c2bf874 | Axel Naumann | 06 February 2014, 14:06:11 UTC | We cannot dlopen a binary that the dictionary is linked in. dyLibName will contain the binary name; dlopen will fail on it. Make this a gDebug Info(), not an Error(). | 06 February 2014, 14:07:37 UTC |
1692adf | Vassil Vassilev | 06 February 2014, 13:36:59 UTC | XFAIL clang's testsuite for now. | 06 February 2014, 13:36:59 UTC |
9293a6b | Vassil Vassilev | 06 February 2014, 13:28:46 UTC | Merge branch 'unloading' | 06 February 2014, 13:28:46 UTC |
5a82b75 | Vassil Vassilev | 29 January 2014, 14:25:08 UTC | Fix iterators. Two things first the iterator wasn't considered to be empty. Second the loop written such assumes chances to the collection within the loop which is not the case. | 06 February 2014, 13:28:10 UTC |
46d7bd7 | Vassil Vassilev | 06 February 2014, 10:49:38 UTC | Fix windows warnings. | 06 February 2014, 13:28:10 UTC |
403878f | Vassil Vassilev | 06 February 2014, 09:41:09 UTC | Get rid of removeIfLast. | 06 February 2014, 13:28:10 UTC |
1be7b7a | Vassil Vassilev | 06 February 2014, 09:36:26 UTC | Add handling for FunctionTemplateDecl-s and ClassTempalateDecls. After the update of removeIfLast, they fall into the 'ordinary' cases. | 06 February 2014, 13:28:10 UTC |
685bcfa | Vassil Vassilev | 06 February 2014, 09:33:40 UTC | If the lookup vector doesn't contain decls remove the entry from the map. The lookup table is highly optimized: a single data member can be a single decl or a vector with decls. In the regular case when clang only adds declarations the single decl gets promoted to a vector. The decl reverter doesn't bother to demote the vector into a single decl (for now) Insted if we see the vector is empty we remove the lookup entry from the map and thus we make the compareState happy. | 06 February 2014, 13:28:10 UTC |
f312e66 | Vassil Vassilev | 06 February 2014, 09:29:27 UTC | Simplify the removeIfLast. This soon will disappear. In the past we used that routine as sanity check whether we are reverting in the correct order. Now we have store/compare state. Moreover this is wrong because the the template specializations create a mess in the order, because the pattern is being added to the context but not the underlying declarations. This breaks the logic of the visitor as it needs to have two completely different code paths for templates and the rest of the decls only because we wanted to keep this logic. Now is a lot simpler. Remove the decl if it was in the context, without providing any sanity checks what so ever. | 06 February 2014, 13:28:10 UTC |
cb5d2c5 | Vassil Vassilev | 06 February 2014, 09:27:06 UTC | Simplify the update of the lookup tables when visiting a redeclarable. The key idea is that the decl that is being detached from the redecl chain can be in registered in the lookup tables. In this case we need to check if there is another decl in the chain and update the lookup table so that the existing decls in the chain could be accessible. | 06 February 2014, 13:28:09 UTC |
6db66b2 | Vassil Vassilev | 06 February 2014, 09:26:30 UTC | Style. | 06 February 2014, 13:28:09 UTC |
92e40ce | Vassil Vassilev | 06 February 2014, 09:25:59 UTC | Move template handling functions in a separate 'section' | 06 February 2014, 13:28:09 UTC |
b3b923d | Vassil Vassilev | 05 February 2014, 10:28:27 UTC | FunctionTemplateDecls don't get registered in the lookup. They have templated decls which are the ones that need to be reverted. | 06 February 2014, 13:28:09 UTC |
3db30cb | Vassil Vassilev | 03 February 2014, 18:43:50 UTC | Adjust to the test with the correct vars. | 06 February 2014, 13:28:09 UTC |
2334a20 | Vassil Vassilev | 03 February 2014, 18:24:04 UTC | Add a test that loads every clang test and unloads it | 06 February 2014, 13:28:09 UTC |
92b39b3 | Vassil Vassilev | 03 February 2014, 18:15:30 UTC | Test improvement of the lookup (should be ignored) | 06 February 2014, 13:28:09 UTC |
2d68143 | Vassil Vassilev | 03 February 2014, 18:14:48 UTC | Keep the files diff files until the end of the session | 06 February 2014, 13:28:09 UTC |
fd8e072 | Vassil Vassilev | 03 February 2014, 18:12:58 UTC | Prepend cling prefix, which makes the file deletion easier. | 06 February 2014, 13:28:08 UTC |
78e8886 | Vassil Vassilev | 30 January 2014, 13:01:23 UTC | Blanks. | 06 February 2014, 13:28:08 UTC |
ffe97a0 | Vassil Vassilev | 30 January 2014, 13:00:43 UTC | Blanks. | 06 February 2014, 13:28:08 UTC |
70eb2be | Vassil Vassilev | 30 January 2014, 12:58:39 UTC | In rawInput mode we want to be as close as possible to the compiler. | 06 February 2014, 13:28:08 UTC |
7151db7 | Vassil Vassilev | 29 January 2014, 16:02:46 UTC | Allow the unloading even in -fsyntax-only mode. At least for verification purposes it is very useful. | 06 February 2014, 13:28:08 UTC |
80a2c14 | Vassil Vassilev | 29 January 2014, 14:48:13 UTC | Add support for C tentative definitions. | 06 February 2014, 13:28:08 UTC |
68660c3 | Vassil Vassilev | 29 January 2014, 14:37:10 UTC | Blanks. | 06 February 2014, 13:28:08 UTC |
764642a | Vassil Vassilev | 29 January 2014, 14:25:08 UTC | Fix iterators. Two things first the iterator wasn't considered to be empty. Second the loop written such assumes chances to the collection within the loop which is not the case. | 06 February 2014, 13:28:08 UTC |
bef9239 | Vassil Vassilev | 29 January 2014, 14:23:49 UTC | In -fsyntax-only mode we don't have llvm::Module. | 06 February 2014, 13:28:08 UTC |
7b561bd | Timur Pocheptsov | 06 February 2014, 13:12:26 UTC | We agreed on 139 :) (+ 128). | 06 February 2014, 13:12:26 UTC |
bd3e960 | Timur Pocheptsov | 06 February 2014, 12:48:47 UTC | Remove + 128 part since it's just a one particular shell's specific. | 06 February 2014, 12:48:47 UTC |
bd5e143 | Timur Pocheptsov | 06 February 2014, 12:33:01 UTC | Fix a comment. | 06 February 2014, 12:33:01 UTC |
01006e1 | Timur Pocheptsov | 06 February 2014, 12:27:47 UTC | Exit on signal with a real signal core (+ high order bit set???) instead of our enumerators (which has no meaning outside ROOT). | 06 February 2014, 12:27:47 UTC |
5834fc8 | Lorenzo Moneta | 06 February 2014, 11:16:59 UTC | Add stressHistfactory test only when ROOT is built with xml support | 06 February 2014, 11:16:59 UTC |
b58d192 | Lorenzo Moneta | 06 February 2014, 10:35:39 UTC | Fix makefile test for stressHistFactory | 06 February 2014, 10:49:24 UTC |
448440d | Lorenzo Moneta | 06 February 2014, 10:34:51 UTC | Remove usage of TRef in HistFactory. It is not needed. Replace with a simple wrapped class to manage histogram pointer (RooStats::HistFactory::HistRef) This should fix a bug reported by Swagato Fix also memory leaks created by cloning histograms in histooworkspace Add a histfactory test from Christian (stresshistFactory ) in $ROOTSYS/test | 06 February 2014, 10:49:24 UTC |
6acd22c | Lorenzo Moneta | 05 February 2014, 14:14:45 UTC | Make TH1::Copy public , to be able to do deep copy of TH1 pointers without using TH1::Clone | 06 February 2014, 10:48:23 UTC |
8fa001c | Axel Naumann | 06 February 2014, 09:16:28 UTC | Expose symbols of the currently loading library to the JIT (ROOT-6015). While dlopening a library, its symbols cannot be found by dlsym(RTLD_GLOBAL,...). The JIT can thus not resolve symbols from the library during dictionary initialization. Thus we need to explicitly include the library. This is now done through the LazyFunctionCreation interface. | 06 February 2014, 09:16:28 UTC |
8282286 | Philippe Canal | 05 February 2014, 22:24:12 UTC | Add missing call to PostLoadCheck. | 05 February 2014, 22:24:12 UTC |
46c2954 | Olivier Couet | 05 February 2014, 16:16:44 UTC | Make sure the hist fill color and frame fill color are different in case of option BAR or HBAR. | 05 February 2014, 16:17:15 UTC |
ab714c8 | Axel Naumann | 05 February 2014, 15:46:12 UTC | Use LC_ALL to unfrenchify (ROOT-6017). | 05 February 2014, 15:46:15 UTC |
91e6825 | Olivier Couet | 05 February 2014, 15:26:15 UTC | fix typos. | 05 February 2014, 15:26:15 UTC |
f39c7f0 | Axel Naumann | 05 February 2014, 14:44:15 UTC | Revert "Search the symbol in all available dylibs (ROOT-6015)." This reverts commit 706b8a1c17c780fb5d28ad2586a8842e45b7f051. DLRT_DEFAULT is alreday searched; it is one of the OpenedHandles. | 05 February 2014, 14:44:49 UTC |
35c4b64 | Axel Naumann | 05 February 2014, 14:44:07 UTC | Revert "Fix fatal typo" This reverts commit 84cb95c01813a49ce278614b7180de005665041d. | 05 February 2014, 14:44:49 UTC |
0ba8120 | Axel Naumann | 05 February 2014, 14:43:58 UTC | Revert "Fix typo a la llvm coding convention." This reverts commit ce6de2269830de3e0eb72dcade4826cb5f3538ab. | 05 February 2014, 14:44:49 UTC |
b5db4bb | Axel Naumann | 05 February 2014, 11:15:42 UTC | Syncronize with our clang repo: minimize diffs. | 05 February 2014, 14:44:49 UTC |
2c1ce4c | Lorenzo Moneta | 05 February 2014, 13:53:37 UTC | Apply a patch from Bertrand to fix warning C4436 from VS (dynamic_cast from virtual base 'class1' to 'class2' in constructor or destructor ) | 05 February 2014, 13:55:27 UTC |
178345c | Lorenzo Moneta | 05 February 2014, 10:47:08 UTC | Copy from 5.34 patches optimisation in HistFactory (FlexibleInt. and PicewiseInt.) | 05 February 2014, 13:50:37 UTC |
ce6de22 | Axel Naumann | 05 February 2014, 10:18:40 UTC | Fix typo a la llvm coding convention. | 05 February 2014, 10:18:40 UTC |
9168a74 | Axel Naumann | 05 February 2014, 10:16:43 UTC | Report if we cannot extract the stdlib -I. | 05 February 2014, 10:17:20 UTC |
f195153 | Philippe Canal | 05 February 2014, 05:06:10 UTC | Avoid using deleted memory. Whenever an emulated TClass is deleted, make sure to remove it from the list of class to update (so that we don't access deleted memory in RegisterModule and trigger the initialization of the class twice). | 05 February 2014, 05:44:37 UTC |
8cfa620 | Philippe Canal | 05 February 2014, 04:09:02 UTC | Delegate the Update of the TClass to the CollectionProxy. This allows to solved ROOT-6040 by avoid to have to initialize the collection proxy just of find that ... we don't need to update it. Having the useless update can lead to nested dlopen. When a dlopen loads a dictionary for a TClass that was already created (in emulated mode), it will tell all the other TClass and all the CollectionProxy to update the possible reference. If at that point we initialize a collection proxy which content is a represented by a TClass that has not yet been loaded but has a rootmap file, we will end up with a nested call to dlopen to loaded the library. | 05 February 2014, 05:43:32 UTC |
84cb95c | Philippe Canal | 04 February 2014, 21:13:36 UTC | Fix fatal typo | 04 February 2014, 21:13:36 UTC |
706b8a1 | Axel Naumann | 04 February 2014, 17:33:38 UTC | Search the symbol in all available dylibs (ROOT-6015). During static init the JIT could not find symbols in a library that the currently loaded library depends on. This was caused by cling not having updated the list of loaded libraries yet - and only that list was searched for symbols. Instead, now the dl handle 0 is searched, using the supposedly optimized symbol lookup tables of the process. This leaves symbol priorities / lookup context to be dealt with: a library might not see the same symbols as the process. | 04 February 2014, 19:07:56 UTC |
f267eee | Philippe Canal | 04 February 2014, 16:08:19 UTC | Fix ROOT-6038 by silencing the class creation during InheritsFrom | 04 February 2014, 16:08:53 UTC |
cebfcfd | Philippe Canal | 04 February 2014, 15:52:20 UTC | Early return in TClass::InheritsFrom | 04 February 2014, 16:08:53 UTC |
d3a54f3 | Philippe Canal | 04 February 2014, 15:51:57 UTC | In TClass::GetClass find typedef info even if it was not search for before | 04 February 2014, 16:08:52 UTC |
489af02 | CristinaCristescu | 04 February 2014, 13:37:31 UTC | TabCompletion: remove printing of object address. | 04 February 2014, 15:11:29 UTC |
98a3ee2 | Axel Naumann | 04 February 2014, 15:08:55 UTC | Add ref-getters for GenericValue; do not copy it unnecessarily. | 04 February 2014, 15:08:55 UTC |
89f8c1e | Timur Pocheptsov | 04 February 2014, 14:41:56 UTC | Add info about "polygon compression". | 04 February 2014, 14:41:56 UTC |
24f4396 | Timur Pocheptsov | 04 February 2014, 14:29:09 UTC | To deal with my terrific polygons, Quartz is required. | 04 February 2014, 14:29:09 UTC |
f2e29e2 | Timur Pocheptsov | 04 February 2014, 14:22:36 UTC | Comments and custom color names. | 04 February 2014, 14:22:36 UTC |
2a57107 | Timur Pocheptsov | 04 February 2014, 14:14:25 UTC | Nice good old tutorial I found and modified to use transparency (Cocoa). | 04 February 2014, 14:14:51 UTC |
6ca563c | Axel Naumann | 04 February 2014, 13:15:01 UTC | Force non-French; ROOT-6017. | 04 February 2014, 13:15:04 UTC |
11436e2 | Bertrand Bellenot | 04 February 2014, 10:17:18 UTC | Fix text output color in stressHistoFit - Make text output color working on any color scheme - Implement color text output on Windows, avoiding weird (control) characters output | 04 February 2014, 10:17:18 UTC |
7c2df3d | Philippe Canal | 04 February 2014, 04:17:04 UTC | 'Quick' fix for ROOT-5876 (and related issues) If you feels like this is using a bandages to fix a leaking dam, you are likely correctly. More technically, this avoid the routine that use GetInterpreterTypeName from re-expanding the name (during the overly complex and obsolete scan through the list of possible names). This avoids infinite recurse for vector<Long64_t> and map<Type,ABC_t> | 04 February 2014, 05:14:10 UTC |
497e663 | Philippe Canal | 04 February 2014, 03:52:56 UTC | Fix TClassEdit::IsDefAlloc Fix a typo (consume key and value) and correct the algorithm in the non-pair case (was forgetting to skip std:: consistently). This solves ROOT-5999. | 04 February 2014, 05:14:09 UTC |
3a642e1 | Philippe Canal | 03 February 2014, 22:21:27 UTC | Work around ROOT-6012, temporary fix for ROOT-6021. The failure in ROOT-6012, means that the error in ROOT-6021, was correct and HepMC::GenParticle's dictionary was not properly loaded. By activating the RegisterTClassUpdate for all already created TClass (from a given dictionary) we make sure that the TClass are properly loaded. | 03 February 2014, 22:26:45 UTC |
515d321 | Axel Naumann | 03 February 2014, 17:55:23 UTC | Be more verbose when unable to detect the stdlib version. | 03 February 2014, 17:55:23 UTC |
7ca15eb | Pere Mato | 03 February 2014, 17:40:16 UTC | Make stressGraphics to work (CTest) | 03 February 2014, 17:40:16 UTC |
b0c2281 | Philippe Canal | 03 February 2014, 14:57:45 UTC | Fix typo in doxygen doc | 03 February 2014, 14:58:05 UTC |
6bdcf38 | Bertrand Bellenot | 03 February 2014, 13:02:56 UTC | Fix compilation on Windows | 03 February 2014, 13:02:56 UTC |
229c3b3 | Sven Kreiss | 01 February 2014, 15:59:45 UTC | Add statconfig to copy constructor. | 03 February 2014, 11:52:30 UTC |
c064379 | Lorenzo Moneta | 31 January 2014, 21:39:37 UTC | When calling Fitter::CalculateMinosErrors switch off the FItCOnfig::MinosErrors flag to avoid that following calls of minimisation computes also the Minos errors. This fixes ROOT-6013 | 03 February 2014, 11:52:30 UTC |
362ab14 | Timur Pocheptsov | 03 February 2014, 08:00:17 UTC | Either auto only, or good old correct declarations. | 03 February 2014, 08:00:17 UTC |
4c02672 | Timur Pocheptsov | 03 February 2014, 07:57:46 UTC | Why auto here???? It looks like we took the code written by Bjarne's grandpa and decided to "make it modern" by adding auto. | 03 February 2014, 07:57:46 UTC |
7555cd8 | Timur Pocheptsov | 02 February 2014, 10:03:05 UTC | '#' can not be a delimiter, only whitespaces/newlin-character can follow after 'fNvar' number. | 02 February 2014, 10:03:05 UTC |
c068ba0 | Philippe Canal | 01 February 2014, 23:58:12 UTC | Disable support for inline namespace in TClassEdit This implementation is to aggresive (strips also non inline namespace inside std) and fails roottest on many platforms. This removal also breaks support for inline namespace (i.e more test fails with libc++ ... :(). | 01 February 2014, 23:58:12 UTC |
9a49c92 | Wim Lavrijsen | 01 February 2014, 06:45:03 UTC | fix for ROOT-5933 | 01 February 2014, 06:45:45 UTC |
50304a4 | Philippe Canal | 01 February 2014, 02:43:07 UTC | Fix handling of partially specialized class template. When handling a partially specialized class template, the index of the template parameter to be replaced was not properly calculated (confusion between the overall index and the index within the non-specialized part). This should fix ROOT-6019. | 01 February 2014, 02:43:07 UTC |
65a5258 | Philippe Canal | 01 February 2014, 02:11:52 UTC | Speed-up THtml. It has been slowed down by the change in TClass::GetBaseClass which now calls TClass::GetClass (the time is spend in the clang lookup/search). Remove the optimization in TClass::InheritsFrom(const char *classname) as it no longer make any difference (TClass::GetClass called in both case) | 01 February 2014, 02:11:52 UTC |
0139e1e | Philippe Canal | 01 February 2014, 00:45:59 UTC | Introduce Transform::Config::DropDefaultArg Currently only implemented for the STL collection. This could eventually be used to allow customization of which default argument to strip and which class template. For now, it is only used to prevent the addition of template default template argument to the STL collection ... which we would any strip. This fixes the secondary issue in ROOT-6020 (one class template instance having more than one corresponding TClass due to the inconsistency in the normalization (not adding vs not stripping) | 01 February 2014, 00:45:59 UTC |
d466df3 | Philippe Canal | 01 February 2014, 00:25:38 UTC | Always go through a TClass to search for base class. To solve ROOT-6020, use TClass::GetClass to resolve the string being passed so that we completely and properly resolve the classname (and reduce the cost of the search). | 01 February 2014, 00:25:38 UTC |
29c5c43 | Philippe Canal | 01 February 2014, 00:23:32 UTC | Move IsStdClass to cling::utils::Analyze | 01 February 2014, 00:23:32 UTC |
158d808 | Wim Lavrijsen | 31 January 2014, 21:43:27 UTC | fixes for ROOT-5836 | 31 January 2014, 21:43:27 UTC |
4c5465e | Philippe Canal | 31 January 2014, 18:46:24 UTC | Use typedef for array types | 31 January 2014, 18:46:24 UTC |
36cc1f4 | Philippe Canal | 31 January 2014, 17:20:43 UTC | Dont't strip the part after std::string. This fix the 2nd part of ROOT-6023 (where the array part was forgotten) | 31 January 2014, 17:20:43 UTC |
e4f4793 | Fons Rademakers | 31 January 2014, 14:58:10 UTC | Fix for broken TSAXParserCallback::{Error,Warning,FatalError}. ROOT-6027. By Oliver Freyermuth. | 31 January 2014, 14:58:10 UTC |
81bfa44 | Timur Pocheptsov | 31 January 2014, 14:22:59 UTC | Remove old and useless "test". | 31 January 2014, 14:22:59 UTC |
b408d13 | Philippe Canal | 31 January 2014, 11:52:31 UTC | Be even more careful to avoid (unintended) initiation. This should fix ROOT-6025. | 31 January 2014, 11:52:31 UTC |
43ef4b6 | Philippe Canal | 31 January 2014, 11:31:11 UTC | Mark std::string as non-split. This is necessary on libcxx (MacOS 10.9) | 31 January 2014, 11:31:25 UTC |
db1b50d | Pere Mato | 31 January 2014, 09:47:25 UTC | Move tutorials when running tests in build directory | 31 January 2014, 09:47:25 UTC |
80d3eea | Bertrand Bellenot | 31 January 2014, 09:18:12 UTC | Merge branch 'master' of https://root.cern.ch/git/root | 31 January 2014, 09:18:12 UTC |
9b00aa5 | Bertrand Bellenot | 31 January 2014, 09:17:06 UTC | Skip the "file:" protocol (if any) at the beginning of the path name in TWinNTSystem::ExpandPathName | 31 January 2014, 09:17:06 UTC |
1989927 | Pere Mato | 31 January 2014, 08:37:59 UTC | The C compiler variable is CMAKE_C_COMPILER | 31 January 2014, 08:37:59 UTC |
42c461a | Timur Pocheptsov | 31 January 2014, 07:57:11 UTC | Found a macro depending on this function. | 31 January 2014, 07:57:11 UTC |
880fe6d | Danilo Piparo | 30 January 2014, 20:56:33 UTC | Fix the CMake build now the library names in the rootmaps are sane. In addition, the creation of the tmva and genvector dicts was simplified. | 31 January 2014, 05:39:20 UTC |
f490ac9 | Philippe Canal | 31 January 2014, 03:47:48 UTC | Revert "In TClass::New() first try the collection proxy, then ClassInfo." This reverts commit f27d31234ae8579f7c05e957bb92262e3f854650. as it seems to destroy the python test :( | 31 January 2014, 03:47:48 UTC |
7e94652 | Philippe Canal | 31 January 2014, 03:31:26 UTC | Handle case of interpreted class with emulated STL inside. If a class is interpreted and contains STL collection (embedded or as a pointer) we issue a warning and do *not* do the I/O. This was already the behavior in case of compiled classes. | 31 January 2014, 03:32:03 UTC |
f27d312 | Axel Naumann | 06 September 2013, 09:58:18 UTC | In TClass::New() first try the collection proxy, then ClassInfo. Additional note by Philippe: Even if we have a ClassInfo, the collection proxy might be emulated, if it is the case then it will use the object incorrectly (resulting in random behavior). This still leaves the possibility that the collection might be allocated as part of a containing interpreted object (i.e. class Event { std::vector<Track> fValues; }), this should lead to an inconsistency warning at TStreamerInfo::Build time. The proper long term solution is to have the CollectionProxy created and compiled at run-time and to have a proper solution for the case where the CollectionProxy can not be generated (because the content does not provide all the features needed for the CollectionProxy operation (for example missing defautl constructor)). | 31 January 2014, 03:32:03 UTC |
3057be2 | Wim Lavrijsen | 30 January 2014, 23:52:03 UTC | better workaround :/ | 30 January 2014, 23:52:03 UTC |
fc13cc1 | Wim Lavrijsen | 30 January 2014, 23:43:58 UTC | ROOT/meta workaround for ROOT-6020 | 30 January 2014, 23:43:58 UTC |
4cc4eda | Philippe Canal | 30 January 2014, 17:32:30 UTC | Add more details to fatal error | 30 January 2014, 17:32:56 UTC |