GitLab

Helps to disambiguate Cabal's compat modules from cabal-install's.

Snapshot deps now behave more like `cabal-dev add-source`. Implemented by adding
a BuildTreeRefType field to the BuildRef index entry.

Fixes #1321.

refold authored May 01, 2013 and Mikhail Glushenkov committed May 02, 2013

Since we're using 'unless' here, we should use (>) instead of (>=).

Apparently, lazy I/O was involved after all. 'openFile' instead of 'withFile'
was deliberate - see the use of 'unsafeInterleaveIO' in
'packageIndexFromCache'. Fixed by adding a strict version of 'getSourceFiles'.

(patch manually merged into the cabal-1.16 branch)

This allows us to make minor changes to packages after they have been
released, without changing the package .tar.gz file. We already keep
the .cabal file outsite the package in the index and use it for
dependency planning. This already lets us do fixes such as making
dependency constraints tighter. Currently we cannot make dep
constraints more relaxed however, since the original .cabal file is
the one used when we get to the actual configure step.

So with this change, we now use the updated .cabal file for the
configure and build too. So there's more fixes we can do post-release.
In particlar, in combination with easier editing on hackage, this
should help us address the problems around the PVP and open or closed
version constraints. It should allow a system of conservative upper
bounds, but allow editing them when new versions of deps are released
and we find that they happen to work fine.

This allows us to make minor changes to packages after they have been
released, without changing the package .tar.gz file. We already keep
the .cabal file outsite the package in the index and use it for
dependency planning. This already lets us do fixes such as making
dependency constraints tighter. Currently we cannot make dep
constraints more relaxed however, since the original .cabal file is
the one used when we get to the actual configure step.

So with this change, we now use the updated .cabal file for the
configure and build too. So there's more fixes we can do post-release.
In particlar, in combination with easier editing on hackage, this
should help us address the problems around the PVP and open or closed
version constraints. It should allow a system of conservative upper
bounds, but allow editing them when new versions of deps are released
and we find that they happen to work fine.

This patch makes 'parsePackageIndex' return a list of IO actions instead of
parsed package data (producing a 'GenericPackageDescription' from a local build
tree reference requires doing IO). Alternatively, we can make
'parsePackageIndex' impure.

Build tree refs are now correctly loaded by `getSourcePackages`; `cabal index
--list` now benefits from package index caching.

Fixes the problem with generating the haddock documentation contents page for
all installed packages. Previously we were (accidentally) telling haddock to
use all versions of each package and haddock would pick the first (lowest
version). Now we correctly do what we were trying to do all along, which is
to pick only the highest version of each package.

e.g. about 3x faster for cabal info pkgname

Rather than 'available' packages, the source now refers consistently to
'source' packages. This is a bit clearer.

Previously when multiple instances of the same package are installed,
e.g. in global and user dbs, we would often end up selecting the wrong
instance. Now we select the user one consistently which will solve the
problem in most (but not all) cases.

For several commands, including install the -v verbosity level had
far too much useless internall stuff in it. Reduced the amount of
output from configuring the compiler, getting installed package and
the dependency planner. The extra detail is still available via -v3.

So we can now use PackageLocation FilePath or Maybe FilePath to
describe what we know about the fetch status of package tarballs.

And remove import list for Types module, just import it all

Move the ambiguity checking to the only use site
Both normal and substring search now return [(PackageName, [pkg])]

Use "maybeDecompress" to handle broken proxies that transparenty decompress network streams. Closes #622, #686. Cabal update could fail in some cases, see http://trac.haskell.org/http/ticket/109283

It used to return Maybe, now it always gives us a PackageIndex.
This depends on an API change in Cabal-1.9.x.

It works by ignoring the possibility that there could be multiple
installed packages sharing the same source package Id. We just pick
the "top most" one which is usually ok. We make no attempt to check
that we are using consistent installed packages.