Here are my patches against branch-2.14 leading up to a new release 2.14.3
that can be built (on Linux) against the whole range of 8.x versions of ghc.

My plan is to push these patches some time tomorrow.

Ganesh, if you happen to find the time to take a quick look, please do so.
Or if you need a bit more time, just tell me and I'll hold off.

11 patches for repository http://darcs.net/releases/branch-2.14:

patch 7f9e8d76769e21fde94395a06d2b2a99cb57b5dc
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Feb 19 09:45:56 CET 2020
  * Setup.hs: allow use of darcs as a cabal subproject
  
  Creation and later installation of the man page make sense only if we build
  a darcs executable. When darcs is used as a cabal subproject, only the
  library is built. So when building we check that we have a darcs exe and
  when installing we check that the manpage is actually there.

patch 447f754fd0613f221cdf6174e67a2e020d71bdb6
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:45:05 CEST 2020
  * for base >=4.13 define fail in MonadFail instances

patch 171b4a816581593f440b080e2618aea59beaca10
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:52:36 CEST 2020
  * replace fail with error in some places
  
  This is nothing but a quick and dirty stop-gap measure to make things
  compile with base >= 4.13. While some of the ocurrences indeed indicate
  bugs, and are therefore correctly replaced with error, there are other
  places where these are clearly normal user-level failures (such as a match
  expression not finding a matching patch). Morre generally, the use of 'fail'
  in monadic code that is not necessarily IO based (ApplyMonad, TreeMonad) has
  always been a sore point in Darcs, and we should address that more cleanly
  in the development version.

patch 37eec5e40add46e4868e6b30f8b2a42f1034aaaa
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:20:42 CEST 2020
  * make Setup.hs build against Cabal >= 3.0
  
  Cabal >= 3.0 no longer has sdist hooks, and rewriteFile and
  autogenModulesDir have been removed from the API.

patch dba3c61f2a57eb2ab788058964b63b83186a38f3
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:12:58 CEST 2020
  * loosen upper bound for haskeline to <0.9
  
  This is required to find a valid build plan for ghc >= 8.8.

patch df04723a659dc500bb4520c56d87d952da72c033
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:14:29 CEST 2020
  * loosen upper bound for shelly to <1.10
  
  This is required to find a build plan with ghc >= 8.8.
  
  A comment in the cabal file claims we cannot use shelly-1.9 because of two
  open issues. However, these issues only concern windows; but for windows we
  use an even stricter bound in a separate build-depend stanza, where we
  require < 1.7.2. I have moved the comment to this other stanza. The upshot
  is that building against ghc versions >= 8.8 is currently not supported on
  windows.

patch 6834a978fe0b8dbc569481c141c922cfef31a9b1
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:23:39 CEST 2020
  * bump version to 2.14.3 in darcs.cabal

patch f4425b3ab99f2ce5eb4b24480623766d8a05c0e9
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 11:54:25 CEST 2020
  * add autogen-modules:Version to library section in cabal file
  
  Note: it is important to ignore cabal warnings that complain about missing
  other-modules entries for the Version module. While these warnings disappear
  if "other-modules: Version" is added to the darcs executable section, this
  will break building with ghc-8.0.

patch c7531db20647c7a8a6cda923f8bd6214f37803d7
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:55:45 CEST 2020
  * loosen upper bound for base to < 4.15
  
  This means we now support building with ghc versions from 8.0 up to 8.10.

patch 12f4d8b19aa95c9e65b10d2c2c4e94f80761a383
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 12:47:18 CEST 2020
  * adapt release/release.sh for use with modern cabal

patch 87fac2a14a25dde5efd8e308a5d53cd4184dd127
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 12:57:07 CEST 2020
  tagged 2.14.3

Attachments

• patch-preview.txt
• setup_hs_-allow-use-of-darcs-as-a-cabal-subproject_0.dpatch
• unnamed

I rebased and updated the bundle because the version file generation was
broken: we must re-generate release/distributed-* from scratch before using
cabal sdist. Also added a few lines to the README and pulled the "remove our
own optimisation settings in darcs.cabal" patch from reviewed to avoid
problems when uploading the tar ball and also to allow the initial cabal
build --disable-optimisation to work (to save time when releasing).

13 patches for repository http://darcs.net/releases/branch-2.14:

patch 7f9e8d76769e21fde94395a06d2b2a99cb57b5dc
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Feb 19 09:45:56 CET 2020
  * Setup.hs: allow use of darcs as a cabal subproject
  
  Creation and later installation of the man page make sense only if we build
  a darcs executable. When darcs is used as a cabal subproject, only the
  library is built. So when building we check that we have a darcs exe and
  when installing we check that the manpage is actually there.

patch 447f754fd0613f221cdf6174e67a2e020d71bdb6
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:45:05 CEST 2020
  * for base >=4.13 define fail in MonadFail instances

patch 171b4a816581593f440b080e2618aea59beaca10
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:52:36 CEST 2020
  * replace fail with error in some places
  
  This is nothing but a quick and dirty stop-gap measure to make things
  compile with base >= 4.13. While some of the ocurrences indeed indicate
  bugs, and are therefore correctly replaced with error, there are other
  places where these are clearly normal user-level failures (such as a match
  expression not finding a matching patch). Morre generally, the use of 'fail'
  in monadic code that is not necessarily IO based (ApplyMonad, TreeMonad) has
  always been a sore point in Darcs, and we should address that more cleanly
  in the development version.

patch 37eec5e40add46e4868e6b30f8b2a42f1034aaaa
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:20:42 CEST 2020
  * make Setup.hs build against Cabal >= 3.0
  
  Cabal >= 3.0 no longer has sdist hooks, and rewriteFile and
  autogenModulesDir have been removed from the API.

patch dba3c61f2a57eb2ab788058964b63b83186a38f3
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:12:58 CEST 2020
  * loosen upper bound for haskeline to <0.9
  
  This is required to find a valid build plan for ghc >= 8.8.

patch df04723a659dc500bb4520c56d87d952da72c033
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:14:29 CEST 2020
  * loosen upper bound for shelly to <1.10
  
  This is required to find a build plan with ghc >= 8.8.
  
  A comment in the cabal file claims we cannot use shelly-1.9 because of two
  open issues. However, these issues only concern windows; but for windows we
  use an even stricter bound in a separate build-depend stanza, where we
  require < 1.7.2. I have moved the comment to this other stanza. The upshot
  is that building against ghc versions >= 8.8 is currently not supported on
  windows.

patch 6834a978fe0b8dbc569481c141c922cfef31a9b1
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:23:39 CEST 2020
  * bump version to 2.14.3 in darcs.cabal

patch c7531db20647c7a8a6cda923f8bd6214f37803d7
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:55:45 CEST 2020
  * loosen upper bound for base to < 4.15
  
  This means we now support building with ghc versions from 8.0 up to 8.10.

patch 8f3b7bb7685d16003ba26050397b3022d0427119
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Jul 19 17:55:14 CEST 2019
  * remove our own optimisation settings in darcs.cabal
  
  Letting cabal be aware of optimisation makes it easier to switch between
  optimised and unoptimised builds, since cabal nowadays puts unoptimised
  builds in a separate directory.
  I tested with a number CPU-intensive test runs (darcs-test -t conflict) that
  this causes no performance regressions: the default optimisation level
  chosen by cabal (-O1) gives the same or better results than the -O2 we
  passed explicitly.

patch c5a52546d0831fe8dfc873d9bb88612c5f8c9ace
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 11:54:25 CEST 2020
  * add autogen-modules:Version to library section in cabal file
  
  Note: it is important to ignore cabal warnings that complain about missing
  other-modules entries for the Version module. While these warnings disappear
  if "other-modules: Version" is added to the darcs executable section, this
  will break building with ghc-8.0.

patch 4f9cdae8307ec21cd49844affe9864e2eb045639
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 15:00:04 CEST 2020
  * add some elementary build instructions to the README.md

patch a8dd180284eb6afc2c38403714f892079bff4552
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 12:47:18 CEST 2020
  * adapt release/release.sh for use with modern cabal

patch 927f669ded76752c86e53e937f72a7babad7fdfc
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 15:19:08 CEST 2020
  tagged 2.14.3

Attachments

• patch-preview.txt
• setup_hs_-allow-use-of-darcs-as-a-cabal-subproject.dpatch
• unnamed

One more remark: the test harness is broken. It runs only the unit
tests, not the shell tests. I have fiddled for a while with pulling
patches related to harness/test.hs but in the end gave up. I don't think
this is critical. Apart from some of the fail -> error replacements and
a few (very conservatively) updated dependencies, the new release should
be functionally identical to 2.14.2.

This updated bundle includes a mitigation for issue2643 and fixes a typo in
patch description of

  * replace fail with error in some places

This rebase highlighted a known problem when rebasing conflicted patches:
contrary to expectations, suspending and then unsuspending a trailing set of
patches may not result in the same state. It took me a number of rounds of
amending the formerly conflicted patch (originally pulled from screened):

  * remove our own optimisation settings in darcs.cabal

to restore the original conflict resolution. I added "[rebased for
branch-2.14]" to the patch name. A darcs compiled from current screened was
used to do the rebase.

IMO this highlights the urgency of getting Ganesh's fullUnwind-refactor of
rebase into screened as soon as possible. I also think we should strive to
make a release 2.16.1 in the near future. This release should contain the
new rebase code but include the darcs-3 format only as technology preview,
subject to further format changes, with appropriate warnings attached.

14 patches for repository http://darcs.net/releases/branch-2.14:

patch 7f9e8d76769e21fde94395a06d2b2a99cb57b5dc
Author: Ben Franksen <ben.franksen@online.de>
Date:   Wed Feb 19 09:45:56 CET 2020
  * Setup.hs: allow use of darcs as a cabal subproject
  
  Creation and later installation of the man page make sense only if we build
  a darcs executable. When darcs is used as a cabal subproject, only the
  library is built. So when building we check that we have a darcs exe and
  when installing we check that the manpage is actually there.

patch 447f754fd0613f221cdf6174e67a2e020d71bdb6
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:45:05 CEST 2020
  * for base >=4.13 define fail in MonadFail instances

patch 509d39b51686d640d75a26e96c8ebd1db471d2db
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 08:52:36 CEST 2020
  * replace fail with error in some places
  
  This is nothing but a quick and dirty stop-gap measure to make things
  compile with base >= 4.13. While some of the ocurrences indeed indicate
  bugs, and are therefore correctly replaced with error, there are other
  places where these are clearly normal user-level failures (such as a match
  expression not finding a matching patch). More generally, the use of 'fail'
  in monadic code that is not necessarily IO based (ApplyMonad, TreeMonad) has
  always been a sore point in Darcs, and we should address that more cleanly
  in the development version.

patch 62907865e80438640ec4701f177ebb0723ccaa2e
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:20:42 CEST 2020
  * make Setup.hs build against Cabal >= 3.0
  
  Cabal >= 3.0 no longer has sdist hooks, and rewriteFile and
  autogenModulesDir have been removed from the API.

patch 46bebd0a1511232b216d387392d964cae6fcac2c
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:12:58 CEST 2020
  * loosen upper bound for haskeline to <0.9
  
  This is required to find a valid build plan for ghc >= 8.8.

patch 5d63e665634c964d7b23dcf9c28efc72d6a6f947
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:14:29 CEST 2020
  * loosen upper bound for shelly to <1.10
  
  This is required to find a build plan with ghc >= 8.8.
  
  A comment in the cabal file claims we cannot use shelly-1.9 because of two
  open issues. However, these issues only concern windows; but for windows we
  use an even stricter bound in a separate build-depend stanza, where we
  require < 1.7.2. I have moved the comment to this other stanza. The upshot
  is that building against ghc versions >= 8.8 is currently not supported on
  windows.

patch 2534270733d7bf96c1c4ab50bfe3fe839978e2a3
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 10:23:39 CEST 2020
  * bump version to 2.14.3 in darcs.cabal

patch 5fbe6c4998cc9c5a134eb5dcbf27b7dab1f19451
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 09:55:45 CEST 2020
  * loosen upper bound for base to < 4.15
  
  This means we now support building with ghc versions from 8.0 up to 8.10.

patch ea6d8f53e2dcf26dca51fd60aebc781b74488e7e
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Jul 19 17:55:14 CEST 2019
  * remove our own optimisation settings in darcs.cabal [rebased for branch-2.14]
  
  Letting cabal be aware of optimisation makes it easier to switch between
  optimised and unoptimised builds, since cabal nowadays puts unoptimised
  builds in a separate directory.
  I tested with a number CPU-intensive test runs (darcs-test -t conflict) that
  this causes no performance regressions: the default optimisation level
  chosen by cabal (-O1) gives the same or better results than the -O2 we
  passed explicitly.

patch e5258386152e607434832285a27dec7596c86ea6
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 11:54:25 CEST 2020
  * add autogen-modules:Version to library section in cabal file
  
  Note: it is important to ignore cabal warnings that complain about missing
  other-modules entries for the Version module. While these warnings disappear
  if "other-modules: Version" is added to the darcs executable section, this
  will break building with ghc-8.0.

patch 8322b50f998b2f2244937d629ad66becf0f4df1e
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 15:00:04 CEST 2020
  * add some elementary build instructions to the README.md

patch cacc6c44c3e8ae94e3aa3456e3f0ded9d00371f5
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Apr 21 12:47:18 CEST 2020
  * adapt release/release.sh for use with modern cabal

patch 82fe27cc6295a2e55936b3b1ff10fcb4966e1e5e
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 10:45:44 CEST 2020
  * mitigate issue2643 with a better error message
  
  If the patch index is corrupt, output the name of the currupt file and
  suggest its removal.

patch 2e321946297422298e914752086523e5f3c3aa26
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 10:48:24 CEST 2020
  tagged 2.14.3

Attachments

• patch-preview.txt
• setup_hs_-allow-use-of-darcs-as-a-cabal-subproject.dpatch
• unnamed

Oops, forgot to add a CHANGELOG entry. Had to obliterate the tag.

2 patches for repository http://darcs.net/releases/branch-2.14:

patch c7afc1c153a22308145eddadeeb592a35e3cfeb5
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 15:27:26 CEST 2020
  * add changelog entry for release 2.14.3

patch 36ed8456ac9f6db8443bc3670bf1e4260bfa38b8
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 15:29:43 CEST 2020
  tagged 2.14.3

Attachments

• add-changelog-entry-for-release-2_14_3.dpatch
• patch-preview.txt
• unnamed

Fix a changelog item that was a bit obscure.

I really hope this is the last time I have to send an update to this bundle.

2 patches for repository http://darcs.net/releases/branch-2.14:

patch 64b5842ba3e32290d9eb479c8281eff4cfcea443
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 15:27:26 CEST 2020
  * add changelog entry for release 2.14.3

patch 0e7dc307b2b1a4f27861588428d753cc468cd137
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 15:32:47 CEST 2020
  tagged 2.14.3

Attachments

• add-changelog-entry-for-release-2_14_3.dpatch
• patch-preview.txt
• unnamed

I got bored waiting for the trustees to give me upload rights, so I started
to eliminate all the warnings. This went well until I hit
Darcs.Test.Patch.Check which was hard to make portable and warning free. So
I restructured the code along the lines indicated in a comment.

This new release now builds with -fwarn-as-error on all compiler versions
from 8.0 up to 8.10.

13 patches for repository http://darcs.net/releases/branch-2.14:

patch 64b5842ba3e32290d9eb479c8281eff4cfcea443
Author: Ben Franksen <ben.franksen@online.de>
Date:   Fri Apr 24 15:27:26 CEST 2020
  * add changelog entry for release 2.14.3

patch 347aeb4b5c1eccfe00956ac318a2123bca9ef9ca
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:26:07 CEST 2020
  * clean up Darcs.Test.Patch.Check
  
  There was a comment in the code about the strangeness of how in this test
  module returning a Boolean was used to communicate a failure, and how this
  should be replaced with a more conventional way of error handling. This is
  now done, using the MaybeT monad transformer.

patch 91eddc2567ba553bfa695af99d758339b2514b49
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:29:58 CEST 2020
  * move cabal-version to the top of darcs.cabal and remove the '>='
  
  This is how it should be done according to the docs.

patch 2dbe7ff7042a2d909640f468c2d1dd39a00ca38b
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:32:02 CEST 2020
  * darcs.cabal: clean up extra-source-files
  
  Also added C source files in addition to header files.

patch a82b0e34630f0a973afa837edff640078ac2b793
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:33:52 CEST 2020
  * darcs.cabal: add -Wno-missing-home-modules
  
  This is to avoid warnings about missing Version module for darcs exe and
  darcs-test, conditional on ghc >= 8.2.

patch dab25e2f7ce6f57a99b9e10ce09aeb1b5648360e
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:36:35 CEST 2020
  * Setup.hs: disable deprecation warnings

patch 8c9c45236167ecb26d78c6cdfa18eb974be57be6
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:37:02 CEST 2020
  * darcs.cabal: remove redundant version ranges 
  
  The cabal docs claim that if the library declares them, then executables and
  test suites don't have to.

patch 80223edc7108090f9c3231e36d4bdb94e1f52f6a
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:40:58 CEST 2020
  * remove redundant imports
  
  This mostly concerns (<>) imported from Darcs.Util.Printer and (<$>) from
  Control.Applicative. These are now consistently exported from Darcs.Prelude.

patch 67326fec2518f09a57c5ee6f397e416c3714e480
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:56:17 CEST 2020
  * harness: remove unsused instance MonadFail Fail

patch 5e45eb29161bd09364e6fbc05078aaf1fa84edff
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:57:36 CEST 2020
  * Darcs.Patch.ReadMonads: remove unused instance MonadFail SM

patch 5a9f410104fbf9095afaf28fe3818693badd0599
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 20:58:51 CEST 2020
  * disable redundant catchall case for ghc>=8.10
  
  It seems ghc-8.10 can now detect some redundant catchall case matches. The
  redundant case is now disabled to avoid a warning.

patch eee9e8e65667d87c0408ae65ed415601c1c040fe
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 21:04:14 CEST 2020
  * no longer export fromJust from Darcs.Prelude
  
  While used in many places, fromJust is a partial function and we should not
  encourage its unfettered use in Darcs.
  
  This eliminates the last warnings. The code, including the test suite, can
  now be built with all supported compiler versions with -fwarn-as-error.

patch 58ee188d443fdc6f8a687ec823ce350e39c3d6c7
Author: Ben Franksen <ben.franksen@online.de>
Date:   Sun Apr 26 21:06:56 CEST 2020
  tagged 2.14.3

Attachments

• add-changelog-entry-for-release-2_14_3.dpatch
• patch-preview.txt
• unnamed

Release 2.14.3 is now published on hackage. I am self-accepting.