Am 10.10.2015 um 21:14 schrieb Tim Northover: > On 10 October 2015 at 04:16, Joachim Durchholz via llvm-dev > <llvm-dev at lists.llvm.org> wrote: >> is this the right place to discuss getting LLVM built using Eclipse, >> as described on http://llvm.org/docs/GettingStarted.html? > > It's probably as good as anywhere, Okay. > but I don't think there's much > that's different from a regular build. As far as I'm aware you just > tell CMake to generate an Eclipse project and hope you've got enough > memory to load it. 32 GB should be enough for everyone ;-) ... oh darnit. Right. I forgot that Eclipse comes configured to use just 512 MB and needs to be reconfigured for more. I just finished a test run with an allowance of 16 GB. The biggest pre-GC memory usage was 6.9 G, the biggest post-GC usage 6.9 GB. From this I come to the following recommendations: - 2 GB as the absolute minimum, - 3 GB "likely workable", and - 7 GB optimum. > I don't really use Eclipse, but it seemed to work OK here after I'd > removed tools/clang/INPUTS/macro_pounder_fn.c (there's probably a way > to tell Eclipse not to bother indexing that file, but I don't know > it). > > What problem are you seeing? It ran straight into GC thrashing, that's what happened - with enough memory, it took a mere 20 minutes (Core i7 with an SSD, YMMV). Now I'm hitting exceptions when reopening the project in Eclipse, and stopping the indexer, which bad though I don't know how bad. It's on Eclipse Mars, so if Luna does not have this problem, don't upgrade your Eclipse yet! I'll go and take this to the CDT bug tracker. Regards, Jo
> Now I'm hitting exceptions when reopening the project in Eclipse, and > stopping the indexer, which bad though I don't know how bad.I definitely think you want to do something about macro_pounder_fn.c. It's *designed* to kill compiler-like things, and the indexer ran fine for me after I'd dealt with it. Cheers. Tim.
Hi Joachim, Le lundi 12 octobre 2015 à 00:15 +0200, Joachim Durchholz via llvm-dev a écrit :> It ran straight into GC thrashing, that's what happened - with enough > memory, it took a mere 20 minutes (Core i7 with an SSD, YMMV). > > Now I'm hitting exceptions when reopening the project in Eclipse, and > stopping the indexer, which bad though I don't know how bad. > It's on Eclipse Mars, so if Luna does not have this problem, don't > upgrade your Eclipse yet!Maybe you are hitting the same problem I had a while ago. It seems to be patched now, but, for me, I had to patch Eclipse myself. Have a look at https://bugs.eclipse.org/bugs/show_bug.cgi?id=419614 Regards, Gael
Am 12.10.2015 um 00:26 schrieb Tim Northover: >> Now I'm hitting exceptions when reopening the project in Eclipse, and >> stopping the indexer, which bad though I don't know how bad. > > I definitely think you want to do something about macro_pounder_fn.c. > It's *designed* to kill compiler-like things, Yeah, I noticed. Torture rack for the macro preprocessor via exponential blowup... > and the indexer ran fine > for me after I'd dealt with it. Surprisingly enough, the indexer didn't hang on it, nor did the exceptions happen in it, so I guess I'll have to look elsewhere. Of course the macro pounder might cause problems once I do more than just import the project, I'll wait and see. Regards, Jo
Am 12.10.2015 um 08:08 schrieb Gaël Jobin via llvm-dev:> Maybe you are hitting the same problem I had a while ago. It seems to > be patched now, but, for me, I had to patch Eclipse myself. Have a look > at https://bugs.eclipse.org/bugs/show_bug.cgi?id=419614In .metadata/.log, I'm seeing a gazillion of "Detected circular reference between inline namespaces" from http://git.eclipse.org/c/cdt/org.eclipse.cdt.git/commit/?id=14f7d3e16c69900b12c74278793c6f6d0133bcb7 I guess that particular workaround is already part of the Eclipse Mars CDT distribution, which is nice. https://bugs.eclipse.org/bugs/show_bug.cgi?id=419614#c26 listed the directories to exclude from indexing to prevent the problem from happening. I'll have to try that and see whether my problem goes away. I submitted my problem as https://bugs.eclipse.org/bugs/show_bug.cgi?id=479528 . Regards, Jo