Re: [Python-Dev] Addition of "pyprocessing" module to standard lib.
On Mon, May 19, 2008 at 5:15 PM, Bill Janssen <janssen@parc.com> wrote:
If you can run a pure Python module that does not depend on any C extension, then that platform has the support needed to run Python.
This is certainly a point of view. One that many end-users wouldn't understand :-).
Perhaps, but it's clear-cut. Is OS X not properly supported because it can't run the _winreg module? I just don't like labeling a platform as unsupported just because ctypes doesn't compile on it. -Brett
On Mon, May 19, 2008 at 5:15 PM, Bill Janssen <janssen@parc.com> wrote:
If you can run a pure Python module that does not depend on any C extension, then that platform has the support needed to run Python.
This is certainly a point of view. One that many end-users wouldn't understand :-).
Perhaps, but it's clear-cut. Is OS X not properly supported because it can't run the _winreg module? I just don't like labeling a platform as unsupported just because ctypes doesn't compile on it.
I assume _winreg runs everywhere Python is said to run, and there's a Windows registry to examine, so I think it's a different kettle of fish. ctypes doesn't run everywhere Python is said to run, and there are dynamic libraries to call into. I think it would be great if we could get some AIX, HP-UX, and Solaris boxes for Thomas to work on. What would motivate IBM, H-P, and Sun to donate such gear? Perhaps each of the companies have an office somewhere that could help with this resource allocation? For instance, talking to the "AIX Collaboration Center" of IBM (aixcc@us.ibm.com)? And these are all SYSV derivatives, aren't they? So perhaps it's some common fix for all three? Bill
Bill Janssen wrote:
On Mon, May 19, 2008 at 5:15 PM, Bill Janssen <janssen@parc.com> wrote:
If you can run a pure Python module that does not depend on any C extension, then that platform has the support needed to run Python. This is certainly a point of view. One that many end-users wouldn't understand :-). Perhaps, but it's clear-cut. Is OS X not properly supported because it can't run the _winreg module? I just don't like labeling a platform as unsupported just because ctypes doesn't compile on it.
I assume _winreg runs everywhere Python is said to run, and there's a Windows registry to examine, so I think it's a different kettle of fish. ctypes doesn't run everywhere Python is said to run, and there are dynamic libraries to call into.
I think it would be great if we could get some AIX, HP-UX, and Solaris boxes for Thomas to work on. What would motivate IBM, H-P, and Sun to donate such gear? Perhaps each of the companies have an office somewhere that could help with this resource allocation? For instance, talking to the "AIX Collaboration Center" of IBM (aixcc@us.ibm.com)?
And these are all SYSV derivatives, aren't they? So perhaps it's some common fix for all three?
OK, I know people in Sun and possibly H-P I could ask, and I may have an arm or two to twist to get to IBM. But worst-case, what budget would the infrastructure committee need for the hardware and (more importantly) if the hardware materialized, would there be manpower to maintain the platforms as "Python supported"? The more libraries that use ctypes to call into native functionality, the more important it becomes to have ctypes run, even if only to implement platform-specific functionality on the given platforms. I would like "being able to call a wide range of native libraries" to be a Python claim on all platforms, even when the native libraries are platform-proprietary. regards Steve -- Steve Holden +1 571 484 6266 +1 800 494 3119 Holden Web LLC http://www.holdenweb.com/
On May 19, 2008, at 7:47 PM, Steve Holden wrote:
OK, I know people in Sun and possibly H-P I could ask, and I may have an arm or two to twist to get to IBM. But worst-case, what budget would the infrastructure committee need for the hardware and (more importantly) if the hardware materialized, would there be manpower to maintain the platforms as "Python supported"?
You can ask me as far as Sun goes. There are probably a variety of options: 1) Get the open source version of VirtualBox and use it to run a virtualized OpenSolaris 2) I can see if we can get folks access to a box running Solaris 3) I can ask and see if there are people at Sun who would be willing to jump in and help Ted
Ted Leung wrote:
On May 19, 2008, at 7:47 PM, Steve Holden wrote:
OK, I know people in Sun and possibly H-P I could ask, and I may have an arm or two to twist to get to IBM. But worst-case, what budget would the infrastructure committee need for the hardware and (more importantly) if the hardware materialized, would there be manpower to maintain the platforms as "Python supported"?
You can ask me as far as Sun goes. There are probably a variety of options:
1) Get the open source version of VirtualBox and use it to run a virtualized OpenSolaris 2) I can see if we can get folks access to a box running Solaris 3) I can ask and see if there are people at Sun who would be willing to jump in and help
We actually have a couple of Solaris buildbots already - as I understand it, the issue there isn't Solaris as such, it's being able to use the Sun compiler instead of GCC to compile ctypes/libffi. Cheers, Nick. -- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia --------------------------------------------------------------- http://www.boredomandlaziness.org
On May 20, 2008, at 2:03 PM, Nick Coghlan wrote:
We actually have a couple of Solaris buildbots already - as I understand it, the issue there isn't Solaris as such, it's being able to use the Sun compiler instead of GCC to compile ctypes/libffi.
Does that mean that we need access to the Sun compiler or that the Sun compiler has bugs which prevent ctypes from compiling? Ted
Ted Leung wrote:
On May 20, 2008, at 2:03 PM, Nick Coghlan wrote:
We actually have a couple of Solaris buildbots already - as I understand it, the issue there isn't Solaris as such, it's being able to use the Sun compiler instead of GCC to compile ctypes/libffi.
Does that mean that we need access to the Sun compiler or that the Sun compiler has bugs which prevent ctypes from compiling?
I don't think anyone's mentioned Solaris in this context, but there are claims that libffi "doesn't build" for AIX and HP-UX. regards Steve -- Steve Holden +1 571 484 6266 +1 800 494 3119 Holden Web LLC http://www.holdenweb.com/
Does that mean that we need access to the Sun compiler or that the Sun compiler has bugs which prevent ctypes from compiling?
I don't think anyone's mentioned Solaris in this context, but there are claims that libffi "doesn't build" for AIX and HP-UX.
I think there was also an issue with using the Sun compiler to build ctypes and/or libffi. Bill
We actually have a couple of Solaris buildbots already - as I understand it, the issue there isn't Solaris as such, it's being able to use the Sun compiler instead of GCC to compile ctypes/libffi.
Does that mean that we need access to the Sun compiler or that the Sun compiler has bugs which prevent ctypes from compiling?
Neither, nor. ctypes (or, rather, libffi) has code specific to gcc (or, rather, the GNU assembler) that makes the Sun compiler reject it. IIRC, it chokes on the attempt to compile assembler code that has C preprocessor macros in it (can't test it right now). Regards, Martin
Martin v. Löwis schrieb:
We actually have a couple of Solaris buildbots already - as I understand it, the issue there isn't Solaris as such, it's being able to use the Sun compiler instead of GCC to compile ctypes/libffi.
Does that mean that we need access to the Sun compiler or that the Sun compiler has bugs which prevent ctypes from compiling?
Neither, nor. ctypes (or, rather, libffi) has code specific to gcc (or, rather, the GNU assembler) that makes the Sun compiler reject it. IIRC, it chokes on the attempt to compile assembler code that has C preprocessor macros in it (can't test it right now).
Could it be a solution to build libffi with gcc, then configure Python with '--with-system--ffi' and compile with the sun compiler, or would this introduce the dependencies on libgcc or whatever again that Ulrich wanted to avoid? Thomas
Could it be a solution to build libffi with gcc, then configure Python with '--with-system--ffi' and compile with the sun compiler, or would this introduce the dependencies on libgcc or whatever again that Ulrich wanted to avoid?
It depends on the processor and the code, but chances are high that this would indeed create a dependency on libgcc. Regards, Martin
Martin v. Löwis wrote:
IIRC, it chokes on the attempt to compile assembler code that has C preprocessor macros in it (can't test it right now).
Could the build process be modified to run the C preprocessor over the assembly language first? -- Greg
Steve Holden wrote:
The more libraries that use ctypes to call into native functionality, the more important it becomes to have ctypes run, even if only to implement platform-specific functionality on the given platforms. I would like "being able to call a wide range of native libraries" to be a Python claim on all platforms, even when the native libraries are platform-proprietary.
Yes, I'd like to see this happen, too. Bill
On Mon, May 19, 2008 at 06:13:11PM -0700, Bill Janssen wrote:
And these are all SYSV derivatives, aren't they? So perhaps it's some common fix for all three?
This reminds of a Tim-ism: ==================
Just for the record, on AIX, the following C program:
Oh no you don't! I followed AIX threads for the first year it came out, but eventually decided there was no future in investing time in baffling discussions that usually ended with "oh, never mind -- turns out it's a bug" <0.9 wink>. Vladimir Marangozov and Tim Peters, 23 Jun 1998 ================== 10 years later, things don't seem to be much different. For ctypes it looks like libffi often fails to compile when not using GCC; http://bugs.python.org/issue1637120 is an AIX bug report. For the curses module, I've noticed that HP-UX seems to require including additional header files or defining _XOPEN_SOURCE to make curses.h work. So I don't think there's a common problem with all these minority platforms, and we really would need developers on all of them. --amk
Bill Janssen wrote:
And these are all SYSV derivatives, aren't they? So perhaps it's some common fix for all three?
I've never personally had the pleasure(?!) of having to get stuff working on all of AIX/HP-UX/Tru64/Solaris, but what little dealings I've had with people who have suggests that the differences between these platforms make the discrepancies between different Linux distributions look utterly trivial by comparison. Cheers, Nick. -- Nick Coghlan | ncoghlan@gmail.com | Brisbane, Australia --------------------------------------------------------------- http://www.boredomandlaziness.org
participants (10)
-
"Martin v. Löwis" -
A.M. Kuchling -
Bill Janssen -
Brett Cannon -
Greg Ewing -
Nick Coghlan -
Steve Holden -
Ted Leung -
Ted Leung
-
Thomas Heller