Skip to contentSkip to conversation

dependencies/cmake: blacklist cmake 3.7

PATCH New Yann E. MORIN 5 participants 21 messages Feb 26, 2017 – Feb 28, 2017

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html
---
 support/dependencies/check-host-cmake.sh | 8 ++++++++
 1 file changed, 8 insertions(+)
support/dependencies/check-host-cmake.sh +8 -0
index 9b63b06..84c26c2 100755
@@ -6,6 +6,9 @@ version_min="${2}"
 major_min="${version_min%.*}"
 minor_min="${version_min#*.}"
 
+# cmake-3.7 incorrectly handles rpath, linking to host libraries
+blacklist_version="3.7"
+
 cmake=`which ${candidate}`
 if [ ! -x "${cmake}" ]; then
     # echo nothing: no suitable cmake found
@@ -27,6 +30,11 @@ version="$(${cmake} --version \
 major="${version%.*}"
 minor="${version#*.}"
 
+if [ "${version}" = "${blacklist_version}" ]; then
+    # echo nothing: no suitable cmake found
+    exit 1
+fi
+
 if [ ${major} -gt ${major_min} ]; then
     echo "${cmake}"
 else
Signature · 1 line
2.7.4
#1 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Hi Yann,

Quoted 32 lines · Yann E. MORIN

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html
---
 support/dependencies/check-host-cmake.sh | 8 ++++++++
 1 file changed, 8 insertions(+)
support/dependencies/check-host-cmake.sh +3 -0
index 9b63b06..84c26c2 100755
@@ -6,6 +6,9 @@ version_min="${2}"
 major_min="${version_min%.*}"
 minor_min="${version_min#*.}"
 
+# cmake-3.7 incorrectly handles rpath, linking to host libraries
+blacklist_version="3.7"
+
 cmake=`which ${candidate}`
 if [ ! -x "${cmake}" ]; then
     # echo nothing: no suitable cmake found

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 33

Quoted 4 lines
support/dependencies/check-host-cmake.sh +1 -0
@@ -27,6 +30,11 @@ version="$(${cmake} --version \
 major="${version%.*}"
 minor="${version#*.}"
 
+if [ "${version}" = "${blacklist_version}" ]; then

Minor nit: there are no quotes around version variables in the rest of this file.

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 38

Quoted 5 lines
support/dependencies/check-host-cmake.sh +4 -0
+    # echo nothing: no suitable cmake found
+    exit 1
+fi
+
 if [ ${major} -gt ${major_min} ]; then

E.g., here.

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 40

Quoted 2 lines
support/dependencies/check-host-cmake.sh +0 -0
     echo "${cmake}"
 else

baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -
#2 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Baruch, All,

On 2017-02-27 06:43 +0200, Baruch Siach spake thusly:

Quoted 38 lines

Hi Yann,

Quoted 35 lines · Yann E. MORIN

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html
---
 support/dependencies/check-host-cmake.sh | 8 ++++++++
 1 file changed, 8 insertions(+)
support/dependencies/check-host-cmake.sh +3 -0
index 9b63b06..84c26c2 100755
@@ -6,6 +6,9 @@ version_min="${2}"
 major_min="${version_min%.*}"
 minor_min="${version_min#*.}"
 
+# cmake-3.7 incorrectly handles rpath, linking to host libraries
+blacklist_version="3.7"
+
 cmake=`which ${candidate}`
 if [ ! -x "${cmake}" ]; then
     # echo nothing: no suitable cmake found
@@ -27,6 +30,11 @@ version="$(${cmake} --version \
 major="${version%.*}"
 minor="${version#*.}"
 
Quoted 4 lines

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 33

Quoted 1 line
support/dependencies/check-host-cmake.sh +1 -0
+if [ "${version}" = "${blacklist_version}" ]; then

Minor nit: there are no quotes around version variables in the rest of this file.

That's because here we are testing strings, while...

Quoted 7 lines

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 38

Quoted 5 lines
support/dependencies/check-host-cmake.sh +4 -0
+    # echo nothing: no suitable cmake found
+    exit 1
+fi
+
 if [ ${major} -gt ${major_min} ]; then

E.g., here.

... here we are testing numbers.

Regards,
Yann E. MORIN.

Quoted 9 lines

support/dependencies/check-host-cmake.sh version="$(${cmake} --version \ · line 40

Quoted 2 lines
support/dependencies/check-host-cmake.sh +0 -0
     echo "${cmake}"
 else

baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#3 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Yann E. MORIN

Hi all,

Quoted 9 lines · Yann E. MORIN

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

J?rg

#4 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

J?rg, All,

Thanks for the follow-up! :-)

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 9 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Quoted 5 lines

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

OK, so what we would have to do (basically):

- copy Modules/Platform/Linux.cmake to Modules/Platform/Buildroot.cmake

- tweak that file so that the two settings (lib32 and lib64) are now FALSE in that file

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Or alternatively:

- add Modules/Platform/Buildroot.cmake, which:
- includes Modules/Platform/Linux.cmake
- sets the the two settings (lib32 and lib64) to FALSE

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

Thoughts?

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#5 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Hi Yann,

Quoted 31 lines · Yann E. MORIN

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 9 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Quoted 5 lines

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

OK, so what we would have to do (basically):

- copy Modules/Platform/Linux.cmake to Modules/Platform/Buildroot.cmake

- tweak that file so that the two settings (lib32 and lib64) are now FALSE in that file

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Or alternatively:

Quoted 7 lines

- add Modules/Platform/Buildroot.cmake, which:
- includes Modules/Platform/Linux.cmake
- sets the the two settings (lib32 and lib64) to FALSE

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

Thoughts?

Again, what about host installed cmake? Can we set lib{32,64} to FALSE directly in toolchain.cmake?

baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -
#6 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Baruch, All,

On 2017-02-27 19:31 +0200, Baruch Siach spake thusly:

Quoted 36 lines · Yann E. MORIN
Quoted 35 lines

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 9 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Quoted 5 lines

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

OK, so what we would have to do (basically):

- copy Modules/Platform/Linux.cmake to Modules/Platform/Buildroot.cmake

- tweak that file so that the two settings (lib32 and lib64) are now FALSE in that file

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Or alternatively:

- add Modules/Platform/Buildroot.cmake, which:
- includes Modules/Platform/Linux.cmake
- sets the the two settings (lib32 and lib64) to FALSE

Quoted 5 lines
Quoted 3 lines

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

Thoughts?

Again, what about host installed cmake?

And this is exactly the kind of reply I expected! ;-)

Quoted 2 lines

Can we set lib{32,64} to FALSE
directly in toolchain.cmake?

I don't think so...

The toolchainfile.cmake sets CMAKE_SYSTEM_NAME, which is used to find the appropriate SystemName module. It means the SystemName.cmake file can only be parsed after tolchainfile.cmake is.

So, anything we set in toolchainfile.cmake would be overriden by the settings in the SystemName.cmake file.

Ergo, we're screwed in this case...

Unless we can tell cmake to look for modules in alternate locations.

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#7 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Yann E. MORIN

J?rg, All,

On 2017-02-27 18:25 +0100, Yann E. MORIN spake thusly:

Quoted 29 lines

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 9 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Quoted 5 lines

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

OK, so what we would have to do (basically):

- copy Modules/Platform/Linux.cmake to Modules/Platform/Buildroot.cmake

- tweak that file so that the two settings (lib32 and lib64) are now FALSE in that file

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Quoted 7 lines

Or alternatively:

- add Modules/Platform/Buildroot.cmake, which:
- includes Modules/Platform/Linux.cmake
- sets the the two settings (lib32 and lib64) to FALSE

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

So I tested that last solution, and it indeed fixes the build. Woot!

Of course, this is only for when we build out own cmake, not when we use the host pre-installed one. I'll try to see tonight if we can do similar for it, too.

Thanks for the hints! :-)

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#8 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Quoted 3 lines · Yann E. MORIN

Of course, this is only for when we build out own cmake, not when we use the host pre-installed one. I'll try to see tonight if we can do similar for it, too.

It may make sense to support this upstream. CMake already has platform files for things such as Cray and Android (among others) which are just special kinds of Linux, like Buildroot.

How often does that toolchain file change? Or does it depend on Buildroot settings?

--Ben

#9 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Ben, All,

On 2017-02-27 13:35 -0500, Ben Boeckel spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 3 lines

Of course, this is only for when we build out own cmake, not when we use the host pre-installed one. I'll try to see tonight if we can do similar for it, too.

It may make sense to support this upstream. CMake already has platform files for things such as Cray and Android (among others) which are just special kinds of Linux, like Buildroot.

How often does that toolchain file change? Or does it depend on Buildroot settings?

The toolchainfile.cmake is really specific to one build: it contains the path to the compiler, linker... It contains place-holders that get replaced by a sed invocation:

https://git.buildroot.org/buildroot/tree/package/pkg-cmake.mk#n236

However, the Modules/Platform/Buildroot.cmake would be just as simple as (which is what I use currently):

include(Platform/Linux)
set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB32_PATHS FALSE) set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB64_PATHS FALSE)

I don;t see that changing anytime soon. So, if that would be aceptable for upstream, then we could submit it, yes.

However, I wonder if that's worth the effort...

After all, this is a setting specific to us. I would have a hard time arguing that upstream cmake should carry all such files for all imaginable buildsystems.

And if the solution is that simple and also works for host pre-installed cmake, then we can carry that ourselves, I believe.

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#10 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Yann, Ben,

Quoted 38 lines · Yann E. MORIN

Ben, All,

On 2017-02-27 13:35 -0500, Ben Boeckel spake thusly:

Quoted 7 lines · Yann E. MORIN
Quoted 5 lines

Of course, this is only for when we build out own cmake, not when we use
the host pre-installed one. I'll try to see tonight if we can do similar
for it, too.

Quoted 8 lines

It may make sense to support this upstream. CMake already has platform
files for things such as Cray and Android (among others) which are just
special kinds of Linux, like Buildroot.

How often does that toolchain file change? Or does it depend on Buildroot settings?

The toolchainfile.cmake is really specific to one build: it contains the
path to the compiler, linker... It contains place-holders that get replaced by a sed invocation:

????https://git.buildroot.org/buildroot/tree/package/pkg-cmake.mk#n23 6

However, the Modules/Platform/Buildroot.cmake would be just as simple as (which is what I use currently):

????include(Platform/Linux)
????set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB32_PATHS FALSE) ????set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB64_PATHS FALSE)

I don;t see that changing anytime soon. So, if that would be aceptable
for upstream, then we could submit it, yes.

Quoted 5 lines

However, I wonder if that's worth the effort...

After all, this is a setting specific to us. I would have a hard time arguing that upstream cmake should carry all such files for all imaginable buildsystems.

I replied to Bard King on the CMake mailing list [1]. From my view, there is an issue that /sysroot/usr/lib32 is not reported by the linker or not treated by CMake as an implicit runtime search path. However, I am neither a CMake nor a toolchain expert...

Quoted 3 lines

And if the solution is that simple and also works for host pre- installed
cmake, then we can carry that ourselves, I believe.

Sounds reasonable!

[1] https://cmake.org/pipermail/cmake/2017-February/065064.html

J?rg

#11 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Yann E. MORIN

J?rg, All,

On 2017-02-27 19:30 +0100, Yann E. MORIN spake thusly:

Quoted 36 lines

On 2017-02-27 18:25 +0100, Yann E. MORIN spake thusly:

Quoted 35 lines

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 11 lines · Yann E. MORIN
Quoted 9 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries from
the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/064970.ht ml

Quoted 5 lines

Brad King from Kitware replied today [1]. In short, Brad does not think there anything wrong about handling the rpath and supposes to load a custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.html

OK, so what we would have to do (basically):

- copy Modules/Platform/Linux.cmake to Modules/Platform/Buildroot.cmake

- tweak that file so that the two settings (lib32 and lib64) are now FALSE in that file

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Or alternatively:

- add Modules/Platform/Buildroot.cmake, which:
- includes Modules/Platform/Linux.cmake
- sets the the two settings (lib32 and lib64) to FALSE

Quoted 7 lines
Quoted 1 line

- tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

So I tested that last solution, and it indeed fixes the build. Woot!

Of course, this is only for when we build out own cmake, not when we use the host pre-installed one. I'll try to see tonight if we can do similar for it, too.

It seems that we could set CMAKE_MODULE_PATH to that effect.

More testing under way...

Regards,
Yann E. MORIN.

Quoted 16 lines

Thanks for the hints! :-)

Regards,
Yann E. MORIN.

Signature · 10 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#12 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Quoted 54 lines · Yann E. MORIN

J?rg, All,

On 2017-02-27 19:30 +0100, Yann E. MORIN spake thusly:

Quoted 51 lines

On 2017-02-27 18:25 +0100, Yann E. MORIN spake thusly:

Quoted 38 lines

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 14 lines · Yann E. MORIN
Quoted 5 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries
from
the host.

Quoted 7 lines

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/0 64970.ht
ml

Quoted 8 lines

Brad King from Kitware replied today [1]. In short, Brad does not think
there anything wrong about handling the rpath and supposes to load a
custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065
063.html

OK, so what we would have to do (basically):

? - copy Modules/Platform/Linux.cmake to
Modules/Platform/Buildroot.cmake

? - tweak that file so that the two settings (lib32 and lib64) are now
????FALSE in that file

? - tweak our support/misc/toolchain.cmake to
set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Quoted 8 lines

Or alternatively:

? - add Modules/Platform/Buildroot.cmake, which:
????- includes Modules/Platform/Linux.cmake
????- sets the the two settings (lib32 and lib64) to FALSE

? - tweak our support/misc/toolchain.cmake to
set(CMAKE_SYSTEM_NAME Buildroot)

So I tested that last solution, and it indeed fixes the build. Woot!

Quoted 7 lines
Quoted 5 lines

Of course, this is only for when we build out own cmake, not when we use
the host pre-installed one. I'll try to see tonight if we can do similar
for it, too.

It seems that we could set CMAKE_MODULE_PATH to that effect.

Sounds promising!

Quoted 1 line

More testing under way...

Thanks!

#13 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Yann E. MORIN
Quoted 44 lines · Yann E. MORIN

J?rg, All,

Thanks for the follow-up! :-)

On 2017-02-27 18:12 +0100, J?rg Krause spake thusly:

Quoted 14 lines · Yann E. MORIN
Quoted 5 lines

cmake-3.7 has a bug in how it handles rpath, linking with libraries
from
the host.

Quoted 7 lines

Until we completely understand the issue, just blacklist cmake- 3.7.

The issue has been reported upstream:
????http://public.kitware.com/pipermail/cmake/2017-February/06497 0.ht
ml

Quoted 8 lines

Brad King from Kitware replied today [1]. In short, Brad does not think
there anything wrong about handling the rpath and supposes to load a
custom platform cmake file instead of the Linux one.

[1] http://public.kitware.com/pipermail/cmake/2017-February/065063.
html

OK, so what we would have to do (basically):

? - copy Modules/Platform/Linux.cmake to
Modules/Platform/Buildroot.cmake

? - tweak that file so that the two settings (lib32 and lib64) are now
????FALSE in that file

? - tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

and we'd be all good?

Or alternatively:

Quoted 8 lines

? - add Modules/Platform/Buildroot.cmake, which:
????- includes Modules/Platform/Linux.cmake
????- sets the the two settings (lib32 and lib64) to FALSE

? - tweak our support/misc/toolchain.cmake to set(CMAKE_SYSTEM_NAME Buildroot)

Thoughts?

I still think this is a bug! A host rpath should not be used when cross-compiling whether lib32 is used or not. Somehow, it feels weird to say that Buildroot is not a Linux platform, in the sense of CMake.

I will reply my thoughts to the CMake mailing list. Lets see what they suppose.

J?rg

#14 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Quoted 3 lines · J?rg Krause

I still think this is a bug! A host rpath should not be used when cross-compiling whether lib32 is used or not. Somehow, it feels weird to say that Buildroot is not a Linux platform, in the sense of CMake.

There's the chance that this was always a problem, but is now being exposed by this. Where is the lib32 rpath coming from? CMake doesn't add rpaths by default, so it has to be coming from somewhere. Is it added manually somewhere? Is a library that used to be found in the sysroot now being found on the host? If so, *that's* the bug which should be fixed. Given how widespread it is, maybe something is set up incorrectly in the toolchain file?

This flag is the same logic as lib64, just with a different suffix. Why is the problem not surfacing on Red Hat platforms which use lib64?

--Ben

#15 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Hi Ben,

Quoted 17 lines · Ben Boeckel
Quoted 5 lines · J?rg Krause

I still think this is a bug! A host rpath should not be used when cross-compiling whether lib32 is used or not. Somehow, it feels weird
to say that Buildroot is not a Linux platform, in the sense of CMake.

There's the chance that this was always a problem, but is now being exposed by this. Where is the lib32 rpath coming from? CMake doesn't add
rpaths by default, so it has to be coming from somewhere. Is it added manually somewhere? Is a library that used to be found in the sysroot now being found on the host? If so, *that's* the bug which should be fixed. Given how widespread it is, maybe something is set up incorrectly
in the toolchain file?

Quoted 3 lines

This flag is the same logic as lib64, just with a different suffix. Why
is the problem not surfacing on Red Hat platforms which use lib64?

For a detailed description, please have a look at my post [1] on the Buildroot mailing list some weeks ago.

[1] http://lists.busybox.net/pipermail/buildroot/2017-February/183579.h
tml

J?rg

#16 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Quoted 2 lines · J?rg Krause

For a detailed description, please have a look at my post [1] on the Buildroot mailing list some weeks ago.

Hmm. Is /usr/lib32 supposed to be an implicit runtime path? If so, why is the compiler not indicating that?

--Ben

#17 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Ben Boeckel

Hi Ben,

Quoted 17 lines · Ben Boeckel
Quoted 5 lines · J?rg Krause

I still think this is a bug! A host rpath should not be used when cross-compiling whether lib32 is used or not. Somehow, it feels weird
to say that Buildroot is not a Linux platform, in the sense of CMake.

There's the chance that this was always a problem, but is now being exposed by this. Where is the lib32 rpath coming from? CMake doesn't add
rpaths by default, so it has to be coming from somewhere. Is it added manually somewhere? Is a library that used to be found in the sysroot now being found on the host? If so, *that's* the bug which should be fixed. Given how widespread it is, maybe something is set up incorrectly
in the toolchain file?

Quoted 3 lines

This flag is the same logic as lib64, just with a different suffix. Why
is the problem not surfacing on Red Hat platforms which use lib64?

Note, that the build errors depend on the host system. I was able to reproduce the build errors on a Debian system, whereas on a Arch, the build does not break. However, all system are affected by having a rpath set to a host path. It just not break the build on all host systems. I did not investigated why...

#18 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

In reply to Yann E. MORIN
Quoted 9 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Yann" == Yann E MORIN <[email protected]> writes:

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html

It is a bit painful that we are back to always building cmake on modern distributions, but OK - Correctness is more important.

Committed, thanks.

Signature · 1 line
Bye, Peter Korsgaard
#19 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Peter, All,

On 2017-02-28 09:51 +0100, Peter Korsgaard spake thusly:

Quoted 5 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Yann" == Yann E MORIN <[email protected]> writes:

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Quoted 7 lines
Quoted 4 lines

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html

It is a bit painful that we are back to always building cmake on modern distributions, but OK - Correctness is more important.

To be noted: I have a branch that re-introduces cmake-3.7 (and later versions, I hope) and fixes this rpath issue.

It works for all types of cmake:

- pre-installed host cmake, tested with 3.1 and 3.7

- built cmake, tested with 3.6 and 3.7

I still need to do more testing, as well as proper commit logs. ;-) Expect it Real Soon Now (TM)! :-)

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 223 225 172 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#20 lore raw

PATCH 1/1 dependencies/cmake: blacklist cmake 3.7

Quoted 20 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Yann" == Yann E MORIN <[email protected]> writes:

Peter, All,
On 2017-02-28 09:51 +0100, Peter Korsgaard spake thusly:

Quoted 5 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Yann" == Yann E MORIN <[email protected]> writes:

cmake-3.7 has a bug in how it handles rpath, linking with libraries from the host.

Quoted 7 lines
Quoted 4 lines

Until we completely understand the issue, just blacklist cmake-3.7.

The issue has been reported upstream:

    http://public.kitware.com/pipermail/cmake/2017-February/064970.html

It is a bit painful that we are back to always building cmake on modern distributions, but OK - Correctness is more important.

To be noted: I have a branch that re-introduces cmake-3.7 (and later versions, I hope) and fixes this rpath issue.

Quoted 8 lines

It works for all types of cmake:

- pre-installed host cmake, tested with 3.1 and 3.7

- built cmake, tested with 3.6 and 3.7

I still need to do more testing, as well as proper commit logs. ;-) Expect it Real Soon Now (TM)! :-)

Ok, good - Thanks!

Signature · 1 line
Bye, Peter Korsgaard