[Lldb-commits] [lldb] [lldb] Add a MemoryRegionInfo cache at a process stop (PR #202509)

[Felipe de Azevedo Piovezan via lldb-commits]

================
@@ -0,0 +1,42 @@
+//===-- MemoryRegionInfoCache.h ---------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLDB_TARGET_MEMORYREGIONINFOCACHE_H
+#define LLDB_TARGET_MEMORYREGIONINFOCACHE_H
+
+#include "lldb/Target/MemoryRegionInfo.h"
+#include "lldb/Utility/RangeMap.h"
+
+namespace lldb_private {
+class MemoryRegionInfoCache {
+public:
+  MemoryRegionInfoCache(Process &process)
+      : m_region_infos(), m_process(process) {}
+
+  /// Remove all cached entries.
+  void Clear();
+
+  /// Remove cached information about region containing \a addr, if any.
+  void Flush(lldb::addr_t addr, lldb::addr_t size);
+
+  /// Locate the memory region that contains load_addr.
+  Status GetMemoryRegionInfo(lldb::addr_t load_addr,
----------------
felipepiovezan wrote:

>having it in a separate class akin to how MemoryCache (a much larger amount of code) seemed natural to me.

I agree a separate class is good and we shouldn't change this bit.

My point is about this:

> Process has a Process::GetMemoryRegionInfo API, and subclasses have a Process::DoGetMemoryRegionInfo,

I think `Process::GetMemoryRegionInfo` should query the cache, if it's there just return.
If it's not there, call the virtual `DoGetMemoryRegion`, then call `MemoryRegionInfoCache::AddRegion`.

The implications of the suggested design are:

1. No more circular dependency between process and the cache
2. No more failure state in `GetMemoryRegionInfo`. It's either in the cache or it is not, no error possible.

https://github.com/llvm/llvm-project/pull/202509