Monday, October 17, 2016

[uitadwod] Stackage

Stackage for Haskell packages has the curious behavior that packages can disappear from it even though they were perfectly fine.  The cause of such a disappearance of say a package B is as follows: package B was originally pulled in as a dependency of another package A, and the maintainer of package A quit, so package A and all its dependencies, including package B, are candidates to be removed from Stackage.  Package B survives only if it has a direct maintainer in Stackage or is a dependency of another maintained package.

Inspired by the many packages that got dropped when lambdabot got removed from Stackage nightly, e.g., brainfuck.

Although the stated goal of Stackage is a curated collection of Haskell packages, each with an explicit maintainer willing to fix bugs and compilation problems (e.g., with new versions of GHC), I have found that a side feature is more useful: the identification of a large mutually compatible collection of packages without version dependency problems.  Such a side feature -- such a collection -- could be computed automatically without having to have a direct or indirect maintainer for each package in the collection.  I wish such a larger collection existed.

Start with, say, Stackage Nightly and expand it to include every package in Hackage that compiles cleanly and is compatible with Stackage Nightly and with every other package in the expanded collection.  There may be tricky cases of mutually incompatible packages in a potential expanded set which will need to be resolved, e.g., the newest version of A requires an old version of B, and the newest version of B requires an old version of A.  Perhaps resolve such conflicts in favor of the choice which causes the expanded set to be as large as possible.

Tangentially, how can one safely build a package (to test whether it compiles cleanly) if one is not sure whether a package's build script is evil?  Probably some kind of operating system container or sandbox.  Identify packages which use simple, presumably safe, build mechanisms, probably pure Haskell, versus packages which do something unusual, e.g., call a Makefile, which ought to be scrutinized before building.  (Inspired by a build script of software, I think maxima computer algebra, which creepily attempted to send email back to the author every time it was compiled.)

Can compiling a carefully crafted source file with GHC allow the author of the source file to perform arbitrary user-level actions within the operating system?


Derek Elkins said...

Answering your last question: it doesn't take careful craftsmanship; arbitrary code can be executed by Template Haskell. It would be easy enough to try to send an email when a Haskell file is compiled via Template Haskell.

Marian Jancar said...

Can Template Haskell perform (actually execute, not just calculate the AST) IO?