Announcing the first release in the stack-2.1 series!
The Stack 2 release represents a series of significant changes
to how Stack works internally. For the vast majority of cases,
these changes are backwards compatible, in that existing projects
will continue to build in the same way with Stack 2 as they did
with Stack 1. The large version bump is due to the fundamental
internal changes to cache handling, database storage (using SQLite
in place of binary files), implicit snapshots (which greatly
improve the precompiled cache), and moving to Pantry. We have also
removed some less used features, as listed in the release notes
below.
See https://haskellstack.org/ for installation and
upgrade instructions.
Changes since v1.9.3
Major changes:
- Switch over to pantry for managing packages. This is a major
change to Stack’s internals, and affects user-visible behavior in a
few places. Some highlights:
- Drop support for multiple package indices and legacy
00-index.tar
style indices. See #4137.
- Support for archives and repos in the
packages
section has been removed. Instead, you must use
extra-deps
for such dependencies.
packages
now only supports local filepaths.
- Add support for Git repositories containing (recursive)
submodules.
- Addition of new configuration options for specifying a “pantry
tree” key, which provides more reproducibility around builds, and
(in the future) will be used for more efficient package content
downloads. You can also specify package name and version for more
efficient config parsing. * NOTE The new
stack freeze
command provides support for
automatically generating this additional information.
- Package contents and metadata are stored in an SQLite database
in place of files on the filesystem. The
pantry
library can be used for interacting with these contents.
- Internally, Stack has changed many datatypes, including moving
to Cabal’s definition of many data types. As a result of such
changes, existing cache files will in general be invalidated,
resulting in Stack needing to rebuild many previously cached builds
in the new version. Sorry :(.
- A new command,
stack freeze
has been added which
outputs project and snapshot definitions with dependencies pinned
to their exact versions.
- The
ignore-revision-mismatch
setting is no longer
needed, and has been removed.
- Overriding GHC boot packages results in any other GHC boot
packages depending on it being no longer available as a dependency,
such packages need to be added explicitly when needed. See [#4510]
(https://github.com/commercialhaskell/stack/issues/4510).
- Cabal solver integration was not updated to support newer
cabal-install
versions so stack solver
command was removed as well as a related option
--solver
from stack new
and stack
init
.
- Upgrade to Cabal 2.4
- Note that, in this process, the behavior of file globbing has
been modified to match that of Cabal. In particular, this means
that for Cabal spec versions less than 2.4,
*.txt
will
match foo.txt
, but not foo.2.txt
.
- Remove the
stack image
command. With the advent of
Docker multistage builds, this functionality is no longer useful.
For an example, please see Building Haskell
Apps with Docker.
- Support building GHC from source (experimental)
- Stack now supports building and installing GHC from source. The
built GHC is uniquely identified by a commit id and an Hadrian
“flavour” (Hadrian is the newer GHC build system), hence
compiler
can be set to use a GHC built from source
with ghc-git-COMMIT-FLAVOUR
stack.yaml
now supports a
configure-options
, which are passed directly to the
configure
step in the Cabal build process. See
#1438
- Remove support for building GHCJS itself. Future releases of
Stack may remove GHCJS support entirely.
- Support for lock files for pinning exact project dependency
versions
Behavior changes:
stack.yaml
now supports snapshot
: a
synonym for resolver
. See #4256
stack script
now passes -i -idir
in
to the ghc
invocation. This makes it so that the
script can import local modules, and fixes an issue where
.hs
files in the current directory could affect
interpretation of the script. See #4538
- When using
stack script
, custom snapshot files
will be resolved relative to the directory containing the
script.
- Remove the deprecated
--upgrade-cabal
flag to
stack setup
.
- Support the
drop-packages
field in
stack.yaml
- Remove the GPG signing code during uploads. The GPG signatures
have never been used yet, and there are no plans to implement
signature verification.
- Remove the
--plain
option for the
exec
family of commands
- Always use the
--exact-configuration
Cabal
configuration option when building (should mostly be a
non-user-visible enhancement).
- No longer supports Cabal versions older than
1.19.2
. This means projects using snapshots earlier
than lts-3.0
or nightly-2015-05-05
will
no longer build.
- Remove the
stack docker cleanup
command. Docker
itself now has
docker image prune
and
docker container prune
, which you can use
instead.
- Interleaved output is now turned on by default, see #4702.
In addition, the
packagename>
prefix is no longer
included in interelaved mode when only building a single
target.
- The
-fhide-source-paths
GHC option is now enabled
by default and can be disabled via the
hide-source-paths
configuration option in
stack.yaml
. See #3784
- Stack will reconfigure a package if you modify your
PATH
environment variable. See #3138.
- For GHC 8.4 and later, disable the “shadowed dependencies”
workaround. This means that Stack will no longer have to force
reconfigures as often. See #3554.
- When building a package, Stack takes a lock on the dist
directory in use to avoid multiple runs of Stack from trampling
each others’ files. See #2730.
- Stack will check occassionally if there is a new version
available and prompt the user to upgrade. This will not incur any
additional network traffic, as it will piggy-back on the existing
Hackage index updates. You can set
recommend-stack-upgrade:
false
to bypass this. See #1681.
stack list-dependencies
has been removed in favour
of stack ls dependencies
.
- The new default for
--docker-auto-pull
is enabled.
See #3332.
Other enhancements:
- Support MX Linux in get-stack.sh. Fixes #4769.
- Defer loading up of files for local packages. This allows us to
get plan construction errors much faster, and avoid some
unnecessary work when only building a subset of packages. This is
especially useful for the curator use case.
- Existing global option
--color=WHEN
is now also
available as a non-project-specific yaml configuration parameter
color:
.
- Adopt the standard proposed at https://no-color.org/, that color should
not be added by default if the
NO_COLOR
environment
variable is present.
- New command
stack ls stack-colors
lists the styles
and the associated ‘ANSI’ control character sequences that stack
uses to color some of its output. See stack ls stack-colors
--help
for more information.
- New global option
--stack-colors=STYLES
, also
available as a non-project-specific yaml configuration parameter,
allows a stack user to redefine the default styles that stack uses
to color some of its output. See stack --help
for more
information.
- British English spelling of ‘color’ (colour) accepted as an
alias for
--color
, --stack-colors
,
stack ls stack-colors
at the command line and for
color:
and stack-colors:
in yaml
configuration files.
- New build option
--ddump-dir
. (See #4225)
- Stack parses and respects the
preferred-versions
information from Hackage for choosing latest version of a package
in some cases, e.g. stack unpack packagename
.
- The components output in the
The main module to load is
ambiguous
message now include package names so they can be
more easily copy-pasted.
- Git repos are shared across multiple projects. See #3551
- Use en_US.UTF-8 locale by default in pure Nix mode so programs
won’t crash because of Unicode in their output #4095
- Add
--tree
to ls dependencies
to list
dependencies as tree. #4101
- Add
--pedantic
to ghci
to run with
-Wall
and -Werror
#4463
- Add
--cabal-files
flag to stack ide
targets
command.
- Add
--stdout
flag to all stack ide
subcommands.
- Use batches when unregistering packages with
ghc-pkg
. (See #2662)
get-stack
script now works on Windows CI machines
of Appveyor, Travis and Azure Pipelines. See #4535/
- Show snapshot being used when
stack ghci
is
invoked outside of a project directory. See #3651
- The script interpreter now accepts a
--extra-dep
flag for adding packages not present in the snapshot. Currently,
this only works with packages from Hackage, not Git repos or
archives.
- When using the script interpreter with
--optimize
or --compile
, Stack will perform an optimization of
checking whether a newer executable exists, making reruns
significantly faster. There’s a downside to this, however: if you
have a multifile script, and change one of the dependency modules,
Stack will not automatically detect and recompile.
stack clean
will delete the entire
.stack-work/dist
directory, not just the relevant
subdirectory for the current GHC version. See #4480.
- Add
stack purge
as a shortcut for stack
clean --full
. See #3863.
- Both
stack dot
and stack ls
dependencies
accept a --global-hints
flag to
bypass the need for an installed GHC. See #4390.
- Add the
stack config env
command for getting shell
script environment variables. See #620.
- Less verbose output from
stack setup
on Windows.
See #1212.
- Add an optional
ignore-expiry
flag to the
hackage-security
section of the
~/.stack/config.yaml
. It allows to disable timestamp
expiration verification just like cabal
--ignore-expiry
does. The flag is not enabled by default so
that the default functionality is not changed.
- Include default values for most command line flags in the
--help
output. See #893.
- Set the
GHC_ENVIRONMENT
environment variable to
specify dependency packages explicitly when running test. This is
done to prevent ambiguous module name errors in
doctest
tests.
get-stack
script now works on Windows CI machines
of Appveyor, Travis and Azure Pipelines. See #4535
- Warn when a Docker image does not include a
PATH
environment variable. See #2472
- When using
system-ghc: true
, Stack will now find
the appropriate GHC installation based on the version suffix,
allowing you to more easily switch between various system-installed
GHCs. See #2433.
stack init
will now support create a
stack.yaml
file without any local packages. See
#2465
- Store caches in SQLite database instead of files.
- No longer use “global” Docker image database
(
docker.db
).
- User config files are respected for the script command. See
#3705,
#3887.
- Set the
GHC_ENVIRONMENT
environment variable to
-
to tell GHC to ignore any such files when GHC is new
enough (>= 8.4.4), otherwise simply unset the variable. This
allows Stack to have control of package databases when running
commands like stack exec ghci
, even in the presence of
implicit environment files created by cabal new-build
.
See #4706.
- Use a database cache table to speed up discovery of installed
GHCs
- You can specify multiple
--test-arguments
options.
See #2226
- Windows terminal width detection is now done. See #3588
- On Windows, informs users if the ‘programs’ path contains a
space character and further warns users if that path does not have
an alternative short (‘8 dot 3’) name, referencing the
local-programs-path
configuration option. See #4726
- Add
--docker-mount-mode
option to set the Docker
volume mount mode for performance tuning on macOS.
Bug fixes:
- Ignore duplicate files for a single module when a Haskell
module was generated from a preprocessor file. See #4076.
- Only track down components in current directory if there are no
hs-source-dirs found. This eliminates a number of false-positive
warnings, similar to #4076.
- Handle a change in GHC’s hi-dump format around
addDependentFile
, which now includes a hash. See
yesodweb/yesod#1551
- Fix
subdirs
for git repos in
extra-deps
to match whole directory names. Also fixes
for subdirs: .
. See #4292
- Fix for git packages to update submodules to the correct state.
See #4314
- Add
--cabal-files
flag to stack ide
targets
command.
- Don’t download ghc when using
stack clean
.
- Support loading in GHCi definitions from symlinked C files.
Without this patch, Stack will try to find object files in the
directory pointed to by symlinks, while GCC will produce the object
files in the original directory. See #4402
- Fix handling of GitHub and URL templates on Windows. See
commercialhaskell/stack#4394
- Fix
--file-watch
not responding to file
modifications when running inside docker on Mac. See #4506
- Using
--ghc-options
with stack script
--compile
now works.
- Ensure the detailed-0.9 type tests work. See #4453.
- Extra include and lib dirs are now order-dependent. See
#4527.
- Apply GHC options when building a
Setup.hs
file.
See #4526.
- Stack handles ABI changes in FreeBSD 12 by differentiating that
version from previous.
- Help text for the
templates
subcommand now
reflects behaviour in stack 1.9 — that it downloads and shows a
help file, rather than listing available templates.
- Fix detection of aarch64 platform (this broke when we upgraded
to a newer Cabal version).
- Docker: fix detecting and pulling missing images with
--docker-auto-pull
, see #4598
- Hackage credentials are not world-readable. See #2159.
- Warnings are dumped from logs even when color is enabled. See
#2997
stack init
will now work for cabal files with
sublibraries. See #4408
- When the Cabal spec version is newer than the global Cabal
version, build against the snapshot’s Cabal library. See #4488
- Docker: fix detection of expected subprocess failures. This
fixes downloading a compatible
stack
executable when
the host stack
is not compatible with the Docker image
(on Linux), and doesn’t show an unnecessary extra error when the
in-container re-exec’ed stack
exits with failure.
- The
stack ghci
command’s
--ghc-options
flag now parses multiple options. See
#3315.
Thanks to all our contributors for this
release:
- Adam Bergmark
- Akshay Mankar
- Aleksey Kozin
- ALeX Kazik
- Alexey Kozin
- Alexey Kuleshevich
- Anders Kaseorg
- Anders Kiel Hovgaard
- Artyom Kazak
- Bastian Krol
- Björn Gohla
- Brad Neimann
- Brandon Chinn
- Colin Woodbury
- Cthulhu.Den
- dadepo
- Daniel Gröber
- David Baynard
- David Spies
- David Vollbracht
- dbaynard
- Emanuel Borsboom
- favonia
- flip111
- Florjan Bartol
- Gleb Popov
- Hussein Ait-Lahcen
- Isumi Feng
- Janfel
- Jurijs Oniscuks
- Kayla Ngan
- Kirill Zaborsky
- Magicloud
- Matt Audesse
- Mauricio Fierro
- Michael Sloan
- Michael Snoyman
- Mihai Maruseac
- Mike Pilgrem
- Neil Mitchell
- NeonGraal
- Niklas Hambüchen
- penteract
- Phil de Joux
- Raghu Kaippully
- Sibi
- Sibi Prabakaran
- Simon Hengel
- skapazzo
- Sylvain HENRY
- Sylvain Henry
- Taekyung
- Timothy Clem
- Tom Sydney Kerckhove
- Valery V. Vorotyntsev
- Vance Palacio
- waddlaw
- 欧阳泽
Subscribe to our blog via email
Email subscriptions come from our Atom feed and are handled by Blogtrottr. You will only receive notifications of blog posts, and can unsubscribe any time.
Do you like this blog post and need help with Next Generation Software Engineering, Platform Engineering or Blockchain & Smart Contracts? Contact us.