The Python language server seems to dislike files that are not part of a
workspace resulting in very long lookup times for specific symbols.
Fixes: QTCREATORBUG-26230
Fixes: QTCREATORBUG-24704
Fixes: QTCREATORBUG-24140
Change-Id: Iceb7a2b3d57aea6554225a74587f619e530e10c9
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Unknown compiler flags are an unsuppressable error in clang.
But we do not want to maintain a blacklist of unsupported GCC and MSVC
options or a whitelist of supported clang options, as both would
constantly go out of date.
As clangd seems to work fine despite the error message, we simply filter
out this type of diagnostic.
Fixes: QTCREATORBUG-27113
Change-Id: Ib32601831eded60daf80eb0ca5cf01bbd71493fa
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: David Schulz <david.schulz@qt.io>
If the client does not react within 20 seconds on the shutdown or exit
message delete the client.
Task-number: QTCREATORBUG-27064
Change-Id: I9680242e471989ea820c4e28d28c500c27a20bb7
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
This way, a warning about missing switch cases will be accompanied by a
matching code action, which allows the user to fix it right away by
clicking on the lightbulb.
Requires clangd >= 15; see https://reviews.llvm.org/D118976.
Change-Id: I11e82264c41e4154f979d28a5e44e72c8158595b
Reviewed-by: David Schulz <david.schulz@qt.io>
Add a private function that logs and sends a basemessage without
checking the current server state and use it for server init,
sendContent and exit.
Change-Id: I6532dfb36ef5bf4d036bb71dbfab4ab29497fc9a
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
If a document is closed between removing a closing client from the
tracked clients in the LanguageClientManager and the actual destruction
of that client we might derefernce an already desctructed document in
resetAssistProvider.
Connect the document closed signal in the client instead of the manager
so Client::closeDocument gets called even if it is not tracked in the
manager anymore.
Fixes: QTCREATORBUG-26534
Change-Id: I7187730600eeb1df54355065eb9cb703a605139e
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
Used there to prevent requesting document symbols from server that do
not support these requests.
Change-Id: Icca64c41415ba7098c769854823431fb05e5ce1d
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
... if the client implementation requests it.
The server is not necessarily aware of our indentation style, so we
might have to apply it to the newly inserted code.
Change-Id: I43518575c7124568da42be3b04a28d7f352f6dc2
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: David Schulz <david.schulz@qt.io>
Introduce an assist processor that merges our built-in quickfixes with
refactoring actions from clangd ("tweaks").
For now, we make it clear which ones are coming from clangd, and we do
not filter duplicate functionality. In the future, we might want to
disable redundant built-in actions if clangd is enabled for the
respective file.
Change-Id: I04842132798c8635dfddf8cfc98cc7a6313fac09
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: David Schulz <david.schulz@qt.io>
The request for highlights of the symbol under the cursor takes a
considerable amount of time on some servers (at least python). Postpone
those requests a bit so that more important requests can be handled
before.
Change-Id: I35776076a4ee9c0c9b7646e960d3d965506154d3
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
If the cursor is right before the "." in a member access expression,
clangd interprets it as belonging to the member instead of the base
expression, which leads to unexpected behavior.
Work around this by sending a cursor position one to the left of the
real one to clangd in such cases.
Change-Id: I429ee9189760ccb02d231acfcb94ab6cfde3cd8d
Reviewed-by: <github-actions-qt-creator@cristianadam.eu>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: David Schulz <david.schulz@qt.io>
Restructure the message handling in the client, so if we get a message
with an id always send a response to the server either with the result
or with an error. Also make sure to not send responses to unreachable
server.
Change-Id: Ie74128069c1678af60871896d5dce45c08e71b05
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
And make use of that in ClangdClient.
When I get a slowdown while typing, it is usually accompanied by the
clangd message "Request cancelled because the document was modified"
occurring in the message window, often many times in a row. I'd like to
find out whether writing to the message window itself is a contributing
factor to the slowdown.
Change-Id: Iff7c459af0aed27d22366b9aade573f51eb5dbc7
Reviewed-by: David Schulz <david.schulz@qt.io>
Since we use the AST in several contexts, it can easily happen that we
need to consult it more than once for the same document revision.
Therefore, it makes sense to cache the AST to prevent redundant requests
for potentially large amounts of data.
Change-Id: I33100e0052ee21bb46e91848d3d2e8a0a96bb860
Reviewed-by: David Schulz <david.schulz@qt.io>
Removes noise from the interface. Using specialized clangd processors
that redirect the proposals for testing purposes is an overall cleaner
solution.
Change-Id: Ia4f3d4720765f0533c5530a2a5c7db3cc971e0be
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
Callers can have legitimate reasons to use it in a non-const manner.
Change-Id: Id91a4708dd95845661b291ce7cc9ee1581bdade8
Reviewed-by: David Schulz <david.schulz@qt.io>
... when doing a decl/def switch via clangd.
The delay is fine for e.g. populating the outline, but shouldn't be used
when dealing with explicit user requests.
Change-Id: I0350ed6daf8220ec3b702a3876fbf0f726da8a67
Reviewed-by: David Schulz <david.schulz@qt.io>
Due to the potential cost of the re-highlighting procedure, we'd like to
delay the semantic tokens update request to a time where the new token
information won't get invalidated right away due to further document
changes. While the document updates in principle have such a delay
already, it is often "sabotaged", for instance due to auto-completion
requests which require the server to know the current document contents
right away.
Therefore, we request new semantic tokens along with a document update
only if it's the "regular" delayed one, and otherwise add an additional
delay.
Change-Id: I074647d1cdfdbcc6aa5fc5ec2a2d5ae6ccd493ba
Reviewed-by: David Schulz <david.schulz@qt.io>
In particular, calling keys() and then take() for all the keys is
wasteful (and obscures what's going on).
Change-Id: Ic66803cf3579a39c23c32f1fc65e2c9399dcc583
Reviewed-by: David Schulz <david.schulz@qt.io>
The pure LSP-based implementation is too inflexible with regards to
activation characters.
Change-Id: I386a60b80f59e1b48254d55c2a47f1a1ffe1f97f
Reviewed-by: David Schulz <david.schulz@qt.io>
This is needed for e.g. ClangdClient, which may need to update even if
the highlighting delta from the server is empty.
Task-number: QTCREATORBUG-26183
Change-Id: I38398c9563fa7a6a1906c5cb57fd1cbcd444cbd9
Reviewed-by: David Schulz <david.schulz@qt.io>
Opening a document in the language client while the server is still
initializing will postpone the opening after the server is fully
initialized. So if a document was scheduled for opening by
ClangModelManagerSupport::onEditorOpened while clangd is still
initializing, skip the superfluous opening in the ClangdClient
initialize callback. Clangd seems to have issues if files are opened
twice, resulting in strange diagnostics after editing the file. This is
reproducible for me when starting Qt Creator with a session that
contains open files.
Change-Id: I200d5c8afb685403f0435e0553f5a475f75e8ea2
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
.. after receiving diagnostics since the code actions are already
inlined into the diagnostics.
Change-Id: I11ed1270344ff5119dd111503d173eef2f340c1d
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
Note that we lose the highlighting for virtual function calls. We need
to amend clangd to add the respective information to its semantic tokens
message.
Also, Qt properties are no longer highlighted as class members. We'll
investigate how to best restore this feature.
Change-Id: I403712aada3d7a8e1c7b7c1277f43f7f64f8450b
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: David Schulz <david.schulz@qt.io>
Note that we used to encode the information about symbol visibility and
static-ness in the icons, which we can't do anymore, because clangd does
not provide this information.
On the upside, this change likely fixes a ton of bugs, as our own outline
was rather "quirky".
Change-Id: I099f11ec4e3c6f52cd461fb43080bbdde3bed5e5
Reviewed-by: David Schulz <david.schulz@qt.io>
Using the document revision causes issues for some servers.
Task-number: QTCREATORBUG-25766
Change-Id: Ic858e19c6fe39e57c9d3124913887aafee0a3cd0
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Server name and version are potentially interesting information.
Change-Id: I4fc9a2556d7fd8285fe0024cf0f83baf95112e65
Reviewed-by: David Schulz <david.schulz@qt.io>
... and make use of it in the clangd client to be able tell when
background indexing has finished.
Change-Id: I0f3c6f9646fd66ababd08c12b2f347da5f1a3729
Reviewed-by: David Schulz <david.schulz@qt.io>
... and make use of it in the clangd client.
We want the progress bar for background indexing to have a better title
than the plain (and non-translated) "indexing" that the clangd server
sends.
Change-Id: Ib75eac370e7c22f0c5bd477f4a4c423283b27e1f
Reviewed-by: David Schulz <david.schulz@qt.io>
If the user has enabled clangd (default is off), we start up one instance
per project when it is opened/changed (including build config switches),
and trigger background indexing.
So far, the index is used to provide results for locators and "Find
Usages".
Per-document functionality such as semantic highlighting and completion
is still provided by libclang.
Change-Id: I12532fca1b9c6278baab560e7238cba6189cde9f
Reviewed-by: David Schulz <david.schulz@qt.io>
Only assign documents to a client if the client was started by a
setting. Programatically started clients should handle assigned
documents manually.
Change-Id: I192c4c011a48e0406f267d3cab3c860f740ddad7
Reviewed-by: Christian Kandeler <christian.kandeler@qt.io>
We can have multiple widgets per document, so we can have multiple
different cursors for the same document. Tracking document highlights
per widget prevents cancelling a potentially valid request for another
text cursor of another widget.
Change-Id: I800d9be5b44b327351252ab90305b8882efa87bb
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
