- User Since
- Aug 15 2016, 6:00 AM (69 w, 2 d)
Mon, Dec 11
Thu, Nov 30
Mon, Nov 27
Updated to fix unittest failure.
Sat, Nov 25
Sat, Nov 18
Or even more abstract (if we assume we don't have to check for presence of llvm_tools_dir):
To be honest, as I said before, I'm entirely confused by the logic there, with all the prepending, appending and shuffling around. But if you believe it gives the correct result, I'm all for it.
Fri, Nov 17
I don't see how this could break anything, and it's a pretty straightforward change.
Thu, Nov 16
Please revert this commit. You've just broken all the stand-alone builds of clang.
Wed, Nov 15
I'm worried that having a different first component of VERSION and SOVERSION could confuse some tools like ldconfig. I'm going to see if we can solve it via different install prefixes with RPATHs first.
Tue, Nov 14
Hmm, that's an interesting option too. I was thinking of using LLVM_VERSION_SUFFIX originally but it obviously did not go to SOVERSION. Can this have any unforeseen consequences? Should VERSION become *major+suffix-0-minor* then?
We = Gentoo. What for is already described = to install libstdc++ and libc++ versions of LLVM simultaneously for rebuild period. How long = for the time being, or until we find a better working solution.
Nov 11 2017
Yay, I've found a way to get the desired linkage! ;-)
Nov 10 2017
I've rewritten the tests to use unittest, so this patch needs to be updated now. However, you may want to wait a while since @compnerd promised to look at it and he might have a way to get uniqueExternal() linkage.
Thanks. I'll update it to incorporate the lately committed test fixes, and push it later today.
Nov 9 2017
Let's try some more reviewers.
Nov 8 2017
I see that you haven't more luck than I did finding a better solution. How about the code completion failure? I suppose I still have to figure that one out on my own ;-).
Duplicate of D39161.
@frutiger, do you maybe happen to have fixes for the two other test failures? Maybe I'm tracking them down unnecessarily ;-).
Nov 7 2017
====================================================================== FAIL: tests.cindex.test_code_completion.test_code_complete_availability ---------------------------------------------------------------------- Traceback (most recent call last): File "/usr/lib64/python2.7/site-packages/nose/case.py", line 197, in runTest self.test(*self.arg) File "/usr/src/llvm/tools/clang/bindings/python/tests/cindex/test_code_completion.py", line 75, in test_code_complete_availability check_completion_results(cr, expected) File "/usr/src/llvm/tools/clang/bindings/python/tests/cindex/test_code_completion.py", line 10, in check_completion_results assert c in completions AssertionError
The diffs aren't very useful since I had to add a class for each test unit and so everything needed reindenting. You can take my word that changes boil down to:
Nov 2 2017
Nov 1 2017
Oct 31 2017
Oct 25 2017
Oct 17 2017
Oct 12 2017
Ok, here's another idea. Since the objects have been compiled already (and they explicitly depend on gtest via their own deps), let's just remove the extraneous gtest dep.
Ok, I think I've found a better solution. Will start submitting patches soonish.
Oct 2 2017
Yes, I did apply it both to the installed CMake files and the external LLVM source tree used by compiler-rt. I can't see where else it could have gotten an old version from, and I don't grep any AddLLVM includes either.
Well, I think I did everything right, yet I still get:
Could you give me a few minutes to test it before merging, please?
Oct 1 2017
Sep 30 2017
Sep 8 2017
It seems that somebody already moved the file in D37490.
Sep 1 2017
Nevermind. I found out what's wrong via looking at the older patch versions.
Aug 31 2017
Aug 30 2017
Aug 29 2017
Thanks a lot for taking care of this.
I'm happy to help with whatever needs to be done to keep breakage to the minimum, provided that we determine some clear path forward that doesn't involve regressing to the half-broken state for non-Android Linux systems and it's doable in the free time I can spare. However, I should point out that it was a very bad idea to hardcode paths all over the projects in the first place.
Aug 28 2017
Thanks for the clarification.