March 2008 - cabal-devel

[Hackage] #229: Build C sources using the system's C compiler directly
by Hackage 15 Aug '08

15 Aug '08

#229: Build C sources using the system's C compiler directly ----------------------------+----------------------------------------------- Reporter: guest | Owner: Type: enhancement | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------+----------------------------------------------- Currently Cabal builds C sources by going via ghc. On some platforms this introduces some limitations (Windows especially, I think). See this conversation for an example where it would be useful: http://www .mail-archive.com/haskell-cafe(a)haskell.org/msg37179.html -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/229> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 3

[Hackage] #225: allow installing just specific bits, like just docs
by Hackage 15 Aug '08

15 Aug '08

#225: allow installing just specific bits, like just docs ----------------------------+----------------------------------------------- Reporter: duncan | Owner: Type: enhancement | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------+----------------------------------------------- Sometimes people would like to build and install just part of a cabal project, e.g. the documentation or one of the libraries or binaries. Specific use-cases are packaging for OS distro's and generating Haddock docs for code that was installed without it (cf the current state of Debian-stable). Igloo proposes the following user interface: ----- As long as docs, license and binaries can be (de)selected individually I don't really mind. If you want a concrete suggestion, we could have `--foo` and `--no-foo` for each thing, with the default being `--all` and the command line being processed left-to-right, e.g. {{{ cabal install --no-haddock-interfaces --docs --no-html }}} would install * all (which wouldn't actually contain anything) * docs (which wouldn't actually contain anything) * haddock-interfaces * license * everything under binaries ----- I broadly agree, but am slightly concerned there are dependencies amongst the various things that his syntax does not respect (e.g. one may need to install the libraries - e.g. one depending on a foreign library - in order to get a binary working). So either we have to vet the command lines or coarsen the flags. I hope Duncan and Igloo can draw on their packaging experience and determine what is actually useful. -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/225> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 6

[Hackage] #212: Overhaul Cabal's testsuite
by Hackage 15 Aug '08

15 Aug '08

#212: Overhaul Cabal's testsuite ----------------------------------+----------------------------------------- Reporter: duncan | Owner: Type: task | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: very hard (<1 week) | Ghcversion: 6.8.2 Platform: | ----------------------------------+----------------------------------------- The current testsuite based on HUnit is almost completely bitrotted. Anyone who wants to take this on has essentially a completely clean slate to work from. The current system is split into two parts. There are hunit tests in many of the modules guarded by #ifdef DEBUG. Then there are external tests in the `tests/` subdirectory. The #ifdef DEBUG style is annoying since it involves using more CPP than seem really necessary. A couple notes from the old `TODO` file: {{{ ** add a make target or command for tests we know will fail? ** setup test suite to run on --push? ** redirect non-hunit outputs to a file? }}} One possibility that is available to us now that was not before is using the hackage collection as part of a regression test. For example it should always be possible to parse all the `.cabal` files in the index. -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/212> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 3

[Hackage] #198: cabal install foo should upgrade foo if there's a newer version
by Hackage 12 Aug '08

12 Aug '08

#198: cabal install foo should upgrade foo if there's a newer version ---------------------------------+------------------------------------------ Reporter: guest | Owner: Type: defect | Status: new Priority: normal | Milestone: Component: cabal-install tool | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: Linux | ---------------------------------+------------------------------------------ if I have foo-1.0 installed, and foo-2.0 is available, "cabal install foo" should upgrade foo to version 2.0 -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/198> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 8

[Hackage] #175: cabal-install needs a new package dependency resolution algorithm
by Hackage 12 Aug '08

12 Aug '08

#175: cabal-install needs a new package dependency resolution algorithm ----------------------------+----------------------------------------------- Reporter: duncan | Owner: Type: task | Status: new Priority: normal | Milestone: Component: cabal-install | Version: 1.2.2.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.4.2 Platform: Linux | ----------------------------+----------------------------------------------- The current package dependency algorithm is pretty simple. It's main failing is that it cannot backtrack if it finds the solution is impossible. What we would like is a new dependency resolution algorithm that can backtrack. We would also like to be able to configure policy issues separately, perhaps as a parameter. It may also be handy to let users or the system add extra version constraints. The input is the set of available packages and the output is a list of packages to install, and the assignment of configuration flags to use for each one. As for the policy input, this might take the form of a function that picks between possible dependencies in a slot, this could then allow us to make choices like preferring installed packages over the latest available package version, or vica versa. See also symptoms of the existing system: * bug #168 -- lack of configurable policy * bug #174 -- lack of backtracking -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/175> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 4

[Hackage] #213: allow wildcards in data-files or extra-source-files?
by Hackage 29 Jul '08

29 Jul '08

#213: allow wildcards in data-files or extra-source-files? ----------------------------+----------------------------------------------- Reporter: duncan | Owner: Type: enhancement | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------+----------------------------------------------- Some packages end up with very very long fields for data-files or extra- source-files. Is this a problem? Should we allow some kind of globing to make it a bit less verbose? {{{ extra-source-files: cbits/*.h c-sources: cbits/*.c }}} On the other hand, there is a danger with globing that you pick up too many files. Perhaps we can counter that by making the globing rather restrictive, like no subdirectory recursion, and only allowing *'s in the final directory component, so no {{{data/*/*}}} patterns. Is there a sweet spot that allows conciseness without including too much accidentally? -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/213> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 6

[Hackage] #232: implement extended syntax for package version ranges
by Hackage 29 Jul '08

29 Jul '08

#232: implement extended syntax for package version ranges ----------------------------+----------------------------------------------- Reporter: duncan | Owner: nominolo Type: enhancement | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------+----------------------------------------------- Revised proposal: http://www.haskell.org/pipermail/cabal- devel/2008-January/001786.html Reproduced here: In order to make it easy for package package authors to describe correct and robust dependencies we (Duncan Coutts and I) propose the following syntactical extension to version range specifications in package files. == Wildcards == {{{ ~ 1.2.* ~~> >= 1.2 && < 1.3 }}} in general {{{ ~ x.y.* ~~> >= x.y && < x.(y+1) }}} The "~" operator should be read as "in the range of". Another option would have been "-", but this is used to specify exact versions in other parts of the Cabal infrastructure , e.g., "cabal install xmonad-0.5". This is the most common specification of package dependencies. The assumption is that changes in the last version number indicate only backwards compatible changes, as is suggested in the new package versioning policy [1]. The use of "*" matches the common usage as a wildcard character. == Upward Ranges == {{{ ~ 1.2.1+ ~~> >= 1.2.1 && < 1.3 }}} in general {{{ ~ x.y.z+ ~~> >= x.y.z && < x.(y+1) }}} The idea here is that sometimes packages need a certain patch level or, as in Cabal's case, a stable release and possible further bugfixes. In those cases the wildcard notation is too liberal, however the explicit range description would be too noisy (and possibly error prone). (Also, using "+" instead of "*" serves as better visual distinction.) -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/232> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 5

[Hackage] #258: case confusion - cabal vs Cabal
by Hackage 23 Jun '08

23 Jun '08

#258: case confusion - cabal vs Cabal ----------------------------+----------------------------------------------- Reporter: guest | Owner: Type: defect | Status: new Priority: normal | Milestone: Component: Cabal library | Version: HEAD Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------+----------------------------------------------- The darcs repo for Cabal is at darcs.haskell.org:/home/darcs/cabal yet it contains a file Cabal.cabal, which indeed claims that the name of the package is Cabal. This confusion of letter case is causing problems for me. Can we agree that Cabal is capitalised, and rename the repo dir? That seems like the simplest way to achieve consistency. -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/258> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 4

[Hackage] #187: Paths_pkgname module is hard to use in a library
by Hackage 12 Jun '08

12 Jun '08

#187: Paths_pkgname module is hard to use in a library ------------------------------+--------------------------------------------- Reporter: duncan | Owner: Type: defect | Status: new Priority: normal | Milestone: Component: Cabal | Version: 1.2.2.0 Severity: normal | Keywords: Difficulty: easy (<4 hours) | Ghcversion: 6.8.1 Platform: Linux | ------------------------------+--------------------------------------------- Most users of the auto-generated {{{Paths_pkgname}}} module are programs. With GHC at least they can get away with not specifying the {{{Paths_pkgname}}} in the {{{Other-Modules}}} field. Get away with in the sense that it does not lead to linker errors because {{{ghc --make}}} does dependency chasing and links in all the necessary .o files. However for a library you must specify every module in {{{Exposed- Modules}}} or {{{Other-Modules}}} or one does get linker errors when you try to use the packages. That's because the missing module's .o files will not have been included in the library .a archive. See bug #128. So for a library one must specify {{{Other-Modules: Paths_pkgname}}} but then of course Cabal cannot find that module since it is not in the normal source directories but in {{{dist/build/autogen}}}. So people have to hack around it by adding {{{dist/build/autogen}}} to the {{{Hs-Source-Dirs}}} field. Obviously that's rather unsatisfactory. Probably the right solution is for Cabal to automatically add that directory to the end of the source dir search path. -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/187> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 5

[Hackage] #230: setup haddock ignores LANGUAGE CPP pragma
by Hackage 09 Jun '08

09 Jun '08

#230: setup haddock ignores LANGUAGE CPP pragma ----------------------------------+----------------------------------------- Reporter: ross(a)soi.city.ac.uk | Owner: Type: defect | Status: new Priority: normal | Milestone: Component: Cabal library | Version: 1.2.3.0 Severity: normal | Keywords: Difficulty: normal | Ghcversion: 6.8.2 Platform: | ----------------------------------+----------------------------------------- `setup haddock` preprocesses sources with cpp if `CPP` is included in the `extensions` field, but not if a source file starts with {{{ {-# LANGUAGE CPP #-} }}} This is confusing, because that works with `setup build`. -- Ticket URL: <http://hackage.haskell.org/trac/hackage/ticket/230> Hackage <http://haskell.org/cabal/> Hackage: Cabal and related projects

1 3