enclose statement in %isdarwin conditional #1865
Conversation
|
Could you explain in more details? How Darwin specific (or maybe it's not?) file is affecting Linux? |
|
Run 'scram b -v dxr' with CMSSW_7_6_0_pre6 on slc6 and you will see that a segfault happens while loading libtrilite.so. Removing this line from the spec file and rebuilding is the only change that removed the segault. |
|
[gartung@cmssrv208 CMSSW_7_6_0_pre6]$ scram setup ~/cmsbuild/slc6_amd64_gcc493/external/py2-dxr-toolfile/1.0-kpegke2/etc/scram.d/py2-dxr.xml Finding a value for PY2_DXR_BASE: Finding a value for LIBDIR: Runtime path settings for PYTHONPATH: Runtime path settings for PATH: Setting up py2-dxr version 1.0-kpegke2:
|
|
[gartung@cmssrv208 CMSSW_7_6_0_pre6]$ gdb -ex r --args python /storage/local/data1/gartung/cmsbuild/slc6_amd64_gcc493/external/py2-dxr/1.0-kpegke2/bin/dxr-build.py -v -f src/dxr.config Program received signal SIGSEGV, Segmentation fault. |
|
The version that works has an additional symbols sqlite3_auto_extension [gartung@cmssrv208 py2-dxr]$ nm -C 1.0-cms6/lib/libtrilite.so | grep sql |
|
Ah... the working and non-working version are built against different sqlite versions... [gartung@cmssrv208 py2-dxr]$ diff 1.0-kpegke2/spec 1.0-cms6/spec 27,31c27,31 < BuildRequires: external+llvm+3.6-kpegke2
|
|
This needs to be debugged, but that's wrong: I hate this package, because it's a hacky mess. I think, I had to patch it again for Clang 3.7.0. Probably this is the place where we need to start looking.
The failing version is 1.0-kpegke2, which has a symbol |
|
On Oct 5, 2015, at 2:34 PM, davidlt <[email protected]mailto:[email protected]> wrote: Could you explain in more details? How Darwin specific (or maybe it's not?) file is affecting Linux? — OK. So it is not because of this line. My best guest is that libtrilite was built with sqlite 3.7 and the environment has sqlite 3.8. By rebuilding py2-dxr, libtrilite is rebuilt and linked against sqlite 3.8. |
|
Well, the Makefile is terrible. It basically links to everything on the system and completely ignores "CMSSW environment". It's also using pkg-config. |
|
On Oct 5, 2015, at 3:32 PM, davidlt <[email protected]mailto:[email protected]> wrote: I hate this package, because it's a hacky mess. I think, I had to patch it again for Clang 3.7.0. The master version of DXR now use ElasticSearch instead of libtrilite.so and sqlite3. It probably also has updates for Clang 3.7.0. |
|
We have ES server (or had) for CMSSW IB logs. |
|
OK. I will submit an updated DXR spec that builds the master branch. On Oct 5, 2015, at 3:47 PM, davidlt <[email protected]mailto:[email protected]> wrote: We have ES server (or had) for CMSSW IB logs. — |
|
Because py2-dxr is becoming hard to maintain I propose we remove it from the cmssw-tool-conf. |
|
@davidlt As long as a rebuild against sqlite 3.8 is forced the segfault goes away. I removed py2-dxr and its requirements from cmssw-tool-conf.spec but I would still like to keep the dxr build option. |
|
@gartung As it is right now this PR cannot be merged. Please do a rebase so we can test it. |
| Requires: python zlib py2-setuptools py2-futures py2-jinja py2-markupsafe py2-ordereddict py2-parsimonious | ||
|
|
||
| Requires: python llvm sqlite zlib py2-setuptools py2-futures py2-jinja py2-markupsafe py2-ordereddict py2-parsimonious | ||
| %define isdarwin %(case %{cmsos} in (osx*) echo 1 ;; (*) echo 0 ;; esac) |
There was a problem hiding this comment.
There is no need for this line.
|
So why do we remove it? We only need to rewrite a small Makefile to look into correct places for us. Should be relatively easy. |
|
Will resubmit with just the last change on a branch. |
@davidlt @smuzaffar @Dr15Jones
This edit cause a segmentation fault in the linux version of this library when this is user