Ok, well, if that's the case I'd like to see about remedying that. Anyone have any thoughts as to how to best go about this? I'm not clear on exactly what info lives where, especially across systems. Entirely manual population would be a (barely) acceptable fallback. On Tue, Mar 18, 2014 at 9:36 AM, Dan Burton wrote:

I have wished for this on multiple occasions. I don't believe such a thing exists as of yet.

-- Dan Burton

On Tue, Mar 18, 2014 at 9:26 AM, David Thomas wrote:

...

Is there a way to extract this? I'm looking to make it easier for newcomers to my project to get things building across different linux distros.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

This is not an immediate solution but I can imagine listing NIX packages as dependencies inside a cabal file, then NIX would create a sandbox with these dependencies and cabal would build in that sandbox. The advantages are that you're not messing around with the underlying operating system's packages, NIX handles all dependencies transitively, and everything is specified declaratively. Sounds like a nice GSoC project :-) You could also do cross GHC version's builds as GHC itself can be sandboxed. Quite intriguing. Michal On Tue, Mar 18, 2014 at 2:16 PM, Michal Antkiewicz < mantkiew@gsd.uwaterloo.ca> wrote:

Certainly NIX is an interesting approach. It already comes with a large base of dependencies, a format for specifying them. NIX can be installed in any Linux distro and serve as an environment for building packages. That might provide a cross-distribution solution to the native dependency problem.

See, a nice post by Oliver Charles

http://ocharles.org.uk/blog/posts/2014-02-04-how-i-develop-with-nixos.html

Michal

On Tue, Mar 18, 2014 at 1:59 PM, David Thomas wrote:

...

Ok, well, if that's the case I'd like to see about remedying that. Anyone have any thoughts as to how to best go about this? I'm not clear on exactly what info lives where, especially across systems. Entirely manual population would be a (barely) acceptable fallback.

On Tue, Mar 18, 2014 at 9:36 AM, Dan Burton wrote:

...

I have wished for this on multiple occasions. I don't believe such a thing exists as of yet.

-- Dan Burton

On Tue, Mar 18, 2014 at 9:26 AM, David Thomas wrote:

...

Is there a way to extract this? I'm looking to make it easier for newcomers to my project to get things building across different linux distros.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Hmm, but then, not all package managers/repositories use the same names for packages. So even if there were a way to extract the required libraries from cabal, feeding that information to the package manager isn't going to be trivial at all. (And frankly, looking at build systems for other languages, I've never seen anything that does this - even autotools doesn't really do a lot more than check whether a library is available). That said, even if we could just get a list of required libraries out of cabal, in a somewhat human and machine readable format, that would be a huge win in itself. On Tue, Mar 18, 2014 at 11:48:33AM -0700, David Thomas wrote:

One option that just occurred to me would be to allow passing a script to cabal that could be passed the extra-libraries (if any), and could install if it knew the relevant OS packages (or NIX packages), or abort with a cleaner error message.

Actually, wrapping ghc might be sufficient (though not ideal).

On Tue, Mar 18, 2014 at 11:29 AM, Michal Antkiewicz < mantkiew@gsd.uwaterloo.ca> wrote:

...

This is not an immediate solution but I can imagine listing NIX packages as dependencies inside a cabal file, then NIX would create a sandbox with these dependencies and cabal would build in that sandbox. The advantages are that you're not messing around with the underlying operating system's packages, NIX handles all dependencies transitively, and everything is specified declaratively. Sounds like a nice GSoC project :-) You could also do cross GHC version's builds as GHC itself can be sandboxed. Quite intriguing.

Michal

On Tue, Mar 18, 2014 at 2:16 PM, Michal Antkiewicz < mantkiew@gsd.uwaterloo.ca> wrote:

...

Certainly NIX is an interesting approach. It already comes with a large base of dependencies, a format for specifying them. NIX can be installed in any Linux distro and serve as an environment for building packages. That might provide a cross-distribution solution to the native dependency problem.

See, a nice post by Oliver Charles

http://ocharles.org.uk/blog/posts/2014-02-04-how-i-develop-with-nixos.html

Michal

On Tue, Mar 18, 2014 at 1:59 PM, David Thomas wrote:

...

Ok, well, if that's the case I'd like to see about remedying that. Anyone have any thoughts as to how to best go about this? I'm not clear on exactly what info lives where, especially across systems. Entirely manual population would be a (barely) acceptable fallback.

On Tue, Mar 18, 2014 at 9:36 AM, Dan Burton wrote:

...

I have wished for this on multiple occasions. I don't believe such a thing exists as of yet.

-- Dan Burton

On Tue, Mar 18, 2014 at 9:26 AM, David Thomas

...

wrote:

...

Is there a way to extract this? I'm looking to make it easier for newcomers to my project to get things building across different linux distros.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Is that cabal file fields for the current project, or cabal file fields for all dependencies? On Tue, Mar 18, 2014 at 1:59 PM, Rogan Creswick wrote:

On Tue, Mar 18, 2014 at 11:48 AM, David Thomas wrote:

...

One option that just occurred to me would be to allow passing a script to cabal that could be passed the extra-libraries (if any), and could install if it knew the relevant OS packages (or NIX packages), or abort with a cleaner error message.

You can do this sort of thing in a custom Setup.hs (using build-type: custom). Depending on the hook you use, you can get access to the cabal file fields there directly.

--Rogan

...

Actually, wrapping ghc might be sufficient (though not ideal).

On Tue, Mar 18, 2014 at 11:29 AM, Michal Antkiewicz < mantkiew@gsd.uwaterloo.ca> wrote:

...

This is not an immediate solution but I can imagine listing NIX packages as dependencies inside a cabal file, then NIX would create a sandbox with these dependencies and cabal would build in that sandbox. The advantages are that you're not messing around with the underlying operating system's packages, NIX handles all dependencies transitively, and everything is specified declaratively. Sounds like a nice GSoC project :-) You could also do cross GHC version's builds as GHC itself can be sandboxed. Quite intriguing.

Michal

On Tue, Mar 18, 2014 at 2:16 PM, Michal Antkiewicz < mantkiew@gsd.uwaterloo.ca> wrote:

...

Certainly NIX is an interesting approach. It already comes with a large base of dependencies, a format for specifying them. NIX can be installed in any Linux distro and serve as an environment for building packages. That might provide a cross-distribution solution to the native dependency problem.

See, a nice post by Oliver Charles

http://ocharles.org.uk/blog/posts/2014-02-04-how-i-develop-with-nixos.html

Michal

On Tue, Mar 18, 2014 at 1:59 PM, David Thomas

...

wrote:

...

Ok, well, if that's the case I'd like to see about remedying that. Anyone have any thoughts as to how to best go about this? I'm not clear on exactly what info lives where, especially across systems. Entirely manual population would be a (barely) acceptable fallback.

On Tue, Mar 18, 2014 at 9:36 AM, Dan Burton

...

wrote:

...

I have wished for this on multiple occasions. I don't believe such a thing exists as of yet.

-- Dan Burton

On Tue, Mar 18, 2014 at 9:26 AM, David Thomas < davidleothomas@gmail.com> wrote:

> Is there a way to extract this? I'm looking to make it easier for > newcomers to my project to get things building across different linux > distros. > > _______________________________________________ > Haskell-Cafe mailing list > Haskell-Cafe@haskell.org > http://www.haskell.org/mailman/listinfo/haskell-cafe > >

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Yeah, it seems that there are a number of potentially good options that are going to vary by distro and situation, which is why I've been thinking "hook to run arbitrary script". Sufficiently populating the lookup table for that script for any given (Project, Distro) shouldn't be hard, and collecting those we should be able to quickly 1) round the rough edges off of the more common cases without requiring too much manual intervention, and 2) have a reasonable path for the less common cases. On Wed, Mar 19, 2014 at 5:48 AM, Alexander V Vershilov < alexander.vershilov@gmail.com> wrote:

In Gentoo Linux we use a common ebuild format that allowes to add native dependencies to packages. To create a ebuild from cabal file we use tool called hackport [1].

This approach allow to bump dependencies when some authors are to lazy to bump them themselves, even in presence of pull requests, and add compatibility patches. Also it's possible to create a desktop-icons, additional files, filter test/bench programs that should not be installed into system. Also it supports a dependency on package flags and profiling.

However this approach doesn't scale on other distros in a easy way, one possibility is to create static packages using emerge, but this area require some more research.

[1] http://hackage.haskell.org/package/hackport

-- Alexander

On 18 March 2014 20:26, David Thomas wrote:

...

Is there a way to extract this? I'm looking to make it easier for newcomers to my project to get things building across different linux distros.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

-- Alexander

Hi Alexander,

Some packages require existing executables, e.g. git-annex require lsof. There were some more examples of packages that doesn't have pkg-config but require additional programs to be installed.

if the package requires system resources that the Cabal file doesn't declare, then this is a bug in the Cabal file, no? Take care, Peter

Hi Alexander,

Such requirement (dependency on executable) can't be covered by the cabal file (at least with 'simple' build type).

it's true that Cabal lacks a field specifically for such run-time dependencies, but in the absence of such a stanza using "build-tools" is a good workaround. The build process can then determine the complete path of, say, lsof and bake it into the binary. It's a work-around, though, abusing build-tools in this way is far from ideal. Take care, Peter

Hi David,

...

what exactly do you mean when by "package requirements"?

Well, the most important thing is to make sure people can successfully build things. Next most important is to minimize the manual intervention required (and the degree to which that manual intervention is spread out through time).

I am sorry, but I still don't understand what exactly you mean. Could you please give a specific example of what exactly it is that you want? Let's use the package "attoparsec" as an example. Could you please write down the information you consider "package requirements" for that package? Take care, Peter

Hi David,

postgresql-libpq clearly depends on the libpq C library (and possibly other things) as listed in the Extra-Libraries field in its cabal file, *as do anything that require postgresql-libpq* which fact does *not* directly show up in their cabal file. Walking the tree and pulling that info is a possibility, but it needs to be done after dependency-resolution in case requirements change version to version.

yes, it's a complicated situation. The information stored in a Cabal file must be "finalized" before Cabal can use it to install the package. According to [1], this process depends on the following parameters: - Cabal flags chosen by the user, - the set of installed (or install-able) packages, - the current architecture and operating system, - the Haskell compiler used to build the package, and - version constraints chosen by the user. Thus, a given package may depend on an external library 'foo' if you compile it with GHC 7.6.3, but it depends on 'bar' if you compile it with GHC 7.4.2, and the same variance occurs recursively in all its dependencies. It would be nice to know the list of external dependencies for any given package, but in fact it's not "the list" -- it's many possible lists, because the outcome depends on the exact environment in which the package is built. A rough estimate of these external dependencies is available from Nix, though. The package postgresql-libpq, for example, has its Hackage home page at [2]. That page links to the NixOS build information at [3], which shows the following dependencies: acl-2.2.52 attr-2.4.47 bash-4.2-p45 binutils-2.23.1 coreutils-8.21 gcc-4.8.2 gcc-wrapper-4.8.2 ghc-7.6.3 glibc-2.18 gmp-5.1.3 gnutar-1.27.1 linux-headers-3.7.1 ncurses-5.9 perl-5.16.3 postgresql-9.2.7 readline-6.2 zlib-1.2.8 That list is not be accurate for all environments, but it's a start. Take care, Peter [1] http://hackage.haskell.org/package/Cabal-1.18.1.2/docs/Distribution-PackageD... [2] http://hackage.haskell.org/package/postgresql-libpq [3] http://hydra.nixos.org/build/9708550#tabs-runtime-deps