Summary:
In the new threading model clangd creates one thread per file to manage
the AST and one thread to process each of the incoming requests.
The number of actively running threads is bounded by the semaphore to
avoid overloading the system.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: klimek, mgorny, jkorous-apple, ioeric, hintonda, cfe-commits
Differential Revision: https://reviews.llvm.org/D42573
llvm-svn: 324356
Summary:
Some STL symbols are defined in inline namespaces. For example,
```
namespace std {
inline namespace __cxx11 {
typedef ... string;
}
}
```
Currently, this will be `std::__cxx11::string`; however, `std::string` is desired.
Inline namespaces are treated as transparent scopes. This
reflects the way they're most commonly used for lookup. Ideally we'd
include them, but at query time it's hard to find all the inline
namespaces to query: the preamble doesn't have a dedicated list.
Reviewers: sammccall, hokein
Reviewed By: sammccall
Subscribers: klimek, ilya-biryukov, jkorous-apple, cfe-commits
Differential Revision: https://reviews.llvm.org/D42796
llvm-svn: 324065
Summary:
Instead of passing Context explicitly around, we now have a thread-local
Context object `Context::current()` which is an implicit argument to
every function.
Most manipulation of this should use the WithContextValue helper, which
augments the current Context to add a single KV pair, and restores the
old context on destruction.
Advantages are:
- less boilerplate in functions that just propagate contexts
- reading most code doesn't require understanding context at all, and
using context as values in fewer places still
- fewer options to pass the "wrong" context when it changes within a
scope (e.g. when using Span)
- contexts pass through interfaces we can't modify, such as VFS
- propagating contexts across threads was slightly tricky (e.g.
copy vs move, no move-init in lambdas), and is now encapsulated in
the threadpool
Disadvantages are all the usual TLS stuff - hidden magic, and
potential for higher memory usage on threads that don't use the
context. (In practice, it's just one pointer)
Reviewers: ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D42517
llvm-svn: 323872
Summary:
For symbols defined inside macros:
* use expansion location, if the symbol is formed via macro concatenation.
* use spelling location, otherwise.
This will fix some symbols that have ill-format location (especial invalid filepath).
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: klimek, ilya-biryukov, jkorous-apple, cfe-commits
Differential Revision: https://reviews.llvm.org/D42575
llvm-svn: 323867
Summary:
Previously, we assume only old.cc includes "old.h", which would
introduce incorrect fixes for the cases where old.h also includes `#include "old.h"`
Although it should not be occurred in real projects, clang-move should handle this.
Old.h:
```
class Foo {};
```
after moving to a new old.h:
```
class Foo {};
```
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: cfe-commits, klimek
Differential Revision: https://reviews.llvm.org/D42639
llvm-svn: 323865
Summary:
We now provide an abstraction of Scheduler that abstracts threading
and resource management in ClangdServer.
No changes to behavior are intended with an exception of changed error
messages.
This patch is preliminary work to allow a revamped threading
implementation that will move the threading code out of CppFile.
Reviewers: sammccall, bkramer, jkorous-apple
Reviewed By: sammccall
Subscribers: hokein, mgorny, hintonda, ioeric, jkorous-apple, cfe-commits, klimek
Differential Revision: https://reviews.llvm.org/D42174
llvm-svn: 323851
Summary:
o Replace the existing clangd::URI with a wrapper of FileURI which also
carries a resolved file path.
o s/FileURI/URI/
o Get rid of the URI hack in vscode extension.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: klimek, ilya-biryukov, jkorous-apple, cfe-commits
Differential Revision: https://reviews.llvm.org/D42419
llvm-svn: 323660
Summary:
This is probably the right behavior for distributed tracers, and makes unpaired
begin-end events impossible without requiring Spans to be bound to a thread.
The API is conceptually clean but syntactically awkward. As discussed offline,
this is basically a naming problem and will go away if (when) we use TLS to
store the current context.
The apparently-unrelated change to onScopeExit are because its move semantics
broken if Func is POD-like since r322838. This is true of function pointers,
and the lambda I use here that captures two pointers only.
I've raised this issue on llvm-dev and will revert this part if we fix it in
some other way.
Reviewers: ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D42499
llvm-svn: 323511
Summary:
This adds checks that our diagnostics emit correct ranges in a bunch of cases,
as promised in D41118.
The diagnostics-preamble test is also converted and extended to be a little more
precise.
diagnostics.test stays around as the smoke test for this feature.
Reviewers: ilya-biryukov
Subscribers: klimek, mgorny, cfe-commits
Differential Revision: https://reviews.llvm.org/D41454
llvm-svn: 323448
Summary:
* truncate symbols from static/dynamic index to the limited number
(which would save lots of cost in constructing the merged symbols).
* add an CLI option allowing to limit the number of returned completion results.
(default to 100)
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: klimek, ilya-biryukov, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D42484
llvm-svn: 323408
Summary:
* For qualified completion (foo::a^)
* unresolved qualifier - use global namespace ("::")
* resolved qualifier - use all accessible namespaces inside the resolved qualifier.
* For unqualified completion (vec^), use scopes that are accessible from the
scope from which code completion occurs.
Reviewers: sammccall, ilya-biryukov
Reviewed By: sammccall
Subscribers: jkorous-apple, ioeric, klimek, cfe-commits
Differential Revision: https://reviews.llvm.org/D42073
llvm-svn: 323189
Summary: I will replace the existing URI struct in Protocol.h with the new URI and rename FileURI to URI in a followup patch.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: jkorous-apple, klimek, mgorny, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D41946
llvm-svn: 323101
Global scope is "" (was "")
Top-level namespace scope is "ns::" (was "ns")
Nested namespace scope is "ns::ns::" (was "ns::ns")
This composes more naturally:
- qname = scope + name
- full scope = resolved scope + unresolved scope (D42073 was the trigger)
It removes a wart from the old way: "foo::" has one more separator than "".
Another alternative that has these properties is "::ns", but that lacks
the property that both the scope and the name are substrings of the
qname as produced by clang.
This is re-landing r322996 which didn't build.
llvm-svn: 323000
Global scope is "" (was "")
Top-level namespace scope is "ns::" (was "ns")
Nested namespace scope is "ns::ns::" (was "ns::ns")
This composes more naturally:
- qname = scope + name
- full scope = resolved scope + unresolved scope (D42073 was the trigger)
It removes a wart from the old way: "foo::" has one more separator than "".
Another alternative that has these properties is "::ns", but that lacks
the property that both the scope and the name are substrings of the
qname as produced by clang.
llvm-svn: 322996
Summary:
- we match on USR, and do a field-by-field merge if both have results
- scoring is post-merge, with both sets of information available
(for now, sema priority is used if available, static score for index results)
- limit is applied to the complete result set (previously index ignored limit)
- CompletionItem is only produces for the returned results
- If the user doesn't type a scope, we send the global scope for completion
(we can improve this after D42073)
Reviewers: ioeric
Subscribers: klimek, ilya-biryukov, mgrang, cfe-commits
Differential Revision: https://reviews.llvm.org/D42181
llvm-svn: 322945
Summary:
This makes performance slower but more predictable (it always processes
every symbol). We need to find ways to make this fast, possibly by precomputing
short queries or capping the number of scored results. But our current approach
is too naive.
It also no longer returns results in a "good" order. In fact it's pathological:
the top N results are ranked from worst to best. Indexes aren't responsible for
ranking and MergedIndex can't do a good job, so I'm pleased that this will make
any hidden assumptions we have more noticeable :-)
Reviewers: hokein
Subscribers: klimek, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D42060
llvm-svn: 322821
Summary:
This test dominates our unit test runtime, and the change speeds it up by 10x.
We lose coverage of some combinations of flags, but I'm not sure that's finding
many bugs.
3300 -> 300ms on my machine (3800 -> 800ms for the whole of CompletionTest).
Reviewers: ilya-biryukov
Subscribers: klimek, cfe-commits
Differential Revision: https://reviews.llvm.org/D42063
llvm-svn: 322547
Summary:
We now hide the static/dynamic split from the code completion, behind a
new implementation of the SymbolIndex interface. This will reduce the
complexity of the sema/index merging that needs to be done by
CodeComplete, at a fairly small cost in flexibility.
Reviewers: hokein
Subscribers: klimek, mgorny, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D42049
llvm-svn: 322480
Summary:
To stay fast, it avoids deserializing anything outside the current file, by
disabling the LoadExternal code completion option added in r322377, when the
index is enabled.
Reviewers: hokein
Subscribers: klimek, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D41996
llvm-svn: 322387
Summary:
For some cases, GoToDefinition will navigate to the forward class
declaration, we should always navigate to the class definition.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: klimek, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D41661
llvm-svn: 322370
Summary:
o We only collect symbols in namespace or translation unit scopes.
o Add an option to only collect symbols in included headers.
Reviewers: hokein, ilya-biryukov
Reviewed By: hokein, ilya-biryukov
Subscribers: klimek, cfe-commits
Differential Revision: https://reviews.llvm.org/D41823
llvm-svn: 322193
Summary:
Use the YAML-format symbols (generated by the global-symbol-builder tool) to
do the global code completion.
It is **experimental** only , but it allows us to experience global code
completion on a relatively small project.
Tested with LLVM project.
Reviewers: sammccall, ioeric
Reviewed By: sammccall, ioeric
Subscribers: klimek, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D41668
llvm-svn: 322191
Summary:
This patch removes hidden items from code completion.
Items can be hidden, e.g., by other items in the child scopes.
This patch addresses a particular problem of a duplicate completion
item for the class in the following example:
struct Adapter { void method(); };
void Adapter::method() {
Adapter^
}
We should probably investigate if there are other duplicates in
completion and remove them, possibly adding assertions that it never
happens.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: cfe-commits, klimek
Differential Revision: https://reviews.llvm.org/D41901
llvm-svn: 322185
Summary:
This enables more execution modes like standalone and Mapreduce-style execution.
See also D41729
Reviewers: hokein, sammccall
Subscribers: klimek, ilya-biryukov, cfe-commits
Differential Revision: https://reviews.llvm.org/D41730
llvm-svn: 322084
Summary:
We currently only collect external-linkage symbols in the collector,
which results in missing some typical symbols (like no-linkage type alias symbols).
This patch relaxes the constraint a bit to allow collecting more symbols.
Reviewers: ilya-biryukov
Reviewed By: ilya-biryukov
Subscribers: klimek, cfe-commits
Differential Revision: https://reviews.llvm.org/D41759
llvm-svn: 322067
Summary:
This improves a few things:
- the insert -> freeze -> read sequence is now enforced/communicated by the
type system
- SymbolSlab::const_iterator iterates over symbols, not over id-symbol pairs
- we avoid permanently storing a second copy of the IDs, and the
string map's hashtable
The slab size is now down to 21.8MB for the LLVM project.
Of this only 2.7MB is strings, the rest is #symbols * `sizeof(Symbol)`.
`sizeof(Symbol)` is currently 96, which seems too big - I think
SymbolInfo isn't efficiently packed. That's a topic for another patch!
Also added simple API to see the memory usage/#symbols of a slab, since
it seems likely we will continue to care about this.
Reviewers: ilya-biryukov
Subscribers: klimek, mgrang, cfe-commits
Differential Revision: https://reviews.llvm.org/D41506
llvm-svn: 321412