(0018468)
Bill Hoffman (manager)
2009-11-17 19:59

I don't see where these sizes are actually used in the CMake c++ but maybe I am missing something. I guess the header should be updated to include the ifdef __APPLE stuff to get the size of stuff done.

(0018475)
Brad King (manager)
2009-11-18 14:08

KWSys needs to be taught to get the size of long and some other types without a try-compile in order to support universal binaries.

For now, try adding "-DCMAKE_TRY_COMPILE_OSX_ARCHITECTURES=i386" to work around the problem.

(0018477)
Sean McBride (reporter)
2009-11-18 15:26

Could KWSys be changed to use fixed sized types like int32_t and int64_t (when it needs something of a fixed size)? Why does it care how big 'long' is?

(0018479)
Brad King (manager)
2009-11-18 15:33

It's a component of KWSys called "FundamentalType" that provides type information about the native types, plus its own equivalent of int32_t and int64_t that does not depend on system headers.

CMake doesn't actually use the component (yet), but since CMake builds on more platforms than any other project that uses KWSys, we enabled all of KWSys for testing purposes. Perhaps this should only be done for dashboard builds, but then the dashboard builds would not be representative of user builds.

(0018482)
Sean McBride (reporter)
2009-11-18 16:41

I looked at FundamentalType.h.in, but only partly understand it....

For sure CMake needs to support a wide variety of systems/compilers/etc. and therefore will need to test the size of native C types to create its own KWint32 (or whatever). But could it not do so only if the standard C99/C++0x types are not present?

(0018492)
Brad King (manager)
2009-11-20 08:39

The following commits fix KWSys FundamentalType. More changes are needed to fix the same problem in Utilities/cmlibarchive.

Enable loose loop constructs in KWSys
/cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt
new revision: 1.151; previous revision: 1.150

Check for 'long long' without computing size
/cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt
new revision: 1.152; previous revision: 1.151
/cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTestsCXX.cxx,v <-- Source/kwsys/kwsysPlatformTestsCXX.cxx
new revision: 1.7; previous revision: 1.6

Create KWSYS_PLATFORM_INFO_TEST macro
/cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTests.cmake,v <-- Source/kwsys/kwsysPlatformTests.cmake
new revision: 1.4; previous revision: 1.3

Fix KWSys FundamentalType for Universal Binaries
/cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt
new revision: 1.153; previous revision: 1.152
/cvsroot/CMake/CMake/Source/kwsys/FundamentalType.h.in,v <-- Source/kwsys/FundamentalType.h.in
new revision: 1.4; previous revision: 1.3
/cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTestsC.c,v <-- Source/kwsys/kwsysPlatformTestsC.c
new revision: 1.4; previous revision: 1.3

(0018547)
Brad King (manager)
2009-11-23 11:36

The following commit addresses the problem in Utilities/cmlibarchive:

libarchive: Use one architecture for try-compiles
/cvsroot/CMake/CMake/Utilities/cmlibarchive/CMakeLists.txt,v <-- Utilities/cmlibarchive/CMakeLists.txt
new revision: 1.14; previous revision: 1.13

(0018548)
Brad King (manager)
2009-11-23 11:47

I've scheduled these fixes for 2.8.1.

I also converted one of our universal binary nightly dashboard builds to use "x86_64;i386" as its list of architectures. This should ensure that the combination continues to work.

Issue History
Date Modified	Username	Field	Change
2009-11-17 18:39	Sean McBride	New Issue
2009-11-17 19:57	Bill Hoffman	Status	new => assigned
2009-11-17 19:57	Bill Hoffman	Assigned To	=> Brad King
2009-11-17 19:59	Bill Hoffman	Note Added: 0018468
2009-11-18 14:08	Brad King	Note Added: 0018475
2009-11-18 14:33	Brad King	Summary	CMake 2.8 cannot build itself as Universal Binary because "CHECK_TYPE_SIZE found different results" => CMake 2.8 cannot build itself as mixed-size Universal Binary because "CHECK_TYPE_SIZE found different results"
2009-11-18 15:26	Sean McBride	Note Added: 0018477
2009-11-18 15:33	Brad King	Note Added: 0018479
2009-11-18 16:41	Sean McBride	Note Added: 0018482
2009-11-20 08:39	Brad King	Note Added: 0018492
2009-11-23 11:36	Brad King	Note Added: 0018547
2009-11-23 11:47	Brad King	Note Added: 0018548
2009-11-23 11:47	Brad King	Status	assigned => closed
2009-11-23 11:47	Brad King	Resolution	open => fixed

Notes
(0018468) Bill Hoffman (manager) 2009-11-17 19:59	I don't see where these sizes are actually used in the CMake c++ but maybe I am missing something. I guess the header should be updated to include the ifdef __APPLE stuff to get the size of stuff done.

(0018475) Brad King (manager) 2009-11-18 14:08	KWSys needs to be taught to get the size of long and some other types without a try-compile in order to support universal binaries. For now, try adding "-DCMAKE_TRY_COMPILE_OSX_ARCHITECTURES=i386" to work around the problem.

(0018477) Sean McBride (reporter) 2009-11-18 15:26	Could KWSys be changed to use fixed sized types like int32_t and int64_t (when it needs something of a fixed size)? Why does it care how big 'long' is?

(0018479) Brad King (manager) 2009-11-18 15:33	It's a component of KWSys called "FundamentalType" that provides type information about the native types, plus its own equivalent of int32_t and int64_t that does not depend on system headers. CMake doesn't actually use the component (yet), but since CMake builds on more platforms than any other project that uses KWSys, we enabled all of KWSys for testing purposes. Perhaps this should only be done for dashboard builds, but then the dashboard builds would not be representative of user builds.

(0018482) Sean McBride (reporter) 2009-11-18 16:41	I looked at FundamentalType.h.in, but only partly understand it.... For sure CMake needs to support a wide variety of systems/compilers/etc. and therefore will need to test the size of native C types to create its own KWint32 (or whatever). But could it not do so only if the standard C99/C++0x types are not present?

(0018492) Brad King (manager) 2009-11-20 08:39	The following commits fix KWSys FundamentalType. More changes are needed to fix the same problem in Utilities/cmlibarchive. Enable loose loop constructs in KWSys /cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt new revision: 1.151; previous revision: 1.150 Check for 'long long' without computing size /cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt new revision: 1.152; previous revision: 1.151 /cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTestsCXX.cxx,v <-- Source/kwsys/kwsysPlatformTestsCXX.cxx new revision: 1.7; previous revision: 1.6 Create KWSYS_PLATFORM_INFO_TEST macro /cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTests.cmake,v <-- Source/kwsys/kwsysPlatformTests.cmake new revision: 1.4; previous revision: 1.3 Fix KWSys FundamentalType for Universal Binaries /cvsroot/CMake/CMake/Source/kwsys/CMakeLists.txt,v <-- Source/kwsys/CMakeLists.txt new revision: 1.153; previous revision: 1.152 /cvsroot/CMake/CMake/Source/kwsys/FundamentalType.h.in,v <-- Source/kwsys/FundamentalType.h.in new revision: 1.4; previous revision: 1.3 /cvsroot/CMake/CMake/Source/kwsys/kwsysPlatformTestsC.c,v <-- Source/kwsys/kwsysPlatformTestsC.c new revision: 1.4; previous revision: 1.3

(0018547) Brad King (manager) 2009-11-23 11:36	The following commit addresses the problem in Utilities/cmlibarchive: libarchive: Use one architecture for try-compiles /cvsroot/CMake/CMake/Utilities/cmlibarchive/CMakeLists.txt,v <-- Utilities/cmlibarchive/CMakeLists.txt new revision: 1.14; previous revision: 1.13

(0018548) Brad King (manager) 2009-11-23 11:47	I've scheduled these fixes for 2.8.1. I also converted one of our universal binary nightly dashboard builds to use "x86_64;i386" as its list of architectures. This should ensure that the combination continues to work.

Relationships