Replace Prelude.head/tail/init/last with NonEmpty

tests: avoid blocking on git tagging

alex 3.2.5 reordered imports

validate.sh: Fix cabal-plan invocation

I haven't understood 100%, but something about having the following in
my ~/.gitconfig:

    [tag]
      gpgsign = true
      forceSignAnnotated = true

means that creating these tags block on my editor exiting and providing
the annotated tag commit message.  I also didn't investigate all the
options, but I tested that this one fixes the issue for me.

In recent versions of cabal-plan the --builddir= flag has become a
non-global flag; i.e. it needs to be placed after the sub-command rather
than before.

`validate.sh` isn't tested in CI so...
[ci skip]

* Restructure Includes3 tests so that I can also test Hackage case.

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

* Compute correct install directories for instantiated Backpack components

Previously, we would compute elabInstallDirs once when configuring a
component, and then reuse the exactly same install directories for
every instantiation of the package.  But this is wrong, since we're
installing the header/object files for each instantiation to a different
directory.  We refactor install directory computation into a helper
function and then call it again at instantiation time to refresh
the install directories.

For some reason, this bug ONLY manifests for packages installed from
Hackage; it seems install dirs are not respected for inplace packages.

Fixes #6005

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

* Test fix

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>

* Copy backpack version-guard from other backpack testcases

Honour suffix and prefix on install

Use Data.List.NonEmpty and avoid some partial functions

Fix typo in Configuration chapter [ci skip]

Start GHCi with the main module in scope

edmundnoble authored 5 years ago and Dale Wijnand committed 5 years ago

Add safe alternatives to some partial functions in
Distribution.Utils.Generic, and removes the originals from
Distribution.Compat.Prelude

Uncomment versions of functions in Distribution.Compat.Parsing
that return non-empty lists

Co-authored-by: Dale Wijnand <dale.wijnand@gmail.com>

Allow forks to trigger CI builds but at the same time exclude builds in forks from triggering notifications.

See also https://docs.travis-ci.com/user/conditions-v1

Bump from GHC 8.6.4 to 8.6.5

CI: Add travis_retry when uploading binaries to S3

But then exclude builds in forks from the notifications.

Javier Neira authored 5 years ago and Daniel Gröber (dxld) committed 5 years ago

Use project config instead of global one to get store-dir and logs-dir.

Fixes #5925 

Fix spelling error in bug report template [ci skip]

Implement --benchmark-options for v2-bench

This implements lots of plumbing to allow the `--benchmark-option(s)`
flags to be used with `v2-bench`, analgous to `v2-test`'s
`--test-option(s)` flag.

Fixes #6209.

On Windows dropExeExtension doesn't actually drop the exe extension.

We have

    splitExtension "foo.exe" == ("foo", ".exe")

but the code is expecting just "exe" for the extension field.

Among other things this makes guessToolFromGhcPath behave
unexpectedly. Since takeVersionSuffix can't see past the extension the
`null suf` case in mkGuesses will trigger which causes the versioned
program path candidates to not be looked for.

Exe dependency less lexeme

Change legacy note to point to cabal.readthedocs nix-style section

doc: Linkify some :pkg-fields:

Allow failures in OSX jobs

- They use 7.10.3 and 8.0.2
- timeout quite often.
- Let's have them, but let them fail without red mark