be broken. However, this doesn't really work as the Hackage blacklist translates to a soft preference in the dependency solver and is thus often ignored. See https://github.com/haskell/cabal/issues/1792 for the gory details. ## Do the right thing automatically This is a carry-over from the 1.20 goals, as we didn't make much progress here. The focus here should be on avoiding manual steps the cabal could do for the user. * Automatically install dependencies when needed. When `cabal build` would fail due to a missing dependency, just install this dependency instead of bugging the user to do it. This will probably have to be limited to sandboxes where we can't break the user's system * GHCi support could be improved by rebinding :reload to rerun e.g. preprocessors automatically. This would enable the users to develop completely from within ghci (i.e. faster edit-save-type-error cycle). We have most of what we need here (i.e. GHC macro support) but someone needs to make the final change to generate a .ghci file to pass in the ghci invocation. ## Faster builds I think we're almost done here. There's really only one remaining thing to do: * Build components and different ways (e.g. profiling) in parallel. We could build both profiling and non-profiling versions in parallel. We could also build e.g. all test suites in parallel. The key challenge here is to coordinate all parallel jobs so we don't spawn too many. ## Support large projects This is also a carry-over from the 1.20 goals. We still don't have a good story for large projects. Sandboxes are too annoying to use if there are 100 add-source deps. We need more automation and more opinionated defaults (e.g. scan the sub-directories from in which cabal was run to find source packages.) What we need most of all here is a design. Perhaps we could try to get together at some Hackathon/ICFP and discuss. ## Issue tracker spring cleaning and test suite improvements The issue tracker is out-of-hand. It's too unwieldy to use for planning our work and get an overview of the most important issues. We should try to close down bugs that haven't had updates in years with extreme prejudice. If the issue is important it will pop up again. We're also severely lacking in the testing department. There are two problems: * There aren't enough tests: the cabal user facing surface is quite larger (lots of features and flags) and many of them are not tested at all, which will lead to regressions as we keep fixing bugs and adding features. * The tests take too long to run: we have too many end-to-end style tests (i.e. build a whole package) and not enough unit style tests. We need to add more of the latter kind. Cheers, Johan ___ cabal-devel mailing list cabal-devel@haskell.org http://www.haskell.org/mailman/listinfo/cabal-devel -- Gregory Collins g...@gregorycollins.net ___ cabal-devel mailing list cabal-devel@haskell.org http://www.haskell.org/mailman/listinfo/cabal-devel