GitLab

Example on how to trigger this:
-have a CMake project with unit tests
-make clean
-try to launch one test that has an executable cleaned by the previous step

(cherry picked from commit 4a4592f3)

clang-format adds/removes "\" for long macros.
If this got added in code before our selected code
we will no longer find our selected code.
Add "\" as fuzzy character.

collect.kde.org was shut down recently, as substitute there is now
a category at store.kde.org

Related: T11415

BUG:377183
FIXED-IN:5.4.5

Ensures that in setups with system KF libs and partial custom build KF
libs in custom paths do not result in include paths making the wrong
headers found first

Passing in a plain int gets localized integer text rendering with
e.g. thousand separator (2.019), which is not what we want here.

By feedback of localization teams the original year span is used, and
not mapped into local calendar systems at least by the active teams,
so just passing in a plain string is fine for now.
Cmp. https://marc.info/?l=kde-i18n-doc&m=157238863619132&w=2

BUG:413390
FIXED-IN:5.4.4

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

Sometimes, for some reason we seem to get problem ranges from libclang that
start and end on the first character of the document. As a result, the final
range computed for the ClangProblem will wrongly start from the beginning of the
document and extend until the actual location of the problem. This commit
introduces a workaround which discards the problematic problem ranges.

Corrects 8c8c7b41

The previous implementation from a0fd2014 and 73fb3cd6 used a mutex
and would attempt to unlock it in the helper thread after locking
it in the main thread. This is UB and triggered an assert in libc++
on Mac (observed on OS X 10.9 and affecting an as yet unknown number of
newer OS versions).

In some projects - I recognized the issue browsing the kbanking part of #kmymoney -
aliased types may point to themselves. I assume an issue with forward-declared types,
but did not investigate this further.
The recursive call of declarationSizeInformation() may then lead to an infinite number
of function calls and consequently to a stack overflow.

The fix ends the recursion if the type is recurring.

Otherwise we may end up returning e.g. Cuda defines/includes for C++
and vice versa.

Unbreaks the basicdeclsandcontexts.cpp test

The cursor pointing to the struct within `typedef struct {...} foo;`
has an empty cursor spelling, but its type spelling is set to `foo`.
This is ideal to disambiguate between different structs. Without that,
we would end up associating uses of different cursors to the first
declaration with an empty name. Now that we set a proper non-empty
name, we can correctly disambiguate and associate the uses correctly.

Sadly, clang_getCompletionParent is broken too for this case - it
doesn't return the typedef name :( We now work-around this limitation
by parsing the USR of the container which _does_ contain that info.
Very hacky but at least this makes it work for both, uses and code
completion! Also, we can get rid of the previous hack which tried to
match the type too, which is probably way more costly than this.

BUG: 409041

This breaks code completion, I'll have to figure out a proper
fix for that too...

This reverts commit a6df33cf.

The cursor pointing to the struct within `typedef struct {...} foo;`
has an empty cursor spelling, but its type spelling is set to `foo`.
This is ideal to disambiguate between different structs. Without that,
we would end up associating uses of different cursors to the first
declaration with an empty name. Now that we set a proper non-empty
name, we can correctly disambiguate and associate the uses correctly.

BUG: 409041

Issue: When you select an inner node (method or member) in classes
panel, then type something in search bar that filters that node out,
Kdevelop crashes. It seems that when nodes get filtered, at some
point currentIndex of QTreeView points to deleted node, which causes
segfault in ClassModel::hasChildren. Manually invalidating
currentIndex before filtering seems to have fixed the issue.

Differential Revision: https://phabricator.kde.org/D22660
BUG: 406931

Make sure include paths containing symbolic links are canonicalized
- and thereby properly resolved - before wrapping them in a Path
object. This is necessary, as class Path is not properly handling
symbolic links.

BUG: 405221

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

CCBUG: 411371

From KTextEditor::TextHintInterface::registerTextHintProvider() docs:
"When the provider is about to be destroyed, make sure to call
unregisterTextHintProvider() to avoid a dangling pointer."
(new requirement with KF5)

The old code never unregistered in case the ContextBrowserPlugin was
deleted (like it happens when disabling the plugin in the settings in
a running KDevelop instance).

CCBUG: 411371