[Lldb-commits] [lldb] [lldb] Fix race in macOS's FindProcessesImpl (PR #204109)

[Jonas Devlieghere via lldb-commits]

================
@@ -684,34 +685,59 @@ static bool GetMacOSXProcessUserAndGroup(ProcessInstanceInfo &process_info) {
   return false;
 }
 
-uint32_t Host::FindProcessesImpl(const ProcessInstanceInfoMatch &match_info,
-                                 ProcessInstanceInfoList &process_infos) {
-  std::vector<struct kinfo_proc> kinfos;
-
+/// Fetches the list of all processes. Returns true on success.
+[[nodiscard]] static bool
+GetProcessList(std::vector<struct kinfo_proc> &kinfos) {
   int mib[3] = {CTL_KERN, KERN_PROC, KERN_PROC_ALL};
 
-  size_t pid_data_size = 0;
-  if (::sysctl(mib, 3, nullptr, &pid_data_size, nullptr, 0) != 0)
-    return 0;
+  // This is an inherently racy API. We have to first query the size for our
+  // buffer and then pass back to sysctl. If more processes spawn between the
+  // size query and the actual call to fetch, then sysctl returns ENOMEM.
+  // We keep retrying until we get a passing result.
+  for (int attempt = 1; attempt < 1024; ++attempt) {
----------------
JDevlieghere wrote:

Nit: Make the retry count a global constant
```suggestion
  static constexpr g_retry_count = 1024;
  for (int attempt = 1; attempt < g_retry_count; ++attempt) {
```

https://github.com/llvm/llvm-project/pull/204109