Hi again,
I just made another test without the reordered headers and I cannot get
the libs to compile.
We actually require the reordering for the library to compile as
otherwise we get the error:
"'SOCKADDR_STORAGE' does not name a type." from winhttp.h
All tests are with MinGW-W64 4.8.2.
Best,
Timo
On 28.04.2014 21:10, Rom Walton wrote:
I've applied patches 1, 3, and 4.
I need to investigate 2 a little bit more. One of the first things
winsock.h does is pull in windows.h.
Can you point me to the recommendation?
----- Rom
-----Original Message-----
From: boinc_dev [mailto:[email protected]] On Behalf Of
Timo Strunk
Sent: Monday, April 28, 2014 12:34 PM
To: [email protected]
Subject: [boinc_dev] [PATCH] MinGW lib build currently broken
Hi everybody,
The commit afb6dcc6f3166f004327ec6fe32f39c577b40b38
"MGR & Client: Massive code clean-up. Remove as much of the
LoadLibrary[...]"
breaks MinGW compilation because of the line #include <winternl.h>
The stock MinGW distribution does not carry this include, but the newer
MinGW-W64(http://sourceforge.net/apps/trac/mingw-w64/browser/trunk/mingw
-w64-headers/include/winternl.h?rev=5224),
does. However also here the compilation fails with the commit.
We (Thomas Koch and me) therefore tried fixing this situation for
MinGW-W64:
The problem is simply that MinGW-W64's winternl.h defines another
THREAD_STATE enum (also in diagnostics_win.h), which is similar to the
one of the BOINC lib, but has members, which carry different names.
The reason this is possible, is that there is no official THREAD_STATE
enum; Microsoft's implementation is a simple DWORD and some #defines for
the STATEs. (I'm definitely no expert on this, so I might be completely
wrong. I did not find any _official_ THREAD_STATE definition).
1.) As I think it will be really hard to change the naming of the MinGW
compiler, the attached patch changes the naming in the BOINC library and
does not define the enums in case of __MINGW32__ defines, as it relies
on the ones found in winternl.h.
In this way we don't change functionality in case of MSVC compiles
(please test, I only checked MinGW), but still allow compilation with
MinGW-W64. Another occurence happens in the installer directory, but I
did
not touch this, because I think it's not intended to be MinGW
compatible anyways.
These are:
0001-Renamed-Thread_State-enum-names-to-mingw-standard-in.patch
0003-Renamed-StateWaiting-again-to-StateWait-as-per-mingw.patch
2.) The attached patches also include <delayimp.h> for MinGW, as it
seems to be required nowadays.
0004-MinGW-also-requires-delayimp-now.patch
This patch here is not required for making everything compile:
3.) winsock.h is now included prior to windows.h as recommended. Also
WIN32_LEAN_AND_MEAN is defined, because I saw no reason not to. If it is
defined somewhere else or even required to be absent and I didn't
notice, please remove it. It might not be good style to define it just
like that in a header.
0002-Include-winsock-before-windows.h.patch
With these 4 patches, mingw still doesn't compile using the supplied
makefile, but requires -fpermissive to be set. The reasons are multiple
deprecated string conversions.
After the commit mentioned at the beginning of the mail Windows 2000
support was dropped for the lib. Maybe the switch to Winsock2.h (Win98
and NT4 and upwards) and ws2tcpip.h (ipv6) should also be made now as
the code still seems to use winsock.h and it should be a simple change
of one include. However: Again, I'm no expert on the subject and there
might be good legacy reasons to keep everything as is.
Best,
Timo
_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.
