RE: Windows Build Broken - Feature Branch

[Conrad T. Pino]

Hi Derek,

> From: Conrad T. Pino
> 
> The unexpected is upon us.  The attached patch:
> 
>    adds "lib/nanosleep.c" to "libcvs" compile
>    adds "#include <winsock.h>" to "windows-NT/unistd.h"
> 
> which gets "nanosleep.c" to compile nicely and "libcvs" builds.
> 
> The picture turns ugly in the "libdiff" compile.  VC6 file "winnt.h"
> blows up repeatedly in a very ugly way even though it compiled nicely
> during "libcvs" build.
> 
> My current hypothesis is "libdiff" conditional compile macros are
> breaking "winnt.h" conditional compile macros since "libdiff" tends
> to include "unistd.h" late in the include chain.
> 
> I'd greatly appreciate a second opinion time permitting of course.

Attached patch and error files are provided to support the following:

I ran more experiments along the lines of copying selected lines from
"winsock.h" into "unistd.h" and this is a dead end because the "select"
function has 2 struct arguments and when they're defined in "unistd.h"
a redefinition error occurs when "winsock.h" follows in same compile.

This implies that Jim Meyering's "sys/select.h" suggestion and all its
variants are dead ends if we can't get a nested "winsock.h" include to
work.

I plan to let Jim Meyering know what we've learned so far.

These are the options as I see them now:

1. fix "libdiff" build and/or sources so nested "winsock.h" includes
work (wish I could offer more here but right now I don't have a clue),

2. ask GNULib team to reimplement "nanosleep" to replace "select" call
with a call to "usleep",

3. CVS project adds "nanosleep" implementation to "windows-NT/woe32.c"

Are you leaning in any particular direction right now?

Conrad

Patch-nanosleep-test.txt
Description: Text document

errors.txt
Description: Text document