svn::talk 0.0.1 - KDevelop
Filter: muesli at gmail dot com 

973 commits found for project KDevelop.
Regular Commiters: rodda, mattr, treat, raggi, dymo, mlaurent, jpetso
Last update: 07:45, 01.09.2006

TimeRevisionAuthorRepliesMessageFriday, 01. September 2006
04:44579423treat5796* Temporary fix
01:32579400treat28672Be quiet!
01:31579399treat10235* Don't save window stuff to the project files. I don't think it makes sense, but perhaps it does. We'll reconsider when we get more advanced window stuff like perspectives. Currently a bug exists in that the window settings are not applied all of the time. It is hit or miss. Kudos to anyone who can spot the underlying bug...


TimeRevisionAuthorRepliesMessageThursday, 31. August 2006
23:53579388treat28463The initialization is in pretty good shape now. THe order of the objects initialized in KDevCore no longer matters.
19:39579307treat0* More work on the initialization of kdevcore objects. Introduce a couple of methods for the objects to handle loading/saving their settings in different situations.
12:10579206rodda0Missed a break
11:31579195rodda0Code completion works, with lots of bugs. Currently disabled - comment out the return in void CppCodeCompletion::cursorPositionChanged() to experiment. Fixed bug with DUContext::allDefinitions() - infinite recursion
09:56579166rodda0Build
09:50579163rodda0Beginnings of code completion support - extract data for the model from a specified context. Not even linked in yet
08:48579152rodda0Didn't end up going with a QReadWriteLock... majority of access was write
08:47579151rodda0Protect shared editor integrator data with a mutex.
07:35579143rodda0I don't wonder why the uses weren't showing to start with anymore - they weren't being built (on purpose) when processing from disk...
05:46579123rodda0Initial support for converting a duchain from its on-disk representation to editor ranges and cursors, and initial highlighting. Makes documents highlighted from just after opening. Seems to be a bug with highlighting of uses though...
04:55579109rodda0Fix editor loading signal bug... if no document was open at startup, the editor integrator's internal QObject would be created on an alternate thread, and thus the signals would be queued connections; however, there are no event loops in the other threads, so the signal would never arrive. Force creation of the internal object on the main thread and assert that it is on the main thread at signal connection time.
04:04579099rodda0The repository can be accessed from multiple threads, therefore protect with a mutex.
03:45579097rodda0Functions to retrieve contexts and uses at certain positions in the document.


TimeRevisionAuthorRepliesMessageWednesday, 30. August 2006
18:15578982rodda0Removed Q_UNUSED
18:14578981rodda0Added some debug which shows up the weird regression in that KTextEditor::Document::completed() does not call KDevEditorIntegratorPrivate::documentLoaded(), even though they are connected. A few other small cleanups
17:20578958rodda0Go back to 2 threads (that wasn't an issue before) Change debug output
17:06578950treat0Getting closer to kosher initialization. Suspend the weaver in construction and resume later after it has had a chance to see the project file settings. Reenable the progress bar.
16:48578943rodda0Assert fix Remove debug
16:38578940treat0* We aren't using these at the moment so quit wasting CPU cycles building/linking
16:34578938rodda0Oooh, much better... now that the types are resolved, we can just use the pointer as the hash key... right?
16:29578936rodda0Start to aggregate types Likely to need more work on performance
16:15578931rodda0Whoops, missed this commit Allow a range to be taken from the object, if the object is to be destroyed.
15:30578920rodda0Introduce type into declaration search in case of multiple matches (which can occurr with functions - there can be any number of definitions with the same identifier (but they must have different types) New test to make sure we're not creating a separate declaration when we encounter a class function (currently, we are... :() I needed a way to get the start token of the FunctionDefinitionAST so I abused the declaration stack - this probably needs a rethink
14:31578902rodda0Valgrind check
14:30578901rodda0Register all uses with their enclosing context; change the way they have highlighting applied (via context, not definition) Some debug statements from earlier - commented out for now (hopefully not needed) Some excessive paranoia added to preprocessjob - perhaps we can remove it later. Don't background parse everything if running under valgrind.
14:27578897rodda0Stability fixes More locking... not really sure if it's needed... will have to get a book on threads. /Disable/ close project action once a project is closed.
11:51578838rodda0Start work on the include system for parsing c++ files. Protected access to the background parser with a mutex. Enabled worker threads to request document addition; the request will block until it has been completed by the main thread. One problem: threadweaver complains about unresolved dependencies when I enable the dependency system. Needs to be debugged, but I've had enough of threads issues for today.
09:10578792rodda0Actually set the file manager in the project so it can be accessed Implement parseJobForDocument
05:08578758rodda0Crash fix - don't allow dangling pointers in parent ranges Took a while to track this as Valgrind went down a different path to when it was running outside of Valgrind.
04:49578753mattr0better handling of source files with output (such as .kcfgc files and .ui files) but it's not perfect
04:33578749rodda0Re-introduce enum types Allow null function argument types (should be improved when the include parsing lands) Fix crash in typedef function declarations
04:31578748mattr0provide a way to get a full path for a file name that is in the project's include path.
04:13578745rodda0Fix signal connection, remove double deletion
04:13578745rodda0Fix signal connection, remove double deletion
03:25578737rodda0Recognise variables as already being defined as well as declared
03:25578737rodda0Recognise variables as already being defined as well as declared
03:22578736rodda0Fix types generated with a list of init-declarators
03:22578736rodda0Fix types generated with a list of init-declarators
03:13578734rodda0Fix function return types by waiting for them to be completed before assignment to the function
03:13578734rodda0Fix function return types by waiting for them to be completed before assignment to the function
02:30578730mattr0compile
02:30578730mattr0compile


TimeRevisionAuthorRepliesMessageTuesday, 29. August 2006
22:37578683treat0Make all KDevCore objects implement the interface. Disable the progress bar in KDevBackgroundParser for now. Reimplement some stuff.
21:17578663treat0Add a common interface for KDevCore objects. This should help sync the startup.
20:05578629treat0Tired of installing these every time. We're not using them right now anyway.
18:53578611rodda0Fix function argument types by using the parameter declaration visit to assign argument types Improve debug output
18:40578605rodda0Add support for forward declarations of classes/unions/structs Need to rethink how we handle forward declarations a bit though...
17:35578588rodda0First attempt at mangled identifiers (likely quite poor)
17:13578581rodda0Add a link to the context in which the definition is found, and add a list of definitions to the context.
16:48578571rodda0Implementation of definition storage
16:43578569rodda0Add definition object to complement declaration
16:32578564rodda0Refactoring; much s/definition/declaration/ DefinitionUse -> Use Fixed an error in the type builder (still a few more to go, heh ;)
13:52578506rodda0Remove unused code
13:40578504rodda0Fix the ranges created by the editor integration Thanks to Roberto :)
13:30578500rodda0First in a series of patches to come from speaking with Roberto yesterday. Use the presence / absence of ParameterDeclarationClauseAST to determine if the current declaration is a function or not. Caused a bit of reorganisation in the builders + the class*definition classes
05:07578355rodda0Start working on hash functionality for types... needs work
04:07578344mattr0disable the file selector again. It's what is causing all the crashing when resizing kdevelop vertically. Now to figure out why.


TimeRevisionAuthorRepliesMessageMonday, 28. August 2006
19:59578269coolo1replaced several URLs with Url and URI with Uri - the patch became a bit larger than expected, so I couldn't have the time to check if it still compiles with the changes done in between
19:29578252treat0* Make way for a new 'mode' button in the settings dialog. This new button will allow the user to specify which file to save the changes to. When a project is opened, the user can thus still save the changes to the local kdeveloprc file instead of the project file...
14:39578149rodda0Very rudimentary support for array types - currently only works if an integer literal is present for the array dimension. We need an expression evaluator... hmm, kdevelop really will have a compiler at the end of this :)
13:48578135rodda0Fix creation of pointer and reference types
09:06578033rodda0Fix tests + errors revealled by them, mostly in the typesystem
07:00578017rodda0Already made this a class member
06:53578016rodda0Weaver improvements in the background parser Locking for the persistent hash
06:51578014rodda5A few important threadweaver fixes 1) Clear m_job in ThreadRunHelper after the job has finished, so that requestAbort doesn't use a dangling pointer when the helper is waiting for the next job. 2) Use isEmpty() rather than isIdle() in WeaverImpl, because the threads are considered active while they wait for more jobs and an infinite loop ensues. 3) Propagate abort requests to child jobs in a job collection (does this need some locking?) I'll commit these to threadweaver once I've discussed my other pending changes with Mirko... CCMAIL:mirko@hackerbuero.org


TimeRevisionAuthorRepliesMessageSunday, 27. August 2006
15:47577810rodda0Remove some old wrong fixmes
15:16577800rodda0Commit a new solution for determining if ExpressionOrDeclarationAST refers to an expression or a definition... check to see if all of the identifiers can be found; if so, it's an expression, otherwise it's a definition. Works ok :)
14:24577774rodda0Highlight functions + classes Remove accidental commit which disabled background parsing of code from the project
13:46577762rodda0Type system starting to work... Moved the const / volatile stuff back to the type... they belong there. Hopefully the member + function stuff can stay in the duchain. Start setting up regression tests.
09:23577641rodda0Move more information from the type system to the definition system
08:11577628rodda0Move const + volatile properties to the definition... I think that's where they belong (although, in the case of function arguments...?)
07:26577607rodda0Introduce repository for c++ types Rework integral types to only generate one type object
06:28577590rodda0Types occur so commonly that they should be shared pointers. Removing TypeInstance, because this really seems to be the job of the definition-use chain. I'll re-introduce class and function definitions shortly.
05:35577580mattr0reenable the context menu


TimeRevisionAuthorRepliesMessageSaturday, 26. August 2006
01:36577273mattr0revert the last commit. I'll need to do something different
01:22577270mattr0add an isBuildable() function to make it easy to tell if a project is buildable. this'll be used in the project manager
01:19577269mattr0remove the make environment function. we'll use KDevEnvironment instead
01:18577268mattr0documentation formatting
01:18577267mattr0make the guibuilder load again
01:18577266mattr0don't go searching for other importers after we've found one
00:31577255rodda0Adding skeleton for dumping types to debug output
00:18577253rodda0Moving the access policy to be a part of definitions, not the types


TimeRevisionAuthorRepliesMessageFriday, 25. August 2006
15:42577113rodda0Set the access policy... However, I plan to move some of this stuff to the definition-use chain...
15:39577112rodda0Ok, so it must be late... compile
15:35577110rodda0Track access policies... but it won't build?? I'm stumped...
15:14577103rodda0None of roberto's code remains
15:13577102rodda0Minor fix If any kdevelop devs would like to help, I'm going to be in need of: 1) mangle/demangle functions - preferably the same as the latest gcc 2) operator == implementations for all of the cpptypes.h classes
14:53577090rodda0More work - untested as yet Next step will be a symbol table...
14:23577084rodda0Progress Some complexities starting to show...
13:52577081rodda0Introducing TypeInstance, a class to hold the definition point of a type, its type, and its identifier.
13:23577068rodda0Starting to flesh out the new c++ type system. The builder still has a long way to go, but it got pretty quickly to this stage...
09:23576946rodda0Refactor manual invocations of the name compiler to a new function, identifierForName(). Convert the type compiler to TypeBuilder, and scrape out the logic (which was verbose and didn't use the visitor pattern)
08:42576933rodda0Moving Roberto's new type system (in its unfinished version) to the duchain area - the directory could do with renaming, but that can be later
08:25576929rodda0Slight cleanups Roberto's copyright didn't apply to any of these files - I wrote this code
08:17576927rodda0Step two towards stackable visitors: spilt the definition builder out from the context builder.
07:56576922rodda0Working towards stackable visitors: separate the use builder from the definition builder.
06:03576902scripty0SVN_SILENT made messages (.desktop file)