hub.darcs.net :: darcs -> darcs-reviewed -> changes

This property occasionally fails with Named patches. Furthermore, we get exceptions when trying to shrink the test complaining about invalid patch names. This needs further investigation.

These properties now work for any Mergeable patch type.

This generalizes RepoPatch properties to any Mergeable patch type. Since we want to generate Named patches with meaningful explicit dependencies, we have to keep track of the names of patches that have been applied to a repo (model). This requires the introduction of yet another (somewhat ad-hoc) class RepoApply to add a method 'patchNames' to all testable patch types. The default implementation that returns [] is used for all patch types except for Named patches.

Recorded as a separate patch for easier review.

This gets rid of the whole Unwitnessed stuff (Darcs.Test.Patch.WSub, Darcs.Test.Patch.Properties.GenericUnwitnessed), renames the module, and mostly returns list of Sealed2 to avoid coercions.

The cheap solution here is to set the permissions of the newly created _darcs/patch_index/ to that of its parent directory _darcs/. Using _darcs/ has the advantage that darcs now automatically fixes the permissions of existing patch_index directories.

This creates a symbolic link to the system's "security" command (if it exists) and adapts the reduced PATH so that it is found by darcs.

• remove test repo before creating it
• remove creation of unused fake 'vi' program (and simplify PATH)
• add comments to the second ("bad editor") test

They changed the default for TLS.CLientSupported to use TLS.ciphersuite_strong instead of the empty list.

Using cabal run instead of runghc inside of Setup.hs ensures that we use the same ghc version that cabal uses, avoiding dependency errors.

Otherwise we would have to add quite a number of extra-deps, even with nightly-2024-05-20. This also changes the conditional for depending on crypton-connection-0.4.0 in the cabal file to ghc >= 9.6.

Versions 1.3.8 up to 1.3.8.4 of the directory package have a bug in their implementation of copyFile on Windows.

GHC 9.10 complains about it, not sure why earlier versions don't.

The NC.settingClientSupported member of the default value for NC.TLSSettings already contains this setting, so we use that default and only change TLS.supportedExtendedMainSecret. This is semantically simpler and more robust in case TLSSettingsSimple acquires additional members. Unfortunately, doing nested record updates is a rather verbose affair in Haskell w/o lenses, so the code is somewhat less readable now.