- cabal-devel - Haskell.org

patch applied (hackage-server): "Fix Restore monad" and 15 others
by devnull＠community.haskell.org 31 Jan '13

31 Jan '13

Thu Jan 24 09:46:43 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Fix Restore monad Ignore-this: d64b04545adb4af8c80b75c67f8c88ff The presence of an explicit RestoreBind constructor meant that the monad laws were violated. M ./Distribution/Server/Framework/BackupRestore.hs -23 +56 Thu Jan 24 10:04:01 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Remove getStateSize from StateComponent Ignore-this: fd113eddc80e2c866cdbe495b2669850 (We can define it once and for all in AbstractStateComponent) M ./Distribution/Server/Features/BuildReports.hs -1 M ./Distribution/Server/Features/Core.hs -1 M ./Distribution/Server/Features/Distro.hs -1 M ./Distribution/Server/Features/Documentation.hs -1 M ./Distribution/Server/Features/DownloadCount.hs -1 M ./Distribution/Server/Features/HaskellPlatform.hs -1 M ./Distribution/Server/Features/Mirror.hs -1 M ./Distribution/Server/Features/PackageCandidates.hs -1 M ./Distribution/Server/Features/PreferredVersions.hs -1 M ./Distribution/Server/Features/Tags.hs -1 M ./Distribution/Server/Features/TarIndexCache.hs -9 +5 M ./Distribution/Server/Features/Upload.hs -3 M ./Distribution/Server/Features/Users.hs -2 M ./Distribution/Server/Framework/Feature.hs -5 +5 Tue Jan 29 13:00:25 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Disentangle htmlFeature (1/N) Ignore-this: 96e266f4cc4f60137ec4e253cf6427ae M ./Distribution/Server/Features/Html.hs -19 +35 Tue Jan 29 14:57:45 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Disentangle htmlFeature (2/N) Ignore-this: 8acaf1b3887228308320996c59d1696a This splits htmlFeature into components for each feature, but so far have made no effort to actually *reduce* the dependencies between these various parts. M ./Distribution/Server/Features/Html.hs -322 +482 Tue Jan 29 15:24:07 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Disentangle htmlFeature (3/N) Ignore-this: d1fae7bb76774b074eb078a560d3e324 Make dependencies more explicit (don't use {..}) M ./Distribution/Server/Features/Html.hs -18 +57 Tue Jan 29 16:35:41 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Disentangle htmlFeature (4/N) Ignore-this: 2db8b7b27a1d0e85e7e2a603af893457 Make the overlap between features more evident M ./Distribution/Server/Features/Html.hs -42 +51 Tue Jan 29 17:27:27 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Start having PackageCandidates implement the CoreFeature interface Ignore-this: deaf61445bf67c6aecf0c63c322fa588 The purpose is that features such as BuildReports or Documentation can then be instantiated by both Core and PackageCandidates. M ./Distribution/Server/Features/Core.hs -4 +4 M ./Distribution/Server/Features/Html.hs -7 +6 M ./Distribution/Server/Features/PackageCandidates.hs -42 +64 Wed Jan 30 09:43:00 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Use "/.." instead of "*" in ServerIntrospect Ignore-this: ac10101fcc3d3007da939d9e23d57221 This brings it inline with what we actually use in the code. M ./Distribution/Server/Features/ServerIntrospect.hs -2 +2 Wed Jan 30 09:52:53 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Document the documentation feature Ignore-this: afc7ae265cdf219e3e2b1a5c74f3e040 M ./Distribution/Server/Features/Documentation.hs -7 +12 Wed Jan 30 12:41:25 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Core cleanup (1/N) Ignore-this: 6a82aa13b3a3ce8e3a95e6172725dd7f Core had withPackageId :: DynamicPath -> (PackageId -> ServerPartE a) -> ServerPartE a withPackageName :: MonadIO m => DynamicPath -> (PackageName -> ServerPartT m a) -> ServerPartT m a But both of these are an instance of a much more general function withPackageInPath :: (MonadPlus m, FromReqURI a) => DynamicPath -> (a -> m b) -> m b So I have removed the first two in favour of the withPackageInPath. Moreover, withPackageInPath belongs in CoreResource more than in CoreFeature -- it doesn't access the package database, but simply extracts a bit from a URL. CoreFeature still contains a whole plethora of withXXX functions; the logical place for these is indeed in Feature rather than Resource, but we don't need quite all the different combinations; we can offer more general functions. Will do that next. The immediate reason for all this cleanup is that we want the PackageCandidates feature to ultimately implement the core API so that other packages (build reports, documentation, tags, mirror, versions, ..?) can work with either. M ./Distribution/Server/Features/Core.hs -3 +6 M ./Distribution/Server/Features/Distro.hs -2 +4 M ./Distribution/Server/Features/Documentation.hs -2 +2 M ./Distribution/Server/Features/Html.hs -26 +22 M ./Distribution/Server/Features/Mirror.hs -4 +12 M ./Distribution/Server/Features/PackageCandidates.hs -5 +10 M ./Distribution/Server/Features/PreferredVersions.hs -3 +12 M ./hackage-server.cabal +1 Thu Jan 31 09:07:44 GMT 2013 edsko(a)well-typed.com * Start removing unnecessary withXXX style Ignore-this: 398ce12fc151ed4e646efeb5bbdfb741 Since these functions don't do any resource management, and live in a MonadPlus anyway, there is no need for withXXX style which just adds clutter. M ./Distribution/Server/Features/Core.hs -2 +4 M ./Distribution/Server/Features/Distro.hs -3 +3 M ./Distribution/Server/Features/Html.hs -77 +75 M ./Distribution/Server/Features/Mirror.hs -32 +32 M ./Distribution/Server/Features/PackageCandidates.hs -21 +24 M ./Distribution/Server/Features/PreferredVersions.hs -4 +4 M ./hackage-server.cabal -1 +1 Thu Jan 31 13:23:48 GMT 2013 edsko(a)well-typed.com * Further cleanup of Core Ignore-this: 1562fbab4220ee28f3381fdee1437d9a Core had this whole plethora of functions: withPackage :: PackageId -> (PkgInfo -> [PkgInfo] -> ServerPartE a) -> ServerPartE a withPackagePath :: DynamicPath -> (PkgInfo -> [PkgInfo] -> ServerPartE a) -> ServerPartE a withPackageAll :: PackageName -> ([PkgInfo] -> ServerPartE a) -> ServerPartE a withPackageAllPath :: DynamicPath -> (PackageName -> [PkgInfo] -> ServerPartE a) -> ServerPartE a withPackageVersion :: PackageId -> (PkgInfo -> ServerPartE a) -> ServerPartE a withPackageVersionPath :: DynamicPath -> (PkgInfo -> ServerPartE a) -> ServerPartE a withPackageTarball :: DynamicPath -> (PackageId -> ServerPartE a) -> ServerPartE a The withXXX style of these functions is unnecessary; it's unclear what precisely these functions do, the names don't make it obvious; and there are too many combinations. In a previous we paved the way for addressing the third concern by introducing packageInPath :: (MonadPlus m, FromReqURI a) => DynamicPath -> m a We now added packageTarballInPath :: MonadPlus m => DynamicPath -> m PackageId replacing withPackageTarball (although it still feels a little adhoc). The other 6 functions have been replaced with two: lookupPackageName :: PackageName -> ServerPartE [PkgInfo] lookupPackageId :: PackageId -> ServerPartE PkgInfo The first finds all versions of a package, while the second finds a specific version of a package, or the most recent if the version in the given identifier is empty. The purpose of both functions should hopefully be evident from their name, and obviously we got rid of the withXXX style. TRANSITION withPackage withPackage returned all packages with the given name, and the most recent version of that package if the version of the package id was Version [] [] or the specific version otherwise. This function was never called directly, but only through withPackagePath. withPackagePath This is a combination of withPackage and withPackageId, which is now packageInPath: BEFORE AFTER withPackagePath dpath $ \_ pkgs -> pkgs <- packageInPath dpath >>= lookupPackage . pkgName withPackagePath dpath $ \pkg _ -> pkg <- packageInPath dpath >>= lookupPackageId (It was only used in one of these two ways, with one argument always ignored, and the first form occurred only once.) withPackageAll This is now lookupPackageName. (Although it was often called like BEFORE AFTER withPackageAll pkgname $ \_ -> _ <- lookupPackageName pkgname We might want to replace that with something like checkPackageExists. withPackageAllPath This used withPackageName (now packageInPath) and returned the package name as well as the packages themselves. BEFORE AFTER withPackageAllPath dpath $ \pkgname pkgs pkgname <- packageInPath dpath pkgs <- lookupPackageName pkgname withPackageAllPath dpath $ \pkgname _ -> pkgname <- packageInPath dpath _ <- lookupPackageName pkgname The second form occurred surprisingly often; I've marked these calls to lookupPackageName as "TODO: Necessary?" because it wasn't obvious to me if (all) of these calls were really required, or whether it was just a "default" option to use withPackageAllPath rather than (what used to be) withPackageName. withPackageVersion This is like lookupPackageId, but explicitly ruled out empty versions. BEFORE AFTER withPackageVersion pkgid $ \pkg -> guard (pkgVersion pkgid /= Version [] []) pkg <- lookupPackageId pkgid (This occurred only once.) withPackageVersionPath Combination of withPackageId (now packageInPath) and withPackageVersion, which was *only* used in the BuildReports module. BEFORE AFTER withPackageVersionPath dpath $ \pkg -> pkgid <- packageInPath dpath guard (pkgVersion pkgid /= Version [] []) pkg <- lookupPackageId pkgid Although the latter is a bit more verbose, it's also a bit more explicit about what's happening, and in many cases the original code contained a subsequent line let pkgid = pkgInfoId pkg anyway (in which case the 'pkg' itself was ignored; I've left in the calls to lookupPackageId regardless, as a check that the package does in fact exist. I'm not 100% all of these are necessary.) M ./Distribution/Server/Features/BuildReports.hs -38 +45 M ./Distribution/Server/Features/Core.hs -83 +57 M ./Distribution/Server/Features/Documentation.hs -26 +27 M ./Distribution/Server/Features/Html.hs -54 +54 M ./Distribution/Server/Features/Mirror.hs -39 +42 M ./Distribution/Server/Features/PackageCandidates.hs -5 +4 M ./Distribution/Server/Features/PackageContents.hs -3 +7 M ./Distribution/Server/Features/PreferredVersions.hs -25 +27 M ./Distribution/Server/Features/Tags.hs -14 +17 Thu Jan 31 15:23:31 GMT 2013 edsko(a)well-typed.com * Add guardValidPackageXXX Ignore-this: ba9516f019aed244dfd06e4c7c4624d8 In our relentness quest to bring Core and Candidates closer together, add guardValidPackageId and guardValidPackageName to CoreResource. Many features just need to check if a package ID is valid, rather than get access to the actual package (BuildReports being one example). This means that these features can easily work with either Core or Candidates (which, after all, use different types of packages). This also addresses the issue mentioned in the previous log, where we used the idiom _ <- lookupPackageXXX to check that a package exists. M ./Distribution/Server/Features.hs -2 +2 M ./Distribution/Server/Features/BuildReports.hs -24 +13 M ./Distribution/Server/Features/Core.hs -1 +12 M ./Distribution/Server/Features/Html.hs -6 +4 M ./Distribution/Server/Features/PreferredVersions.hs -7 +9 M ./Distribution/Server/Features/Tags.hs -2 +2 Thu Jan 31 15:32:04 GMT 2013 edsko(a)well-typed.com * Remove withXXX from BuildReports Ignore-this: bd91f01277dffdde57ef3623dd7bf87d .. while we were here anyway .. M ./Distribution/Server/Features/BuildReports.hs -40 +38 Thu Jan 31 15:45:49 GMT 2013 edsko(a)well-typed.com * Remove withXXX style from Upload Ignore-this: 691782ddc84077e053242ff5ec121bec In preparation for cleaning up Candidates, still. M ./Distribution/Server/Features/Documentation.hs -12 +12 M ./Distribution/Server/Features/Html.hs -20 +18 M ./Distribution/Server/Features/PackageCandidates.hs -4 +4 M ./Distribution/Server/Features/PreferredVersions.hs -52 +52 M ./Distribution/Server/Features/Upload.hs -29 +30 Thu Jan 31 15:50:14 GMT 2013 edsko(a)well-typed.com * Documentation now relies on CoreResource only Ignore-this: 9377ef7f88fc721963161d8eddb0d3d0 This means that BuildReports and Documentation should now (in principle) be able to work with Core as well as Candidates. M ./Distribution/Server/Features.hs -1 +1 M ./Distribution/Server/Features/Documentation.hs -9 +7

1 0

Cleaning up the bug tracker
by Johan Tibell 28 Jan '13

28 Jan '13

Cabal hackers, With 477 open issues, it's hard to use the bug tracker efficiently e.g. for prioritizing work. Gregory Collins and I have started cleaning up some dead bugs. We won't get through all of them, so I encourage other to join in a try to reproduce bugs and add a note to the bug if it's reproducible or not. Feel free to drop me an email if there are any bugs you think could be closed. -- Johan

3 2

patch applied (hackage-server): "Restrict uploads to members of the package maintainer group" and 4 others
by devnull＠community.haskell.org 25 Jan '13

25 Jan '13

Fri Jan 25 17:31:02 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Restrict uploads to members of the package maintainer group Ignore-this: c7e40d08b81e18d3d52e6c39c0426a0 We can do this now that we can pre-populate the groups when we import from the old hackage server. M ./Distribution/Server/Features/PackageCandidates.hs -1 +1 M ./Distribution/Server/Features/Upload.hs -6 +1 Fri Jan 25 17:35:10 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Only add the uploader to the maintainer group on normal uploads Ignore-this: f1f062df0d09dc997b3580054c134e25 Previously the mirroring client would also get added to the maintainer group. M ./Distribution/Server/Features/PackageCandidates.hs +2 M ./Distribution/Server/Features/Upload.hs -9 +2 Fri Jan 25 17:47:11 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Improve a couple package upload error messages Ignore-this: 35528e111a04c660be22bc36bb264f84 M ./Distribution/Server/Features/Upload.hs -7 +18 Fri Jan 25 17:48:16 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Return ETags when serving tarballs Ignore-this: eaca4f4574be1f47c7a57f775052e0c5 Note that nowhere yet do we actually do the if-not-match check. M ./Distribution/Server/Features/Documentation.hs -1 +3 M ./Distribution/Server/Features/PackageCandidates.hs -11 +13 M ./Distribution/Server/Features/PackageContents.hs -14 +16 M ./Distribution/Server/Framework/ResponseContentTypes.hs -4 +15 M ./Distribution/Server/Util/ServeTarball.hs -10 +11 Fri Jan 25 17:49:50 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Add a note about a wrong error response in some circumstances Ignore-this: dd01ed0d65d9b0817a3317660a3aa6e3 M ./Distribution/Server/Features/Users.hs +3

1 0

[PATCH] Fix building dynamic/profiling executables that use TH
by Patrick Palka 24 Jan '13

24 Jan '13

We need to build a _vanilla_ executable the first time. The previous code was essentially a no-op as it did not change the flags used to build the executable the first time. --- Cabal/Distribution/Simple/GHC.hs | 13 ++++++------- 1 file changed, 6 insertions(+), 7 deletions(-) diff --git a/Cabal/Distribution/Simple/GHC.hs b/Cabal/Distribution/Simple/GHC.hs index 5f2926a..84745d6 100644 --- a/Cabal/Distribution/Simple/GHC.hs +++ b/Cabal/Distribution/Simple/GHC.hs @@ -866,15 +866,14 @@ buildExe verbosity _pkg_descr lbi ghcOptExtra = ["-no-hs-main" | not isHaskellMain ] } - -- For building exe's for profiling that use TH we actually - -- have to build twice, once without profiling and the again - -- with profiling. This is because the code that TH needs to - -- run at compile time needs to be the vanilla ABI so it can - -- be loaded up and run by the compiler. + -- For building profiling exe's or dynamic exe's that use TH + -- we actually have to build twice, once the vanilla way and + -- then the profiling/dynamic way. This is because GHC can only + -- use vanilla object files to link & run the appropriate + -- compile-time TH code. when ((withProfExe lbi || withDynExe lbi) && EnableExtension TemplateHaskell `elem` allExtensions exeBi) $ - runGhcProg compileOpts { ghcOptNoLink = toFlag True } - --TODO: do we also need to play the static vs dynamic games here? + runGhcProg staticOpts { ghcOptNoLink = toFlag True } runGhcProg compileOpts { ghcOptNoLink = toFlag True } -- 1.7.10.4

1 0

patch applied (hackage-server): "Add the tarIndexCacheFeature to the list of features"
by devnull＠community.haskell.org 24 Jan '13

24 Jan '13

Thu Jan 24 22:09:07 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Add the tarIndexCacheFeature to the list of features Ignore-this: 7b300d7026741cb130afc99f70fd5dbc M ./Distribution/Server/Features.hs -1 +2

1 0

patch applied (hackage-server): "Fix a couple minor content type issues" and 3 others
by devnull＠community.haskell.org 24 Jan '13

24 Jan '13

Wed Jan 9 21:46:16 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Fix a couple minor content type issues Ignore-this: b8d62ee739230922efe0db9bc85d92b1 M ./Distribution/Server/Framework/ResponseContentTypes.hs -1 +1 M ./Distribution/Server/Util/ServeTarball.hs -1 +3 Wed Jan 9 21:46:42 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Fix a warning Ignore-this: a7e5bbcf3a8a43caf319d12a061ac05b M ./Distribution/Server/Util/ServeTarball.hs -4 +4 Wed Jan 9 21:47:03 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * Warn rather than fail for uploading docs for non-existant packages Ignore-this: e47ca7c566f4ff5253d03a57de4c8d48 e.g. base-2.x is not uploaded, but the old hackage has docs for it. M ./ImportClient.hs +3 Thu Jan 24 21:30:35 GMT 2013 Duncan Coutts <duncan(a)well-typed.com> * We need a more recent zlib version Ignore-this: 1f3a044a1eb8ec7509506972f02bfa43 M ./hackage-server.cabal -1 +1

1 0

Re: Darcs.haskell.org down again
by Ian Lynagh 23 Jan '13

23 Jan '13

On Wed, Jan 23, 2013 at 10:17:24AM +0000, Simon Peyton-Jones wrote: > > It seems that 'abbott' is down again. Two days in a row suggests something is wrong. Is there anything either of you can do to help? I'm not sure exactly what caused the problems, but I've: * Disabled http://hackage.haskell.org/cgi-bin/hackage-scripts/archive.tar in /etc/apache2/sites-available/hackage * Disabled git integration with the GHC trac * Removed gitweb Hopefully that will help. Thanks Ian -- Ian Lynagh, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

2 1

patch applied (hackage-server): "Have the Documentation use the TarIndexCache"
by devnull＠community.haskell.org 23 Jan '13

23 Jan '13

Wed Jan 23 17:15:07 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Have the Documentation use the TarIndexCache Ignore-this: b58eda75c156c31b316723965eaa5412 This means that we don't keep all the tar indices in memory but deserializable them on every request, thereby saving memory (and the Documentation feature is simpler now that the tar index caching is done in a different feature). I noticed however that the build client no longer works; in particular, uploading documentation does not work (I think the URL is wrong; .../package/doc rather than .../package/docs, and it tries to upload a .tar.gz while we expect a .tar). M ./Distribution/Server/Features.hs +1 M ./Distribution/Server/Features/Documentation.hs -23 +39 M ./Distribution/Server/Features/Documentation/State.hs -7 +7 M ./Distribution/Server/Features/TarIndexCache.hs -1 +8

1 0

patch applied (hackage-server): "Factor out common PackageContents/PackageCandidates code"
by devnull＠community.haskell.org 23 Jan '13

23 Jan '13

Wed Jan 23 15:10:44 GMT 2013 Edsko de Vries <edsko(a)well-typed.com> * Factor out common PackageContents/PackageCandidates code Ignore-this: 5d0e3699077e3943eecfa391b3faeaf9 There is still a bit of duplication, but this is "uninteresting" code (in the sense that there are no real design decisions involved, just getting information from various places). In particular, the code that finds a changelog in a tarball (looking for ChangeLog, CHANGELOG, etc.) is now in a separate module. M ./Distribution/Server/Features/PackageCandidates.hs -57 +31 M ./Distribution/Server/Features/PackageContents.hs -61 +29 A ./Distribution/Server/Packages/ChangeLog.hs M ./Distribution/Server/Util/ServeTarball.hs -1 +1 M ./hackage-server.cabal +2

1 0

Cabal and cross compilation
by Nathan Hüsken 23 Jan '13

23 Jan '13

Hey, I am working on getting ghc to cross compile to android. When trying to get haskeline to compile. I want to change the cabal file such that it sets a flag when compiling for android. For that I changed cabal so that it recognizes android as a OS. But cabal seems to get its os information from System.Info.os, and from what I can tell this always returns the host os and not the target os. Am I getting this right, is cabal unaware of the target os? How can we change this? Regards, Nathan

2 1