Why does this work? Is this a know conflict?

We obviously have no CI on this change.

someone has report this problem http://matplotlib.1069221.n5.nabble.com/Runtime-Error-on-Solaris-Error-Closing-Dupe-File-Handle-td45601.html

why this work:

on smartos:

#sed -n 512,525p /usr/include/unistd.h
#if !defined(XOPEN_OR_POSIX) || defined(__EXTENSIONS)
extern void setusershell(void);
#endif /* !defined(XOPEN_OR_POSIX)|| defined(__EXTENSIONS) /
extern unsigned sleep(unsigned);
#if !defined(XOPEN_OR_POSIX) || defined(__EXTENSIONS)
extern int stime(const time_t *);
#endif / !defined(XOPEN_OR_POSIX) || defined(__EXTENSIONS) _/
#if defined(XPG4)
/ EXTENSIONS makes the SVID Third Edition prototype in stdlib.h visible _/
extern void swab(const void *_RESTRICT_KYWD, void *RESTRICT_KYWD, ssize_t);
#endif / defined(_XPG4) */
#if !defined(__XOPEN_OR_POSIX) || defined(_XPG4_2) || defined(EXTENSIONS)
extern int symlink(const char *, const char *);
extern void sync(void);

swab was declared 'if defined(_XPG4)'. So add 'undef _XPG4' to avoid the twice declaration of swab.
Moreover, I find one commit d0d21 in 2010 which was related this issues

#git show d0d21
commit d0d2110
Author: John Hunter [email protected]
Date: Fri Jul 2 18:06:25 2010 +0000

fix swab bug so mpl will compile on solaris with cxx 6

svn path=/trunk/matplotlib/; revision=8484

diff --git a/CHANGELOG b/CHANGELOG
index c49196d..af6023d 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,3 +1,7 @@
+2010-07-02 Modified CXX/WrapPython.h to fix "swab bug" on solaris so

•       mpl can compile on Solaris with CXX6 in the trunk.  Closes
  

•       tracker bug 3022815 - JDH
  

  2010-06-30 Added autoscale convenience method and corresponding
  pyplot function for simplified control of autoscaling;
  and changed axis, set_xlim, and set_ylim so that by
  diff --git a/CXX/WrapPython.h b/CXX/WrapPython.h
  index 118a874..ef97cf1 100644
  --- a/CXX/WrapPython.h
  +++ b/CXX/WrapPython.h
  @@ -48,6 +48,9 @@
  #if defined(_XPG4)
  #undef _XPG4
  #endif
  +#if defined(_XPG3)
  +#undef _XPG3
  +#endif
  #endif

  // Python.h will redefine these and generate warning in the process

What should I do if we need setup CI env for this?
I believe many person who use matplotlib on Solaris well also met this problem

Is there any links that you could include as a comment so help guide future developers?

no link. I can update this patch with comment about that. Is it ok?

That'll have to do, I suppose. I don't think any of the regular developers have much experience developing on a sun-like system, so we don't know what the implications are of undef-ing that macro.

@WeatherGod
Ok. I find another link. I think they're same issues
https://github.com/FreeCAD/FreeCAD/blob/master/src/CXX/WrapPython.h

For CI would would need access to a solaris (like) box to run on.

(slightly better formatted version git log)

09:43 $ git show d0d2110f3fd6ba00403f949cc519760bb3ee3e2e
commit d0d2110f3fd6ba00403f949cc519760bb3ee3e2e
Author: John Hunter <[email protected]>
Date:   Fri Jul 2 18:06:25 2010 +0000

    fix swab bug so mpl will compile on solaris with cxx 6

    svn path=/trunk/matplotlib/; revision=8484

diff --git a/CHANGELOG b/CHANGELOG
index c49196d..af6023d 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,3 +1,7 @@
+2010-07-02 Modified CXX/WrapPython.h to fix "swab bug" on solaris so
+           mpl can compile on Solaris with CXX6 in the trunk.  Closes
+           tracker bug 3022815 - JDH
+
 2010-06-30 Added autoscale convenience method and corresponding
            pyplot function for simplified control of autoscaling;
            and changed axis, set_xlim, and set_ylim so that by
diff --git a/CXX/WrapPython.h b/CXX/WrapPython.h
index 118a874..ef97cf1 100644
--- a/CXX/WrapPython.h
+++ b/CXX/WrapPython.h
@@ -48,6 +48,9 @@
 #if defined(_XPG4)
 #undef _XPG4
 #endif
+#if defined(_XPG3)
+#undef _XPG3
+#endif
 #endif

 // Python.h will redefine these and generate warning in the process

attn @mdboom It looks like this is something we lost in the CXX-ectomy

@FrankYu Thanks for your diligent footwork on this!

Can you put in the XDG3 undef as well?

Given that we still have not cut 1.5.2 and we now have win+mac+linux testing in this branch I think this should be backported to 1.5.x

backported to v1.5.x via bf6e785