Some of bundled LADSPA plugin causes crash on some system #4030
Comments
|
@Martin1995, That said, we have made some changes to TAP plugins very recently, so that's a good place to start looking. #3960 If you think some programmers can help you iron this out locally, we have a dedicated |
|
While trying to reinstall lmms i had looked a littlebit around inside the PKGBUILD file that is used on every version of Arch to install it from the user repository and it seems that is using the latest sources from github |
|
That [git branch] changes with time. It's currently stable-1.2 branch but will switch back to master when stable-1.2 is released. The stable-1.2 branch is a good branch to run, it's modern and stable, however this doesn't really help point at the problem. When an undefined symbol occurs it's often a sign that a precompiled binary has changed. Since we do not yet offer any precompiled binaries for Linux we can't own this bug. As a temporary measure, you can move the tap plugins out of the way. LMMS does not require them to run. If you can reproduce this when building [manually] from source, it would help us identify it. This would help us gather more information about your build environment. |
|
ok i will try this tomorrow |
|
If i'm trying to build it directly from github then i get some config errors out of cmake (see attached txt-file) |
|
@Martin1995, |
|
ok my fault, im not so familiar with git, because i'm not a programmer i only tinker around with arduino and shell scripts. |
That's fine, submodules are not very intuitive to deal with in git yet. Thanks for the update. |
|
oh i forgot testing the software itself, compiling worked now but if i start the software three errors are gone but the software crashes anyway. i think libQt5Core.so.5 is the culprit but im not sure backtrace |
|
@Martin1995 thanks for the backtrace. This is a regression caused by #3872. I'm tagging @BumblingCoder and reopening. |
tresf
changed the title
I got similar crash when I tried to load |
|
As @PhysSong has illustrated, the TAP issue is a much less severe error that happened to appear before the segfault, it should be handled as a standalone plugin and not segfault LMMS. So there were actually two unrelated issues in the same bug report. We'll leave open so that the TAP issue can be confirmed as fixed as well. @PhysSong thanks for the background info on the compiler. |
|
Oh i hadn't known that you also have a precompiled portable version, just downloaded it and the portable version works fine on Arch. I think i have also to ask for help in the Arch Linux forum, somebody could also know a way to fix this Problem on Arch. |
|
Problem was solved by itselft, i don't know how and actually what cuased the Problemn, but i'm happy that LMMS works again. Thank you all for your help. |
|
I can reproduce the |
|
@PhysSong - a question - if you open |
|
It's related to LADSPA plugin build issue with new compilers, so that's not related. |
|
For some reason, It seems to work fine on |
PhysSong
changed the title
|
It could be solved by this patch: diff --git a/plugins/LadspaEffect/tap/CMakeLists.txt b/plugins/LadspaEffect/tap/CMakeLists.txt
index 0ae8ae54e..fd017555d 100644
--- a/plugins/LadspaEffect/tap/CMakeLists.txt
+++ b/plugins/LadspaEffect/tap/CMakeLists.txt
@@ -12,16 +12,10 @@ FOREACH(_item ${PLUGIN_SOURCES})
INSTALL(TARGETS "${_plugin}" LIBRARY DESTINATION "${PLUGIN_DIR}/ladspa")
SET_TARGET_PROPERTIES("${_plugin}" PROPERTIES PREFIX "")
# Don't strip if "Debug" or "RelWithDebInfo"
- IF(LMMS_BUILD_APPLE)
- SET_TARGET_PROPERTIES("${_plugin}" PROPERTIES LINK_FLAGS "${LINK_FLAGS} -Bsymbolic -lm")
- ELSEIF(NOT LMMS_BUILD_OPENBSD)
- SET_TARGET_PROPERTIES("${_plugin}" PROPERTIES LINK_FLAGS "${LINK_FLAGS} -shared -Wl,-no-undefined -Wl,-Bsymbolic -lm")
- ENDIF()
- IF(LMMS_BUILD_LINUX OR LMMS_BUILD_HAIKU)
- SET_TARGET_PROPERTIES("${_plugin}" PROPERTIES LINK_FLAGS "${LINK_FLAGS}")
- ENDIF()
+
+ target_link_libraries ("${_plugin}" m)
ENDFOREACH()
We basically have to use |
|
Recent gcc compiler use vector math when both Notes about dropped flags:
|
|
@PhysSong Thanks for the detailed explanation; However, the same error still exist when specifically adding This is the command before my patch (no libmvec): And this is after my patch (have libmvec): Am I missing another difference other than the place of |
No, that's THE reason. Some compiler/linker flags may lead to different results when their places are changed. In this case, @Reflexe Do you have this issue on |
|
@Reflexe Do you have a backtrace for |
It would only happen when I compile lmms from the console with Comping with QtCreator is ok. |
|
Strange. I'll try 1.2 with from both manual |
|
Test summary:
|
|
@PhysSong The difference between QtCreator and manually running cmake was probably the compiler (QtCreator: Clang, Default: Gcc) |
|
It seems like we're setting compiler flags for SWH in 1.2 in wrong way. The second lmms/plugins/LadspaEffect/swh/CMakeLists.txt Lines 20 to 24 in 407973a As of CMake 2.8.12,
Agreed. That works for old version of CMake, too. For some issues with adding flags/libraries, I opened a new issue #4430.
I found that |
Add math library using target_link_libraries instead of setting LINK_FLAGS properties with target_link_libraries.
if i try to start lmms on my Arch Linux it crashes after the splashscreen since today.
The console gives me a little bit more output:
Edited by tresf, added backtrace:
can you help me please, because the git version was the better running solution for my system.
The text was updated successfully, but these errors were encountered: