Problem in LLVM CMake modules

Hi again,
     I'm trying to upgrade my LLVM bindings in Java from 2.9 to 3.1. To do so, I regenerated the JNI bindings from fresh LLVM 3.1 headers, and did a slight rewrite of my CMakeLists.txt file for building the C code.

     Problem is, cmake no longer finishes at all. I receive the following output, and then it just runs forever (while still responding to a CTRL-C):

eli@eli-netbook:~/Programs/jllvm/src/org/jllvm/bindings$ cmake .
-- The C compiler identification is GNU
-- The CXX compiler identification is GNU
-- Check for working C compiler: /usr/bin/gcc
-- Check for working C compiler: /usr/bin/gcc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Found Java: /usr/lib/jvm/java-6-openjdk-amd64/bin/java (found version "1.6.0.24")
-- Found JNI: /usr/lib/jvm/java-6-openjdk-amd64/jre/lib/amd64/libjawt.so

Running CMake with --debug-output yields no additional information, and running cmake with gdb attached just gives me a backtrace with no signs of abnormal operation.

I've tried the old "comment stuff out and see what causes the symptom to reappear" method of debugging, and it's definitely the llvm_map_components_to_libraries(REQ_LLVM_LIBRARIES all) line in my CMakeLists.txt. I've attached CMakeLists.txt for reference.

Could anyone perhaps tell me why CMake can't finish?

Cheers,
--Eli Gottlieb

CMakeLists.txt (1.62 KB)

Eli Gottlieb <eligottlieb@gmail.com> writes:

    I'm trying to upgrade my LLVM bindings in Java from 2.9 to 3.1. To
do so, I regenerated the JNI bindings from fresh LLVM 3.1 headers, and
did a slight rewrite of my CMakeLists.txt file for building the C
code.

    Problem is, cmake no longer finishes at all. I receive the
following output, and then it just runs forever (while still
responding to a CTRL-C):

Yep, llvm_map_components_to_libraries gets confused by the existence of
both gtest and gtest_main and enters an infinite loop. A workaround is
to not pass "all" to llvm_map_components_to_libraries but a list of
required components.

Óscar Fuentes <ofv@wanadoo.es> writes:

Yep, llvm_map_components_to_libraries gets confused by the existence of
both gtest and gtest_main and enters an infinite loop. A workaround is
to not pass "all" to llvm_map_components_to_libraries but a list of
required components.

This patch *seems* to fix the problem (cmake regexps are not thoroughly
documented):

--- a/cmake/modules/LLVM-Config.cmake
+++ b/cmake/modules/LLVM-Config.cmake
@@ -160,7 +160,7 @@ function(explicit_map_components_to_libraries out_libs)
     list(REVERSE expanded_components)
     list(APPEND processed ${lib})
     # Find the maximum index that doesn't have to be re-processed:
- while(NOT "${expanded_components}" MATCHES "^${processed}.*" )
+ while(NOT "${expanded_components}" MATCHES "^${processed}(;)|(.*)")
       list(REMOVE_AT processed -1)
     endwhile()
     list(LENGTH processed cursor)

However, using "all" as the component to be linked is still broken, as
it passes to the linker -lLTO_static and -lprofile_rt-static, which are
non-existant libraries on a static build.

On the case of LTO_static the fix probably is

diff --git a/tools/lto/CMakeLists.txt b/tools/lto/CMakeLists.txt
index 9112976..60fc902 100644
--- a/tools/lto/CMakeLists.txt
+++ b/tools/lto/CMakeLists.txt
@@ -23,5 +23,4 @@ endif()

if( NOT BUILD_SHARED_LIBS )
   add_llvm_library(${LTO_STATIC_TARGET_NAME} ${SOURCES})
- set_property(TARGET ${LTO_STATIC_TARGET_NAME} PROPERTY OUTPUT_NAME "LTO")
endif()

In the case of profile_rt-static, the fix probably is:

diff --git a/runtime/libprofile/CMakeLists.txt b/runtime/libprofile/CMakeLists.txt
index 414ad00..2794c4d 100644
--- a/runtime/libprofile/CMakeLists.txt
+++ b/runtime/libprofile/CMakeLists.txt
@@ -9,9 +9,6 @@ set(SOURCES
   )

add_llvm_library( profile_rt-static ${SOURCES} )
-set_target_properties( profile_rt-static
- PROPERTIES
- OUTPUT_NAME "profile_rt" )

add_llvm_loadable_module( profile_rt-shared ${SOURCES} )
set_target_properties( profile_rt-shared

And do we really want gtest, gtest_main and maybe other internal
libraries to be listed when the user requires "all" LLVM components?

Óscar Fuentes <ofv@wanadoo.es> writes:

This patch *seems* to fix the problem (cmake regexps are not thoroughly
documented):

--- a/cmake/modules/LLVM-Config.cmake
+++ b/cmake/modules/LLVM-Config.cmake
@@ -160,7 +160,7 @@ function(explicit_map_components_to_libraries out_libs)
     list(REVERSE expanded_components)
     list(APPEND processed ${lib})
     # Find the maximum index that doesn't have to be re-processed:
- while(NOT "${expanded_components}" MATCHES "^${processed}.*" )
+ while(NOT "${expanded_components}" MATCHES "^${processed}(;)|(.*)")
       list(REMOVE_AT processed -1)
     endwhile()
     list(LENGTH processed cursor)

The patch above doesn't make any sense. This looks more correct:

--- a/cmake/modules/LLVM-Config.cmake
+++ b/cmake/modules/LLVM-Config.cmake
@@ -160,7 +160,7 @@ function(explicit_map_components_to_libraries out_libs)
     list(REVERSE expanded_components)
     list(APPEND processed ${lib})
     # Find the maximum index that doesn't have to be re-processed:
- while(NOT "${expanded_components}" MATCHES "^${processed}.*" )
+ while(NOT "${expanded_components}" MATCHES "^${processed};|$")
       list(REMOVE_AT processed -1)
     endwhile()
     list(LENGTH processed cursor)