STR #3271

GitHub FLTK Project   FLTK News RSS Feed  
  FLTK Apps      FLTK Library      Forums      Links     Login 
 Home  |  Articles & FAQs  |  Bugs & Features  |  Documentation  |  Download  |  Screenshots  ]
 

Return to Bugs & Features | Post Text | Post File | Prev | Next ]

STR #3271

Application:FLTK Library
Status:5 - New
Priority:4 - High, e.g. key functionality not working
Scope:2 - Specific to an operating system
Subsystem:Core Library
Summary:Socket for MS-WINDOWS is UINT_PTR
Version:1.4-feature
Created By:szukw000
Assigned To:Unassigned
Fix Version:Unassigned
Update Notification:

Receive EMails Don't Receive EMails

Trouble Report Files:

Post File ]
Name/Time/Date Filename/Size top right image
 
#1 szukw000
05:55 Dec 27, 2015
Socket-r10957.dif
5k
 
 
#2 szukw000
08:40 Dec 28, 2015
socket.txt
3k
 
bottom left image   bottom right image

Trouble Report Comments:

Post Text ]
Name/Time/Date Text top right image
 
#1 szukw000
05:55 Dec 27, 2015
winsock2.h:

typedef UINT_PTR        SOCKET;

winfried
 
 
#2 AlbrechtS
03:05 Dec 28, 2015
https://msdn.microsoft.com/en-us/library/windows/desktop/aa383751%28v=vs.85%29.aspx#UINT_PTR

<quote>
This type is declared in BaseTsd.h as follows:

#if defined(_WIN64)
 typedef unsigned __int64 UINT_PTR;
#else
 typedef unsigned int UINT_PTR;
#endif
</quote>
 
 
#3 AlbrechtS
03:26 Dec 28, 2015
Thanks for the report, and thanks for the patch. Please tell us why you think that we should apply your patch, i.e. compiler warnings/messages, the platform and compiler and so on.

On a first glance I can see that MS defines SOCKET as unsigned, whereas the Linux definition is int. This makes writing portable code difficult, and the FLTK API uses int for add_fd() etc.. If we were going to change that I believe that we would break existing programs - at least we'd introduce compiler warnings in existing programs that use (currently appropriate) casts. Maybe, or not...

Also, there is a comment in the code your patch proposes to change:

// Keep avoiding having the socket deps at that level but mke sure it will work in both 32 & 64 bit builds

So the author (I didn't check who that was) did intentionally not include winsock2.h in FL/Fl.H, and we should probably not change this.

According to the MSDN docs cited above the only real difference to the effect of your patch (in FL/Fl.H) would be the (un-)signedness of FL_SOCKET in the /32-Bit/ Windows case. I assume that the existing code intended to keep the WIN32 case as it was before (int, as on all other platforms anyway), but added the _WIN64 case for some reason - probably compiler warnings.


Conclusion: The current API and ABI uses "int" in all cases, the proposed patch would break the ABI at least on 64-Bit Windows, hence this is not going to be applicable in FLTK 1.3.x.

I don't know if it would be wise to do such an API change in FLTK 1.4 though. I tend to not doing it (leave the API using "int") although Windows users might have to live with casts.

Comments, anybody?
 
 
#4 szukw000
08:49 Dec 28, 2015
The file 'socket.txt' contains the answers to comment #3.

One could avoid to include winsock2.h, but:

MS-WINDOWS on 32-Bit machine:

'int' is 32-Bits wide, 'unsigned int' is 32-Bits wide.

MS-WINDOWS on 64-Bit machine:

'int' is 32-Bits wide, 'unsigned int' is 32-Bits wide.

So if you cast a 64-Bits word to 32-Bits, you obviously will
loose data.

winfried
 
 
#5 AlbrechtS
16:13 Sep 01, 2016
Bumped version from 1.3 to 1.4-feature.
Changed priority from 3 (moderate) to 4 (high) for FLTK 1.4.

We know that the definition of FL_SOCKET for Windows, particularly 64-bit versions, is critical, but we can't change the API/ABI in FLTK 1.3.x.

I hope that we can find a better solution in FLTK 1.4, hence the version bump.
 
 
#6 szukw000
22:04 Sep 01, 2016
Supposed the definitions of MS-WINDOWS and of LINUX/CYGWIN do not
change:

#if defined(WIN32) && !defined(__CYGWIN__)
# if defined(_WIN64)
#  define FL_SOCKET unsigned __int64
# else
#  define FL_SOCKET unsigned int
# endif
#else
# define FL_SOCKET int
#endif

It would not be necessary to include 'winsock2.h':

-  static void add_fd(int fd, Fl_FD_Handler cb, void* = 0);
+  static void add_fd(FL_SOCKET fd, Fl_FD_Handler cb, void* = 0);

etc.

winfried
 
 
#7 ianmacarthur
14:49 Feb 13, 2018
For 1.4, while ABI is still in flux, this typedef is possibly our best choice here?  
 
#8 ianmacarthur
14:52 Feb 13, 2018
For 1.3.x - maybe we are stuck with hoping that no one uses a socket with a number that is "too large"... Which is *probably* OK on most practical implementations, AFAIK.

I'm not entirely clear why MS decided to make this 64-bit, TBH, when there are so many other places in their API's where they probably should have gone 64-bit and decided not too. It seems an odd choice here...
 
bottom left image   bottom right image

Return to Bugs & Features | Post Text | Post File ]

 
 

Comments are owned by the poster. All other content is copyright 1998-2021 by Bill Spitzak and others. This project is hosted by The FLTK Team. Please report site problems to 'erco@seriss.com'.