Co-authored-by: Ncarlossanlop <carlossanlop@users.noreply.github.com>

This reverts commit eaa1f05d.

Publishes crossgen as an AOT binary on Windows+Linux x64+ARM64, otherwise publishes as an R2R single file.

Closes #60016

Microsoft.CodeAnalysis.NetAnalyzers
 From Version 7.0.0-preview1.22174.1 -> To Version 7.0.0-preview1.22180.6
Co-authored-by: Ndotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>

Enabling because the underlying issues appear to have been fixed.

https://github.com/dotnet/runtime/issues/56085 
https://github.com/dotnet/runtime/issues/52460

* Refactor metadata reading infrastructure.

* Update src/libraries/System.Text.Json/src/System/Text/Json/Serialization/Converters/Object/ObjectConverter.cs

`ValueNumStore::GetVNFunc` is very hot, in fact so hot that it shows up
right at the top in a profile of SPMI over libraries.pmi, next to
`fgMorphSmpOp`. Instead of copying all args just return a pointer to the
arguments which allows all the cases to be unified and makes the
function small enough for MSVC to inline it.

Most of the work here is to make sure we do not end up with undefined
behavior. I also removed some of the duplication by using a templated
struct for func apps instead of the old VNDefFuncNArg.

Windows x64 SPMI over libraries.pmi:
PIN before: 329991512865
PIN after: 326098430289 (-1.19%)

* Use ROS<byte> instead of byte[] on S.S.C.Cose

* Add TrySign and improve Sign implementation to use less byte[]

* Add TrySign tests

* Address using scope feedback

* Address src feedback

* Refactor tests to avoid duplicated ones

* Remove invalid asserts in Crypto code

* Fix 'new()' without the type on the left-hand side

* Fix ThreadStatic issues in tests

* * Don't use ArrayPool in SignCore
* Add comment describing reusability of encoded protected headers
* Cache toBeSigned

* Don't cache toBeSigned for detached content

* Address nits in tests

Bring back direct dependencies on System.Runtime.CompilerServices.Unsafe for other tfms than net7.0 (#67385)

* Add dependency on Unsafe package

With the removal of the Unsafe package, a few projects regressed which
previously had a direct dependency on the latest version of Unsafe
(6.0.0) by now depending on the transitive Unsafe dependency of i.e.
System.Memory which is a very old version (4.5.*).

Fixing those projects to declare the direct dependency again.

* Cleanup

Found this because it was asserting in one of the Pri-0 tests. We were trying to generate a special instantiating stub for an interface type that was not a dynamic interface castable implementation. I'm adding a regression test too because it has a visible outside behavior besides the assert.

https://github.com/dotnet/runtime/pull/64860 added a TestWaitForExitValidation test that would fail when run on platforms that did not support RemoteExecutor.

This change adds a `ConditionalTheory` to make sure that doesn't happen.

* Load R2R images on osx-arm64 according to Apple's guidelines

* Address review comments

* use MAP_PRIVATE for the mapped file
* correct the condition

* Use PAL_JitWriteProtect once per relocation

* Flush instruction cache during relocation on Apple Silicon

* Fix typo in comment

* Use PAL_JitWriteProtect in the same spots where protection is toggled

This allows to keep the instruction cache flush logic identical between platforms.

* Revert the change for Intel to avoid any regression

* Always unmap the file to avoid leaks + log errors

As discussed offline, according to the current merged wrapper
architecture, for architecture-variant tests all versions
are compiled and the "wrong" variants get skipped at runtime
using the ConditionalFact attribute. Sadly for some of the ILPROJ
tests JITting the "wrong" variant ends up crashing the JIT during
Crossgen2 compilation. We can follow up by figuring out whether
the crash is intentional but for now a simple way to work around
this problem (hitting just a small number of tests) is to mark
them as out-of-process tests so that the wrong variants get
filtered out early and R2R compilation is not triggered for them.

Thanks

Tomas

* Set side-effect flags in constructor for HWIs

* Add test

* Experiment: NULL_CHECK LDOBJs

* Take HWI stores into account in hoisting

* Also fix ordinary stores while we're at it

The contract for whether the LHS of a global ASG(IND, ...)
should be marked with GLOB_REF is not clearly specified.

* Add tests

* Remove "inst_RV_TT" usages from non-XARCH code

a) It is clearer and cheaper to directly use the emitter anyway.
b) "inst_RV_TT" will be made XARCH-only in an upcoming change.

* Refactor "inst_TT" and "inst_RV_TT"

a) Make them XARCH-only.
b) Rewrite them using "OperandDesc".

* Memory containment for FIELD_LIST operands on x86

Enable broader containment for FIELD_LIST operands on x86.

The primary benefits come from containing loads that long
decomposition has left (e. g. LCL_FLDs):

```diff
-       mov      ecx, dword ptr [ebp+08H]
-       mov      edx, dword ptr [ebp+0CH]
-       push     edx
-       push     ecx
+       push     dword ptr [ebp+0CH]
+       push     dword ptr [ebp+08H]
```

* emitIns_A: add function header

* Remove the "UNDONE: ..." printing

Insert an assert instead about what we expect and handle GCInfo-wise.

* Fix Wrong exception for Open*Stream after connection is closed

* Don't handle InvalidState

* Update src/libraries/System.Net.Quic/src/System/Net/Quic/Implementations/MsQuic/MsQuicStream.cs
Co-authored-by: NNatalia Kondratyeva <knatalia@microsoft.com>
Co-authored-by: NNatalia Kondratyeva <knatalia@microsoft.com>

Port Access violation occurring in System.DirectoryServices.ActiveDirectory.ForestTrustRelationshipInformation fix from .netfx to .net-core (#66726)

* foresttrust port from dotnetfx

* fixed nullable reference error

* Fix ordering in default interface method lookup

The previous implementation would find a default implementation in the current type before looking for non-default implementation in the base. Default implementations should be looked at last.

In fact the default implementation in the current type is unreachable if there's a non-default implementation in the base type - the compiler shouldn't even have bothered to emit it into the dispatch map. That can be a separate fix - this fix is still logically correct and necessary to get variant dispatch corner cases correctly.

* Regression test

Deopt is currently disabled for gsharedvt methods, so disable AOTing
gsharedvt methods with clauses to avoid hitting an assert in
emit_llvmonly_landing_pad ().

For OpenSSL 3, we need to always commit the chain to clear out the untrusted intermediates. Otherwise, we started getting details about the partial chain that we don't map to codes.

This fixes the behavior so that an AKI/SKI mismatch reports as a partial chain.

This test was disabled for macOS arm64, adding MEM_RESERVE_EXECUTABLE fixes the issue.

Tracking: https://github.com/dotnet/runtime/issues/60152

This reverts commit 2453f168.

Grammar, consistency fixes

Right now, the Debug.Fail just shows the literal text `Unrecognized
X509VerifyStatusCode:Interop+Crypto+X509VerifyStatusCode`, which isn't
very helpful.

* perf: wasm: Collect helix logs, and artifacts when the build fails

This will be helpful in debugging build issues, and random crashes that
we have been seeing on helix.

* Fix misplaced double quote

* perf: Use '--run-isolated' with wasm projects

* Fix SuperPMI implementation of `appendClassName`

Change #67135 added more and different calls to `appendClassName`, exposing
that it was not correctly implemented in SuperPMI. Fix this, especially, to
handle the case of passing a zero length to determine the correct size output buffer.

Also, add `eeTryGetClassSize` to wrap `getClassSize` and use it in two cases in the
importer that only use the output in JitDump, but was preventing a clean JitDump
replay of some collections.

Change the JIT-EE GUID because the SuperPMI collection data structures were changed.

* Update semantics of `appendClassName` API

Document the semantics more clearly.

Update EE and crossgen2 to conform to the same semantics.

Notable changes:
1. ppBuf argument is allowed to be nullptr, if `*pnBufLen` is zero.
2. If `*pnBufLen` is greater than zero, `*ppBuf` and `*pnBufLen` are
advanced by the number of characters actually written rather than the
number in the full class name string. This will differ in the case of
truncation due to a small output buffer. The advance does not include
the null terminator. Thus, `*ppBuf` will point at the null terminator
on return, and `*pnBufLen` will be at least `1`.
3. The EE will copy as much as possible (i.e., will truncate if necessary),
rather than crashing on a too-small output buffer.
4. Crossgen2 now advances `*ppBuf` and `*pnBufLen` as noted above.
5. SuperPMI implements these new semantics.

* Fix non-Windows build

Set `JIT_BUILD` when building SuperPMI. This disables contracts and some PAL debugging,
including the debugreturn.h debugging that was causing a build break.

Also, adjust a couple loops based on code review feedback.

Change these default values when the jit targets x64 or arm64:

* COMPlus_TC_QuickJitForLoops=1
* COMPlus_TC_OnStackReplacement=1

The upshot is that on x64/arm64 more methods will be jitted at Tier0,
and we will rely on OSR to get out of long-running Tier0 methods.

Other architectures continue to use the old behavior for now, as
OSR is not yet supported for x86 or arm.