Potential incorrect values in `getUsableMemory()` and `getFreeSwapMemory()` on Windows

[marcosfrm]

The Windows implementation of getMemoryInfo() appears to have a calculation error.

cbang/src/cbang/os/win/WinSystemInfo.cpp

Lines 68 to 83 in f6386fd

	uint64_t WinSystemInfo::getMemoryInfo(memory_info_t type) const {
	MEMORYSTATUSEX info;
	info.dwLength = sizeof(MEMORYSTATUSEX);
	GlobalMemoryStatusEx(&info);
	switch (type) {
	case MEM_INFO_TOTAL: return (uint64_t)info.ullTotalPhys;
	case MEM_INFO_FREE: return (uint64_t)info.ullAvailPhys;
	case MEM_INFO_SWAP: return (uint64_t)info.ullAvailPageFile;
	case MEM_INFO_USABLE:
	return (uint64_t)(info.ullAvailPhys + info.ullAvailPageFile);
	}
	return 0;
	}

In getUsableMemory(), the code sums ullAvailPhys + ullAvailPageFile. However, according to the Windows API documentation, ullAvailPageFile already includes both free RAM and available pagefile space ("The maximum amount of memory the current process can commit ...": https://learn.microsoft.com/en-us/windows/win32/api/sysinfoapi/ns-sysinfoapi-memorystatusex). This causes the free RAM to be counted twice.

Similarly, getFreeSwapMemory() returns ullAvailPageFile directly, which includes free physical RAM instead of showing only the isolated swap space.

Suggested fix:

• MEM_INFO_USABLE: Should just return info.ullAvailPageFile. Note that this represents the unallocated system commit limit (free RAM + free swap), which differs fundamentally from Linux's MemAvailable. It is impossible to calculate a true MemAvailable equivalent using only the MEMORYSTATUSEX structure; achieving that requires switching to PERFORMANCE_INFORMATION and calculating it via perfInfo.PhysicalAvailable * perfInfo.PageSize (which includes the system cache/standby list: https://learn.microsoft.com/en-us/windows/win32/api/psapi/ns-psapi-performance_information), as done in Python's psutil:

    unsigned long long totalPhys, availPhys, totalSys, availSys, pageSize;
    PERFORMANCE_INFORMATION perfInfo;

    if (!GetPerformanceInfo(&perfInfo, sizeof(PERFORMANCE_INFORMATION))) {
        psutil_oserror();
        return NULL;
    }
    // values are size_t, widen (if needed) to long long
    pageSize = perfInfo.PageSize;
    totalPhys = perfInfo.PhysicalTotal * pageSize;
    availPhys = perfInfo.PhysicalAvailable * pageSize;
    totalSys = perfInfo.CommitLimit * pageSize;
    availSys = totalSys - perfInfo.CommitTotal * pageSize;

https://github.com/giampaolo/psutil/blob/v7.2.2/psutil/arch/windows/mem.c#L23-L39

• MEM_INFO_SWAP should return info.ullAvailPageFile - info.ullAvailPhys to isolate the pagefile, which is roughly equivalent to Linux's SwapFree in /proc/meminfo.

Here is how Zabbix does it (it separates total swap and available swap):

		ms_ex.dwLength = sizeof(MEMORYSTATUSEX);

		(*zbx_get_GlobalMemoryStatusEx())(&ms_ex);

		real_swap_total = ms_ex.ullTotalPageFile > ms_ex.ullTotalPhys ?
				ms_ex.ullTotalPageFile - ms_ex.ullTotalPhys : 0;
		real_swap_avail = ms_ex.ullAvailPageFile > ms_ex.ullAvailPhys ?
				ms_ex.ullAvailPageFile - ms_ex.ullAvailPhys : 0;

https://github.com/zabbix/zabbix/blob/7.4.11/src/libs/zbxsysinfo/win32/swap.c#L130-L137

Note: this issue and its proposed fix were generated with help from Gemini and require manual review.

[marcosfrm]

Gemini is going nuts! The ullAvailPhys field in MEMORYSTATUSEX is equivalent to the PhysicalAvailable field in PERFORMANCE_INFORMATION-- that's exactly what you'd expect for MEM_INFO_USABLE. No need to call GetPerformanceInfo(). Something like:

  switch (type) {
  case MEM_INFO_TOTAL:
    return (uint64_t)info.ullTotalPhys;

  case MEM_INFO_FREE:
  case MEM_INFO_USABLE:
    return (uint64_t)info.ullAvailPhys;

  case MEM_INFO_SWAP:
    return info.ullAvailPageFile > info.ullAvailPhys ?
      (uint64_t)(info.ullAvailPageFile - info.ullAvailPhys) : 0;
  }

[marcosfrm]

Here's the output of the relevant MEMORYSTATUSEX fields on a system with low available memory (values in MB):

ullTotalPhys:                    7988
ullAvailPhys:                    1098
ullAvailPageFile:                2847
ullAvailPageFile - ullAvailPhys: 1749

ullAvailPageFile + ullAvailPhys: 3945

In Task Manager, the reported free memory is "1.1 GB". Therefore, using ullAvailPageFile + ullAvailPhys as MEM_INFO_USABLE is completely wrong.

And returning ullAvailPageFile as MEM_INFO_SWAP seems to overestimate the value... since it's not used anywhere, I think we can just adopt Zabbix's logic instead (ullAvailPageFile - ullAvailPhys).