chore(path): Make a commonly triggered assert more verbose ...

We see this assert triggered frequently, even sometimes in production,
so at least make it easier to track down the data involved.

Fix skipRegistry parsing unittest ...

Contrary to the comment in the unittest, the code is checking for the
wrong value, should be `default_` not `all`. The unittest passed
because the function threw an exception not because of the value of
skipRegistry but because there's a missing '}'.

Signed-off-by: Andrei Horodniceanu <a.horodniceanu@proton.me>

update version to v1.39.0-beta.1

Merge remote-tracking branch 'upstream/stable' into merge_stable

update version to v1.38.1

Filesystem: Provide the string writeFile overload at the interface level ...

The string overload is just a convenience function and there is little
chance that it wll ever need to be overloaded.

MockFS: Simplify existsFile / existsDirectory

MockFS: Simplify implementation of setAttributes / setTimes ...

Hoist the lookup from FSEntry to MockFS to reduce the duplication.

Refactor filesystem abstraction to and use it in production ...

The filesystem abstraction needs to be known to production to avoid
too much code duplication. So far we have used functions in
`PackageManager` that the `TestPackageManager overrides.

With this change, we can remove them, and instead just pass an interface.
On the long run, we should be able to do dependency injection simply
by passing a different object (or overriding) the `Dub` class.

An initial attempt made `FSEntry` inherit from the `Filesystem` interface,
however that attempt was doomed as it conflates two different abstractions:
`FSEntry` is a "node" in the filesystem, which we want to act on in one
of multiple ways: move to another node, or take an action. On the other hand,
`Filesystem` is a more generic interface and takes a path as most of its
function's argument. A `path` and an `FSEntry` have a lot of overlap
(they are a reference to a node in a tree) and the resulting interface
would have been very clunky.

Trivial: Pass an absolute path when building TestDub in legacy UT ...

This way we avoid a call to getcwd.

Merge remote-tracking branch 'upstream/stable'

CI: Reduce dependency on std.path / std.file in one unittest ...

This will make migrating to the abstracted IO much easier.

Drop support for LDC v1.26.0 ...

We only support the last 10 versions, so LDC should have been upgraded in the previous PR.
In addition, we can remove a superfluous __VERSION__ statement, which haseen superfluous
for quite a while now.

Look up `dub.selections.json` files in parent directories too (#2272) ...

* Look up `dub.selections.json` files in parent directories too

In case the root package directory doesn't contain a `dub.selections.json`
file, look in parent directories too and load the first (deepest) one.

This allows using a 'central' `dub.selections.json` file for a repository
containing multiple dub projects, making it automatically apply to all
builds in that source tree if located in the repository root directory
(unless a local `dub.selections.json` overrides it).

Such an 'inherited' selections file is never mutated when running dub for
a nested project, i.e., changes are always saved to a *local*
`dub.selections.json` file.

Merge remote-tracking branch 'upstream/stable' into merge_stable

update version to v1.38.1-rc.1

LDC: Don't keep around temporary object files for static libs (#2690) ...

* LDC: Don't keep around temporary object files for static libs

This was a workaround for older LDC versions. Since LDC v1.25,
`-cleanup-obj` implies that the temporary object files are placed
into a unique temp directory (per compiler invocation), so that
object file collisions across parallel dub invocations cannot happen
anymore.

Trivial: Remove internal deprecations ...

Neither commandline not dub.internal are part of the public API for the library,
hence those deprecations were unnecessary.

Merge pull request #2928 from Geod24/mlang/Fix2830 ...

VersionRange: Always output minor version for ~> and ^

Tests: Implement the ability to mock the registry too ...

So far tests have limited themselves to the local state,
but we want to expand them to test the behavior of dub
when fetching packages, so we need a convenient way
to add packages to the registry.

FSEntry: Set a default time for all entries ...

Otherwise we are not able to write the entries in a Zip file,
as we get an exception from std.zip.
This is a workaround and a proper fix would be to implement
a Clock system.

Work around file locking dependency in PackageManager ...

The `dub.internal.utils : lockFile` function does not land itself to mocking quite well,
as it exposes a struct containing a `std.stdio : File` instance.
The path of least resistance, for now, is to access `store_` directly from
TestPackageManager by overriding `store`.

Do not use `Package.storeInfo` in `PackageManager.store` ...

`Package.storeInfo` does IO directly which conflicts with our
dependency injection approach. Instead inline the code and
use `this.writeFile` directly.

ToolchainRequirements: Handle '*' while parsing PackageRecipe ...

PR 2926 added the possibility that '*' is serialized to PackageRecipe
instead of '>=0.0.0' but we need the parser to handle it.

VersionRange: Always output minor version for ~> and ^ ...

Dub was correct in outputting only the major version,
however this can lead to some confusion for users
as pointed out in #2830. Accepting ~>5 but only exposing
two states (~>5.0 and ~>5.0.0) will hopefully make
things a bit more obvious.

Fix #2901: VersionRange.toString() doesn't handle >0.0.0 ...

The `VersionRange.toString` method was not ideal and dropping
this rare case. A tiny improvement that was also done in the
process is to give higher priority to `matchesAny` so that
we get a more user-friendly output, given how common it is
(e.g. for `ToolchainRequirement`).

FSEntry: Throw an Exception when trying to read non-existent file ...

FSEntry.lookup can return null, and it would lead to a SEGV
with the attribute check, so add the forgotten check first.

PackageManager: Make lookup lazy again ...

We want to avoid loading the whole cache whenever lookup is called,
as most invocations of dub should result in a call to lookup,
since dependency resolution is much rarer than building with an
existing dub.selections.json.

Clear an unprinted deprecation during dub build ...

This deprecation is currently not printed by DMD,
but shows up if '-de' is used while compiling.
This solution is a workaround, as a proper fix
would mean not using the function in the first place,
but that requires a more in-depth rework.

Move the I/O part of Project.loadSelections to PackageManager ...

By doing so we can remove a sizeable amount of duplication as well
as the last `Dub.loadPackage` override. The resulting API is not
ideal, for example it doesn't allow to distinguish between parsing
errors and non-existent file, but will have to do for now.