| msg89118 - (view) |
Author: Ronald Oussoren (ronaldoussoren) *  |
Date: 2009-06-08 21:35 |
| Apple just announced that MacOSX 10.6 ("Snow Leopard") will be released in September, and will only support intel systems. This means that the --with-universal-archs option is not 100% usable if you want to build a version of Python that specifically targets this release of the OS. The attached patch adds the option '--with-universal-archs=intel', which will build a 32-bit/64-bit framework for intel systemsn (that is i386 and x86_64). Note: the patch is for the trunk, I'll port it to 3.x after review. |
|
|
| msg89119 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2009-06-08 21:51 |
| What do you mean by "not 100% usable"? I would hope that a four-way binary still works just fine, no? So what are the restrictions? |
|
|
| msg89129 - (view) |
Author: Ronald Oussoren (ronaldoussoren) * |
Date: 2009-06-09 00:17 |
| The details of snow leopard are under NDA, but based on public information : * Snow Leopard will only support Intel-based systems * Apple's "rosetta" dynamaic translation subsystem for running PPC code on an intel system only supports 32-bit PPC code. There is therefore no reason to include 64-bit PPC code in builds for Snow Leopard, and I wouldn't be surprised if SL would ship without 64- bit PPC code in system frameworks. |
|
|
| msg89132 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2009-06-09 04:32 |
| > The details of snow leopard are under NDA, but based on public > information : This is all fine, and really not surprising. My question is: Why is that causing limited usability? I would expect that Snow Leopard just ignores the PPC bits in the universal binaries, and just accesses the x86 bits. The binaries might be larger than necessary; people bothered by that could easily strip them using ditto (IIUC). I'm fine with an option of not building the PPC bits - I'm just puzzled by the claim that, without the patch, it will not be "100% usable". |
|
|
| msg89141 - (view) |
Author: John Szakmeister (jszakmeister) * |
Date: 2009-06-09 09:32 |
| I think Ronald is trying to say that a 10.6 SDK is likely not to support building 64-bit binaries at all for the PPC (since there won't be 64-bit versions of the supporting libraries). So you need this patch, if you're going to build against it. I don't think he's trying to claim that fat binaries won't run if there is 64-bit PPC code in them. FWIW, it's the --with-universal-archs option that he said wasn't 100% usable, not the binaries. |
|
|
| msg89164 - (view) |
Author: Martin v. Löwis (loewis) * |
Date: 2009-06-09 20:36 |
| > FWIW, > it's the --with-universal-archs option that he said wasn't 100% usable, > not the binaries. Ah, ok. That clarifies it. |
|
|
| msg92066 - (view) |
Author: Mattias Stahre (plux) |
Date: 2009-08-29 20:20 |
| Is there any progress on this? |
|
|
| msg92302 - (view) |
Author: Ronald Oussoren (ronaldoussoren) * |
Date: 2009-09-06 12:05 |
| I've attached a new patch "arch-intel-v2.patch" that relects my current thinking about this patch. This adds two new options to the "--with-univeral-archs" option for configure: "intel" and "3-way". The former builds a universal binary with the i386 and x86_64 architectures, the latter builds one with the i386, x86_64 and ppc architectures. The "intel" build is most useful when your code only needs to run on modernish systems, the latter allows you to run scripts on PPC systems as well. Both can be build on Snow Leopard systems. The current options "all" and "64-bit" cannot be used when you perform your build and target the default SDK because the system doesn't ship with PPC64 suppport anymore. I intend to apply this patch for 2.7 and 3.2 unless there is serious opposition against this. The patch will not be backported to earlier versions. |
|
|
| msg92649 - (view) |
Author: Ronald Oussoren (ronaldoussoren) * |
Date: 2009-09-15 19:17 |
| Committed in 74806 (trunk), 74807 (2.6), 74808 (3.x), 74809 (3.1) |
|
|