Re: rpcrt4: Dynamically load iphlpapi

[Rob Shearman]

2008/8/5 Juan Lang <juan.lang(a)gmail.com>:
> This has a similar motivation to the last patch I sent.  In rpcrt4's
> case, iphlpapi is needed at most once per process, so dynamically load
> it rather than keeping it around all the time.
> @@ -304,18 +304,23 @@ #define ADDRESS_BYTES_NEEDED 6
>
> static RPC_STATUS RPC_UuidGetNodeAddress(BYTE *address)
>  {
> +    static const WCHAR iphlpapi[] = { 'i','p','h','l','p','a','p','i',0 };
>      int i;
>      DWORD status = RPC_S_OK;
>
> ULONG buflen = sizeof(IP_ADAPTER_INFO);
>      PIP_ADAPTER_INFO adapter = HeapAlloc(GetProcessHeap(), 0, buflen);
> +    HMODULE lib = LoadLibraryW(iphlpapi);
> +    DWORD (WINAPI *pGetAdaptersInfo)(PIP_ADAPTER_INFO, PULONG) = NULL;

Hi Juan,

Why manually load iphlpapi here instead of using the automatic delayed
import mechanism like you used for ws2_32? I assume it was because you
reasoned that a hard dependency was unnecessary because the code has a
fallback path, but I think you needed to make this clear in the
changelog.

-- 
Rob Shearman

rpcrt4: Dynamically load iphlpapi

[Juan Lang]

> Why manually load iphlpapi here instead of using the automatic delayed
> import mechanism like you used for ws2_32? I assume it was because you
> reasoned that a hard dependency was unnecessary because the code has a
> fallback path, but I think you needed to make this clear in the
> changelog.

I can try to make the changelog clearer, but that wasn't my
motivation.  As I mentioned in the email with the patch, iphlpapi is
only needed once per process by rpcrt4, so I manually load it so that
I can also unload it as soon as it's done.  This is in contrast to
ws2_32, where iphlpapi may be needed more than once.
--Juan

rpcrt4: Dynamically load iphlpapi

[Rob Shearman]

2008/8/10 Juan Lang <juan.lang(a)gmail.com>:
> > Why manually load iphlpapi here instead of using the automatic delayed
> > import mechanism like you used for ws2_32? I assume it was because you
> > reasoned that a hard dependency was unnecessary because the code has a
> > fallback path, but I think you needed to make this clear in the
> > changelog.
>
> I can try to make the changelog clearer, but that wasn't my
> motivation.  As I mentioned in the email with the patch, iphlpapi is
> only needed once per process by rpcrt4, so I manually load it so that
> I can also unload it as soon as it's done.  This is in contrast to
> ws2_32, where iphlpapi may be needed more than once.

Ah, yes, I missed that part. The patch is fine with me.

-- 
Rob Shearman