[stable] Follow-up on dlang#9221
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

Fix Issue 20649 - Trait isZeroInit gives false in certain scenarios
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

The dmd binary must have permission to be executable otherwise it would be
impossible to launch it without manually setting permissions.

This patch sets the 644 permission by default for all the other files installed
through installRelativeFiles while the 755 permissions will be set for the dmd binary.

This fixes the issue #20662

(Fix #20662) Set the DMD binary as executable
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

Fix Issue 20610 - const in a .tupleof loop is ignored
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

…e to __ArrayCast not being written

Fix Issue 20130 - ICE when casting from string to other array type du…
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

Fix Issue 20656 - cannot compile live function without -preview=dip1021
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

Fix Issue 20687 - Allow member function address as const initializer
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

…rash dmd

…ializer"

Fix Issue 20692 - Nested alias declarations and getMember trait can c…

…gnored

`Type`-derived AST nodes do not support the same kind of semantic analysis that
`Dsymbol`-derived one do. The issue was that in some codes, semantic analysis on
the type was performed before it was performed on the declaration,
and chasing those cases turned out to be endless, hence the minor refactoring.

[stable] Fix issue 20700 -  Forward references leads to extern(C++, class|struct) being ignored
merged-on-behalf-of: unknown

dlang#10960)

Fix Issue 18651 - ice: assert in glue.d:777 when building these three…
merged-on-behalf-of: Mathias LANG <pro.mathias.lang@gmail.com>

The early expression sema of `exp.e1`, the assert condition, was a major
problem. E.g.:

```
struct S
{
    int a;

    this(this) {}

    bool opEquals(S) @safe
    {
        return true;
    }
}

void foo()
{
    S a;
    assert(S() == a); // S().opEquals(a)
}
```

Here, an lvalue `a` is passed to `S.opEquals()`, which takes the arg by
value. The early `CallExp` sema already lowers the argument expression
`a` to a temporary, to something like
`S __copytmp = (__copytmp = a).__postblit()`. *That* expression (instead
of `a`) was previously stored as a further `__assertOp` temporary, and
the original argument replaced by that temporary (without postblit).

Previous lowering:
assert(S(0).opEquals(((S __assertOp3 = (S __copytmp2 = (__copytmp2 = a).__postblit();) , __assertOp3 = __copytmp2;) , __assertOp3)), _d_assert_fail(S(0), __assertOp3));

New lowering:
assert(S(0).opEquals(((S __copytmp2 = (__copytmp2 = a).__postblit();) , __copytmp2)), _d_assert_fail(S(0), a));

A related 2nd issue concerns rvalue arguments - the early `CallExp` sema
detects that no temporary is needed (move; elide postblit and dtor in the
caller); the rvalue was previously correctly stored as `__assertOp`
temporary, but as it is now an lvalue (possibly accessed a 2nd time for
the `_d_assert_fail()` call if the assertion fails), it must be copied +
postblitted. This can be fixed by deferring the `CallExp` sema until the
arguments have been replaced by required temporaries (and hence possibly
promoted from rvalues to lvalues).

void rval()
{
    static S nonPure()
    {
        __gshared int counter;
        ++counter;
        return S(123);
    }
    assert(S() == nonPure());
}

Previous lowering:
assert(S(0).opEquals(((S __assertOp12 = nonPure();) , __assertOp12)), _d_assert_fail(S(0), __assertOp12));

New lowering:
(S __assertOp11 = nonPure();) , assert(S(0).opEquals(((S __copytmp12 = (__copytmp12 = __assertOp11).__postblit();) , __copytmp12)), _d_assert_fail(S(0), __assertOp11));

For druntime's test/exceptions/src/assert_fail.d.

[stable] Fix misc. issues wrt. temporaries for -checkaction=context lowering
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

…d errors occurred in compilation

When running semantic on __traits(compiles), errors are gagged, so
the saved errors count is global.gaggedErrors, and not global.errors.

So the condition always evaluated as true if any real errors occured
during compilation, and every subsequent __traits(compiles) errors too,
whether it is valid syntax or not.

fix Issue 20730 - [REG 2.091] __traits(compiles) fails if any ungagged errors occurred in compilation
merged-on-behalf-of: Rainer Schuetze <rainers@users.noreply.github.com>

The const flag being set on the return type triggered switch errors

Cleaning up after dlang#10952 and dlang#10988.

Remove qualifiers from toPrec return type
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

[stable] 2 small frontend fix-ups after v2.091.0
merged-on-behalf-of: Nicholas Wilson <thewilsonator@users.noreply.github.com>

PR dlang#6621 took the code for pending `ConditionalDeclaration`s from
doc.d but missed the early return.

Fix Issue 20742 - dmd -X (JSON output) includes uncompiled symbols
merged-on-behalf-of: Steven Schveighoffer <schveiguy@users.noreply.github.com>

Conflicts:
	test/fail_compilation/fob2.d