GitLab

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'.

Misc fixes here and there

We don't show that command in 'sandbox help' output by default, but if the user
is running 'list-sources', she probably wants to know about it.

Also make it possible to list ignored build tree refs (those present in the
index file but not in the source package DB).

This is like showInstalledPackageInfoField, but uses a simpler output
form. Used by ghc-pkg for its --simple-output mode.

Implement 'sandbox add-source --snapshot'.

Unlike 'cabal-dev', we don't create a tarball, but simply copy the files to a
private location.

Fixes #1143.

Implement per-compiler timestamps

Build detailed test suite pieces with own ComponentLocalBuildInfo.

Bug #1287. Type 'detailed-0.9' test suites were failing to build because the
ComponentLocalBuildInfo for the test suite was being used to build the library
and stub executable. The test suite's ComponentLocalBuildInfo should only be used
to create the CLBIs for the library and stub executable, as they have their own
CLBI constructors. Proactively make the same fix for benchmarks, though there is
no build failure there yet.

Detect bug #1287.

Improvements to 'reinstallAddSourceDeps'

I decided to go this route instead of passing compiler & platform to
'reinstallAddSourceDeps' directly because it makes the side-effect of
'configure' easier to test.

It'll be created on demand by 'configureAction' or 'install'.

They're only used by the sandbox code ATM.

Allows us to reconfigure on reinstall and pass compiler & platform to
'reinstallAddSourceDeps'.

We don't want to defult to building executables the dynamic way too.

Minimal rebuilds for add-source dependencies