llvm dev - May 2014 - [LLVMdev] [PATCH] Use GCC_INSTALL_PREFIX for rpath if set.

This way, an LLVM compiled after setting GCC_INSTALL_PREFIX will work correctly
if you don't want to install LLVM into the GCC_INSTALL_PREFIX.
---
 CMakeLists.txt | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 9ec3e33..c85a028 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -450,7 +450,12 @@ if (APPLE)
   set(CMAKE_INSTALL_RPATH "@executable_path/../lib")
 else(UNIX)
   if(NOT DEFINED CMAKE_INSTALL_RPATH)
-    set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
+    if(DEFINED GCC_INSTALL_PREFIX)
+      set(CMAKE_INSTALL_RPATH "${GCC_INSTALL_PREFIX}/lib")
+    else()
+      set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
+    endif()
+
     if (${CMAKE_SYSTEM_NAME} MATCHES FreeBSD)
       set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}
-Wl,-z,origin")
     endif()
-- 
1.9.1.423.g4596e3a

This looks like it would remove the origin/../lib RPATH which would break
finding libraries related to Clang such as libc++?

It makes sense to add the GCC install prefix to the RPATH though. I wonder,
should we try to do something about the lib vs. lib{32,64} business? I
don't know that either of these does the right thing there.


On Thu, May 29, 2014 at 3:37 PM, Russell Harmon <eatnumber1 at google.com>
wrote:
> This way, an LLVM compiled after setting GCC_INSTALL_PREFIX will work
> correctly
> if you don't want to install LLVM into the GCC_INSTALL_PREFIX.
> ---
>  CMakeLists.txt | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/CMakeLists.txt b/CMakeLists.txt
> index 9ec3e33..c85a028 100644
> --- a/CMakeLists.txt
> +++ b/CMakeLists.txt
> @@ -450,7 +450,12 @@ if (APPLE)
>    set(CMAKE_INSTALL_RPATH "@executable_path/../lib")
>  else(UNIX)
>    if(NOT DEFINED CMAKE_INSTALL_RPATH)
> -    set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
> +    if(DEFINED GCC_INSTALL_PREFIX)
> +      set(CMAKE_INSTALL_RPATH "${GCC_INSTALL_PREFIX}/lib")
> +    else()
> +      set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
> +    endif()
> +
>      if (${CMAKE_SYSTEM_NAME} MATCHES FreeBSD)
>        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}
> -Wl,-z,origin")
>      endif()
> --
> 1.9.1.423.g4596e3a
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140529/2dcc1f44/attachment.html>

The behavior of automatically detecting libraries installed in the same
prefix feels like magic to me anyway. Perhaps it would be better to have
project-level variables for specifying the path to them individually? The
default value for them could remain as origin/../lib which would keep from
breaking things.

On Thu May 29 2014 at 7:06:17 PM, Chandler Carruth <chandlerc at
google.com>
wrote:
> This looks like it would remove the origin/../lib RPATH which would break
> finding libraries related to Clang such as libc++?
>
> It makes sense to add the GCC install prefix to the RPATH though. I
> wonder, should we try to do something about the lib vs. lib{32,64}
> business? I don't know that either of these does the right thing there.
>
>
> On Thu, May 29, 2014 at 3:37 PM, Russell Harmon <eatnumber1 at
google.com>
> wrote:
>
>> This way, an LLVM compiled after setting GCC_INSTALL_PREFIX will work
>> correctly
>> if you don't want to install LLVM into the GCC_INSTALL_PREFIX.
>> ---
>>  CMakeLists.txt | 7 ++++++-
>>  1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/CMakeLists.txt b/CMakeLists.txt
>> index 9ec3e33..c85a028 100644
>> --- a/CMakeLists.txt
>> +++ b/CMakeLists.txt
>> @@ -450,7 +450,12 @@ if (APPLE)
>>    set(CMAKE_INSTALL_RPATH "@executable_path/../lib")
>>  else(UNIX)
>>    if(NOT DEFINED CMAKE_INSTALL_RPATH)
>> -    set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
>> +    if(DEFINED GCC_INSTALL_PREFIX)
>> +      set(CMAKE_INSTALL_RPATH "${GCC_INSTALL_PREFIX}/lib")
>> +    else()
>> +      set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
>> +    endif()
>> +
>>      if (${CMAKE_SYSTEM_NAME} MATCHES FreeBSD)
>>        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}
>> -Wl,-z,origin")
>>      endif()
>>
> --
>> 1.9.1.423.g4596e3a
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140530/e13ecd9a/attachment.html>

On 29 May 2014 18:37, Russell Harmon <eatnumber1 at google.com>
wrote:
> This way, an LLVM compiled after setting GCC_INSTALL_PREFIX will work
correctly
> if you don't want to install LLVM into the GCC_INSTALL_PREFIX.
> ---
>  CMakeLists.txt | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/CMakeLists.txt b/CMakeLists.txt
> index 9ec3e33..c85a028 100644
> --- a/CMakeLists.txt
> +++ b/CMakeLists.txt
> @@ -450,7 +450,12 @@ if (APPLE)
>    set(CMAKE_INSTALL_RPATH "@executable_path/../lib")
>  else(UNIX)
>    if(NOT DEFINED CMAKE_INSTALL_RPATH)
> -    set(CMAKE_INSTALL_RPATH "\$ORIGIN/../lib")
I do depend on $ORIGIN/../lib.

This makes sure that a shared build of llvm find the correct libraries
both in the build and the install directory. It also means that the
install directory is relocatable.

Cheers,
Rafael