When building Python on Solaris, I don't get the os.mknod function. This
seems to be a combination of two errors:
1. The definition of posix_mknod() in posixmodule.c is surrounded by:
#if defined(HAVE_MKNOD) && defined(HAVE_MAKEDEV)
It works fine if I remove the HAVE_MAKEDEV define.

2. The reason why HAVE_MAKEDEV doesn't work, is that the Python
configure script only looks for makedev in <sys/types.h>, while on
Solaris you need to include <sys/mkdev.h> as well.

cc -V gives:
cc: Sun C 5.9 SunOS_sparc Patch 124867-01 2007/07/12
uname -a gives:
SunOS zelda 5.9 Generic_117171-07 sun4us sparc FJSV,GPUZC-M

It would be best if you could contribute a patch to fix this. The source
of configure is configure.in; you need autoconf to generate configure
from it.

The patch seems to be trivial. I need review, specially compiling under non-Solaris OS.

I plan to commit this patch to 2.6, 2.7, 3.1 and 3.2.

Must I update "NEWS"?

Compiles fine under Debian stable.
You shouldn't commit to 2.6 until 2.6.5 is released, though. If you want to do so, contact Barry first.

You do need a NEWS entry.

Please, do the final review, ready for commiting.

I have asked about committing to 2.6 in the python-committers mailing list.

This will not block 2.6.5 but is a candidate for 2.6.6.

Some questions: 
- why patch don't update  posixmodule.c and remove  defined(HAVE_MAKEDEV) ?
- how is defined HAVE_DEVICE_MACROS in pyconfig.h

About changes in configure.in - I'm not sure that they are correct (more later).

I just did the minimal change. I don't know if removing "defined(HAVE_MAKEDEV)" is safe.

"HAVE_DEVICE_MACROS" is defined if "configure" finds "makedev()" macro.

Clarify your comment about configure.in changes being wrong.

Jesús Cea Avión wrote:
>
> Jesús Cea Avión<jcea@jcea.es>  added the comment:
>
> I just did the minimal change. I don't know if removing "defined(HAVE_MAKEDEV)" is safe.

The python build system is full with minimal changes and result is a big 
mess. Did you found which revision add ".. defined(HAVE_MAKEDEV)" in 
posix*.c ?

> "HAVE_DEVICE_MACROS" is defined if "configure" finds "makedev()" macro.

And what is result on you platform ?

> Clarify your comment about configure.in changes being wrong.

Usually just adding #include <xxxx> will break test on platforms where 
xxxx is missing.

The correct implementation will depend from above.

Cases:
1) platform is no longer supported - it is save to remove test case from 
configure and defined(HAVE_MAKEDEV) from posixmodule.c

2) HAVE_DEVICE_MACROS is defined for you . Then check how is written 
test for HAVE_DEVICE_MACROS and make test for makedev similar.

Roumen

jcea, why did  you make this a release blocker for 2.6.5?

Sorry, Barry. Don't know what happened. Probably write a followup without reloading the page and "undid" previous changes.

I keep my issues permanently open in firefox tabs. I will try to be more careful.

Roumen Petrov (rpetrov) wrote:
>>
>> Jesús Cea Avión<jcea@jcea.es>  added the comment:
>>
>> I just did the minimal change. I don't know if removing "defined(HAVE_MAKEDEV)" is safe.

> The python build system is full with minimal changes and result is a
> big mess. Did you found which revision add ".. defined(HAVE_MAKEDEV)"
> in posix*.c ?

The code was added in 2002-07-30 by nnorwitz, to support OSF1 (Dec Unix). Is that platform still supported?. I don't see any related  buildbot.

>> "HAVE_DEVICE_MACROS" is defined if "configure" finds "makedev()" macro.

> And what is result on you platform ?

It is defined if I add the new "include".


> Cases:
> 1) platform is no longer supported - it is save to remove test case from 
> configure and defined(HAVE_MAKEDEV) from posixmodule.c

Do we have a list of officially supported?.

> 2) HAVE_DEVICE_MACROS is defined for you . Then check how is written 
> test for HAVE_DEVICE_MACROS and make test for makedev similar.

Investigating the issue, I could say that HAVE_MAKEDEV should be deleted, but I don't have access to a OSF1 machine for trying.

How should I proceed?

My browser keeps deleting nosy.  Sorry.

Roumen, please read.

>>> "HAVE_DEVICE_MACROS" is defined if "configure" finds "makedev()" macro.
>> And what is result on you platform ?
>It is defined if I add the new "include".

a) Why you touch test case for "AC_MSG_CHECKING(for major, minor, and makedev)
AC_TRY_LINK([
#if defined(MAJOR_IN_MKDEV)
#include <sys/mkdev.h>
#elif defined(MAJOR_IN_SYSMACROS)
#include <sys/sysmacros.h>
#else
#include <sys/types.h>
#endif
],[
  makedev(major(0),minor(0));
],[
  AC_DEFINE(HAVE_DEVICE_MACROS, 1,
	    [Define to 1 if you have the device macros.])
  AC_MSG_RESULT(yes)
],[
  AC_MSG_RESULT(no)
])"
?
Sorry I did't understand why this test case fail for you.


b) PEP 11 list unsupported platforms and OSF1 is not listed.

c) So code if from revision 27820 that fix "SF patch #584245, get python to link on OSF1 (Dec Unix)" . The current patch as is will break OSF1 and other OS-es (as example linux don't define this header). Just adding new include will fail test case and will left as undefined HAVE_MAKEDEV => mknod won't be added for those platforms.

I think this small change can slip in after beta.

Please, Review. Roumen?.

I plan to commit this to 2.7, 2.6, 3.1 and 3.2, if you agree.

Jesus, this looks ok, have you checked the changes work fine under e.g. Linux?
I don't think DEC Unix is supported anymore, actually I'm not sure anyone still uses it.

Pitrou, the code configures correctly in Linux.

Reading PEP11, I don't see the procedure to propose a platform for deprecation. Just mailing python-dev?

I will delay the patch committing a couple of days, just in case somebody else want to comment.

Thanks for the review, Pitrou.

> Reading PEP11, I don't see the procedure to propose a platform for
> deprecation. Just mailing python-dev?

Yes.

> Thanks for the review, Pitrou.

You can call me Antoine.

Informal europeans! :-).

Hi, Antoine, I am Jesús (sunny Spain!).

Deprecation request mailed to python-dev.

Sorry Jesús, right now I cannot test new patch it but It looks goodand I 
expect first test to succeed so that second won't be tested as result 
OCF_XXX to be defined. If there is no spelling error it is fine.

Roumen

Patch committed.

trunk (2.7): r80574
2.6: r80575
py3k (3.2): r80576
3.1: r80577