Summary: Depends on D80198.
This patch implies ASTs might be built with stale preambles without
blocking for a fresh one. It also drops any guarantees on every preamble
version being built. In case of multiple preamble build requests, in
addition to being debounced.
Any preamble requested with a WantDiags::Yes will always be built, this
is ensured by blocking enqueueing of any subsequent reqest.
AST worker will still block for initial preamble to reduce duplicate
work.
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, jfb, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D80293
Summary:
This is achieved by calculating newly added includes and implicitly
parsing them as if they were part of the main file.
This also gets rid of the need for consistent preamble reads.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, mgrang, arphaman, jfb, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D77392
Summary:
We want to be sure they don't cause AST rebuilds or evict items from the cache.
D77847 is going to start sending spurious no-op changes (in case the preamble
was invalidated), this is cheap enough but we shouldn't regress that in future.
Reviewers: kadircet
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, jfb, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D78048
Summary:
This assertion was bad. It will show up once we start running preamble
thread async. Think about the following case:
- Update 1
builds a preamble, and an AST. Caches the AST.
- Update 2
Invalidates the cache, preamble hasn't changed.
- Update 3
Invalidates the cache, preamble hasn't changed
- Read
builds AST using preamble v1, and caches it.
preamble for v2 gets build, cache isn't invalidated since preamble is same.
generateDiags tries to reuse cached AST but latest version is 3 not 2, so
assertion fails.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D77664
Summary:
FileInputs are only written by ASTWorker thread, therefore it is safe
to read them without the lock inside that thread. It can still be read by other
threads through ASTWorker::getCurrentCompileCommand though.
This patch also gets rid of the smart pointer wrapping FileInputs as there is
never mutliple owners.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D77309
Summary:
This is another step for out-of-order preamble builds. To keep the
diagnostic behavior same, we only build ASTs either with "usable" preambles,
the ones that are fully applicable to a given ParseInput, or after building a
new preamble. Which is the same behaviour as what we do today. ASTs
built in the latter is called golden ASTs.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76725
Summary:
TUStatus api had a single thread in mind. This introudces a section
action to represent state of the preamble thread. In the file status extension,
we keep old behavior almost the same. We only prepend current task with a
`parsing includes` if preamble thread is working. We omit the idle thread in the
output unless both threads are idle.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76304
Summary:
First step to enable deferred preamble builds. Not intending to land it
alone, will have follow-ups that will implement full deferred build
functionality and will land after all of them are ready.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76125
Summary:
This helps us prevent races when scheduler (or any other thread) tries
to read a request while it's still running.
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D75927
Summary:
This ties to an LSP feature (diagnostic versioning) but really a lot
of the value is in being able to log what's happening with file versions
and queues more descriptively and clearly.
As such it's fairly invasive, for a logging patch :-\
Key decisions:
- at the LSP layer, we don't reqire the client to provide versions (LSP
makes it mandatory but we never enforced it). If not provided,
versions start at 0 and increment. DraftStore handles this.
- don't propagate magically using contexts, but rather manually:
addDocument -> ParseInputs -> (ParsedAST, Preamble, various callbacks)
Context-propagation would hide the versions from ClangdServer, which
would make producing good log messages hard
- within ClangdServer, treat versions as opaque and unordered.
std::string is a convenient type for this, and allows richer versions
for embedders. They're "mandatory" but "null" is a reasonable default.
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D75582
Summary:
Otherwise they can force us to build lots of snapshots that we don't need.
Particularly, try to do this for operations that are frequently
generated by editors without explicit user interaction, and where
editing the file makes the result less useful. (Code action
enumeration is a good example).
https://github.com/clangd/clangd/issues/298
This doesn't return the "right" LSP error code (ContentModified) to the client,
we need to teach the cancellation API to distinguish between different causes.
Reviewers: kadircet
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, jfb, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D75602
The goal of this patch is to maximize CPU utilization on multi-socket or high core count systems, so that parallel computations such as LLD/ThinLTO can use all hardware threads in the system. Before this patch, on Windows, a maximum of 64 hardware threads could be used at most, in some cases dispatched only on one CPU socket.
== Background ==
Windows doesn't have a flat cpu_set_t like Linux. Instead, it projects hardware CPUs (or NUMA nodes) to applications through a concept of "processor groups". A "processor" is the smallest unit of execution on a CPU, that is, an hyper-thread if SMT is active; a core otherwise. There's a limit of 32-bit processors on older 32-bit versions of Windows, which later was raised to 64-processors with 64-bit versions of Windows. This limit comes from the affinity mask, which historically is represented by the sizeof(void*). Consequently, the concept of "processor groups" was introduced for dealing with systems with more than 64 hyper-threads.
By default, the Windows OS assigns only one "processor group" to each starting application, in a round-robin manner. If the application wants to use more processors, it needs to programmatically enable it, by assigning threads to other "processor groups". This also means that affinity cannot cross "processor group" boundaries; one can only specify a "preferred" group on start-up, but the application is free to allocate more groups if it wants to.
This creates a peculiar situation, where newer CPUs like the AMD EPYC 7702P (64-cores, 128-hyperthreads) are projected by the OS as two (2) "processor groups". This means that by default, an application can only use half of the cores. This situation could only get worse in the years to come, as dies with more cores will appear on the market.
== The problem ==
The heavyweight_hardware_concurrency() API was introduced so that only *one hardware thread per core* was used. Once that API returns, that original intention is lost, only the number of threads is retained. Consider a situation, on Windows, where the system has 2 CPU sockets, 18 cores each, each core having 2 hyper-threads, for a total of 72 hyper-threads. Both heavyweight_hardware_concurrency() and hardware_concurrency() currently return 36, because on Windows they are simply wrappers over std:🧵:hardware_concurrency() -- which can only return processors from the current "processor group".
== The changes in this patch ==
To solve this situation, we capture (and retain) the initial intention until the point of usage, through a new ThreadPoolStrategy class. The number of threads to use is deferred as late as possible, until the moment where the std::threads are created (ThreadPool in the case of ThinLTO).
When using hardware_concurrency(), setting ThreadCount to 0 now means to use all the possible hardware CPU (SMT) threads. Providing a ThreadCount above to the maximum number of threads will have no effect, the maximum will be used instead.
The heavyweight_hardware_concurrency() is similar to hardware_concurrency(), except that only one thread per hardware *core* will be used.
When LLVM_ENABLE_THREADS is OFF, the threading APIs will always return 1, to ensure any caller loops will be exercised at least once.
Differential Revision: https://reviews.llvm.org/D71775
Summary:
- This option forces a preamble rebuild to handle the odd case
of a missing header file being added
Reviewers: sammccall
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, jfb, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D73916
Summary:
Currently we delay AST rebuilds by 500ms after each edit, to wait for
further edits. This is a win if a rebuild takes 5s, and a loss if it
takes 50ms.
This patch sets debouncepolicy = clamp(min, ratio * rebuild_time, max).
However it sets min = max = 500ms so there's no policy change or actual
customizability - will do that in a separate patch.
See https://github.com/clangd/clangd/issues/275
Reviewers: hokein
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D73873
This is how it should've been and brings it more in line with
std::string_view. There should be no functional change here.
This is mostly mechanical from a custom clang-tidy check, with a lot of
manual fixups. It uncovers a lot of minor inefficiencies.
This doesn't actually modify StringRef yet, I'll do that in a follow-up.
Summary:
This is the initial version. The cross-file rename is purely based on
the index.
It is hidden under a command-line flag, and only available for a small set
of symbols.
Reviewers: ilya-biryukov, sammccall
Subscribers: merge_guards_bot, MaskRay, jkorous, arphaman, kadircet, usaxena95, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D69263
This much better reflects what is (now) in this header.
Maybe a rename to ParsedTU would be an improvement, but that's a much
more invasive change and life is too short.
ClangdUnit is dead, long live ClangdUnitTests!
llvm-svn: 370862
Summary:
Those errors are exposed at the first character of a file,
for a lack of a better place.
Previously, all errors were stored inside the AST and report
accordingly. However, errors in command-line argument parsing could
result in failure to produce the AST, so we need an alternative ways to
report those errors.
We take the following approach in this patch:
- buildCompilerInvocation() now requires an explicit DiagnosticConsumer.
- TUScheduler and TestTU now collect the diagnostics produced when
parsing command line arguments.
If pasing of the AST failed, diagnostics are reported via a new
ParsingCallbacks::onFailedAST method.
If parsing of the AST succeeded, any errors produced during
command-line parsing are stored alongside the AST inside the
ParsedAST instance and reported as previously by calling the
ParsingCallbacks::onMainAST method;
- The client code that uses ClangdServer's DiagnosticConsumer
does not need to change, it will receive new diagnostics in the
onDiagnosticsReady() callback
Errors produced when parsing command-line arguments are collected using
the same StoreDiags class that is used to collect all other errors. They
are recognized by their location being invalid. IIUC, the location is
invalid as there is no source manager at this point, it is created at a
later stage.
Although technically we might also get diagnostics that mention the
command-line arguments FileID with after the source manager was created
(and they have valid source locations), we choose to not handle those
and they are dropped as not coming from the main file. AFAICT, those
diagnostics should always be notes, therefore it's safe to drop them
without loosing too much information.
Reviewers: kadircet
Reviewed By: kadircet
Subscribers: nridge, javed.absar, MaskRay, jkorous, arphaman, cfe-commits, gribozavr
Tags: #clang
Differential Revision: https://reviews.llvm.org/D66759
llvm-svn: 370177
Now that we've moved to C++14, we no longer need the llvm::make_unique
implementation from STLExtras.h. This patch is a mechanical replacement
of (hopefully) all the llvm::make_unique instances across the monorepo.
Differential revision: https://reviews.llvm.org/D66259
llvm-svn: 368944
Summary:
clangd supports a -j option to limit the amount of threads to use for parsing
TUs. However, when using -background-index (the default in later versions of
clangd), the parallelism used by clangd defaults to the hardware_parallelisn,
i.e. number of physical cores.
On shared hardware environments, with large projects, this can significantly
affect performance with no way to tune it down.
This change makes the -j parameter apply equally to parsing and background
index. It's not perfect, because the total number of threads is 2x the -j value,
which may still be unexpected. But at least this change allows users to prevent
clangd using all CPU cores.
Reviewers: kadircet, sammccall
Reviewed By: sammccall
Subscribers: javed.absar, jfb, sammccall, ilya-biryukov, MaskRay, jkorous, arphaman, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D66031
llvm-svn: 368498
Summary:
By exposing a callback that can guard code publishing results of
'onMainAST' callback in the same manner we guard diagnostics.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: javed.absar, MaskRay, jkorous, arphaman, kadircet, hokein, jvikstrom, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D64985
llvm-svn: 366577
Summary:
In practice, opening Foo.h will still often result in Foo.cpp making the
second index build instead of the first, as the rebuild policy doesn't
know to wait.
Reviewers: kadircet
Subscribers: ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D64575
llvm-svn: 365888
Summary:
This makes addDocument non-blocking and would also allow code completion
(in fallback mode) to run when worker waits for the compile command.
Reviewers: sammccall, ilya-biryukov
Reviewed By: ilya-biryukov
Subscribers: javed.absar, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D60607
llvm-svn: 358400
Summary:
When calling TUScehduler::runWithPreamble (e.g. in code compleiton), allow
entering a fallback mode when compile command or preamble is not ready, instead of
waiting. This allows clangd to perform naive code completion e.g. using identifiers
in the current file or symbols in the index.
This patch simply returns empty result for code completion in fallback mode. Identifier-based
plus more advanced index-based completion will be added in followup patches.
Reviewers: ilya-biryukov, sammccall
Reviewed By: sammccall
Subscribers: sammccall, javed.absar, MaskRay, jkorous, arphaman, kadircet, jdoerfert, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D59811
llvm-svn: 357916
Summary:
This enables include insertion by adding canonical includes into
preambledata.
Reviewers: ioeric, ilya-biryukov
Subscribers: javed.absar, MaskRay, jkorous, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D57508
llvm-svn: 353054
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
The new guideline is to qualify with 'llvm::' explicitly both in
'.h' and '.cpp' files. This simplifies moving the code between
header and source files and is easier to keep consistent.
llvm-svn: 350531
Summary:
Previously, removeDoc followed by an addDoc to TUScheduler resulted in
racy diagnostic responses, i.e. the old dianostics could be delivered
to the client after the new ones by TUScheduler.
To workaround this, we tracked a version number in ClangdServer and
discarded stale diagnostics. After this commit, the TUScheduler will
stop delivering diagnostics for removed files and the workaround in
ClangdServer is not required anymore.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, jfb, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D54829
llvm-svn: 347468
Summary:
- Reads are never executed if canceled before ready-to run.
In practice, we finalize cancelled reads eagerly and out-of-order.
- Cancelled reads don't prevent prior updates from being elided, as they don't
actually depend on the result of the update.
- Updates are downgraded from WantDiagnostics::Yes to WantDiagnostics::Auto when
cancelled, which allows them to be elided when all dependent reads are
cancelled and there are subsequent writes. (e.g. when the queue is backed up
with cancelled requests).
The queue operations aren't optimal (we scan the whole queue for cancelled
tasks every time the scheduler runs, and check cancellation twice in the end).
However I believe these costs are still trivial in practice (compared to any
AST operation) and the logic can be cleanly separated from the rest of the
scheduler.
Reviewers: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D54746
llvm-svn: 347450
Summary:
To enable this, TUScheduler has to provide a way to run async tasks without
needing a preamble or AST!
Reviewers: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, jfb, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D53644
llvm-svn: 345268
Standardize on the most common namespace setup in our *.cpp files:
using namespace llvm;
namespace clang {
namespace clangd {
void foo(StringRef) { ... }
And remove redundant llvm:: qualifiers. (Except for cases like
make_unique where this causes problems with std:: and ADL).
This choice is pretty arbitrary, but some broad consistency is nice.
This is going to conflict with everything. Sorry :-/
Squash the other configurations:
A)
using namespace llvm;
using namespace clang;
using namespace clangd;
void clangd::foo(StringRef);
This is in some of the older files. (It prevents accidentally defining a
new function instead of one in the header file, for what that's worth).
B)
namespace clang {
namespace clangd {
void foo(llvm::StringRef) { ... }
This is fine, but in practice the using directive often gets added over time.
C)
namespace clang {
namespace clangd {
using namespace llvm; // inside the namespace
This was pretty common, but is a bit misleading: name lookup preferrs
clang::clangd::foo > clang::foo > llvm:: foo (no matter where the using
directive is).
llvm-svn: 344850
Summary:
Most of its functionality is moved into ClangdLSPServer.
The decoupling between JSONRPCDispatcher, ProtocolCallbacks, ClangdLSPServer
was never real, and only served to obfuscate.
Some previous implicit/magic stuff is now explicit:
- the return type of LSP method calls are now in the signature
- no more reply() that gets the ID using global context magic
- arg tracing no longer relies on RequestArgs::stash context magic either
This is mostly refactoring, but some deliberate fixes while here:
- LSP method params are now by const reference
- notifications and calls are now distinct namespaces.
(some tests had protocol errors and needed updating)
- we now reply to calls we failed to decode
- outgoing calls use distinct IDs
A few error codes and message IDs changed in unimportant ways (see tests).
Reviewers: ioeric
Subscribers: mgorny, ilya-biryukov, javed.absar, MaskRay, jkorous, arphaman, jfb, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D53387
llvm-svn: 344737
Summary:
- DynamicIndex doesn't implement ParsingCallbacks, to make its role clearer.
ParsingCallbacks is a separate object owned by the receiving TUScheduler.
(I tried to get rid of the "index-like-object that doesn't implement index"
but it was too messy).
- Clarified(?) docs around DynamicIndex - fewer details up front, more details
inside.
- Exposed dynamic index from ClangdServer for memory monitoring and more
direct testing of its contents (actual tests not added here, wanted to get
this out for review)
- Removed a redundant and sligthly confusing filename param in a callback
Reviewers: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D51221
llvm-svn: 341325
Summary:
After code completion inserts a header, running signature help using the old
preamble will usually fail. So we add support for consistent preamble reads.
Reviewers: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, kadircet, cfe-commits
Differential Revision: https://reviews.llvm.org/D51438
llvm-svn: 341076
Summary:
Will be used for updating the dynamic index on updates to the open files.
Currently we collect only information coming from the preamble
AST. This has a bunch of limitations:
- Dynamic index misses important information from the body of the
file, e.g. locations of definitions.
- XRefs cannot be collected at all, since we can only obtain full
information for the current file (preamble is parsed with skipped
function bodies, therefore not reliable).
This patch only adds the new callback, actually updates to the index
will be done in a follow-up patch.
Reviewers: hokein
Reviewed By: hokein
Subscribers: kadircet, javed.absar, ioeric, MaskRay, jkorous, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D50847
llvm-svn: 340401
Summary:
Fix an inconsistent behavior of using `LastBuiltPreamble`/`NewPreamble`
in TUScheduler (see the test for details), AST should always use
NewPreamble. This patch makes LastBuiltPreamble always point to
NewPreamble.
Preamble rarely fails to build, even there are errors in headers, so we
assume it would not cause performace issue for code completion.
Reviewers: ilya-biryukov
Reviewed By: ilya-biryukov
Subscribers: javed.absar, ioeric, MaskRay, jkorous, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D50695
llvm-svn: 340001
Summary:
This allows implementations like different symbol indexes to know what
the current active file is. For example, some customized index implementation
might decide to only return results for some files.
Reviewers: ilya-biryukov
Reviewed By: ilya-biryukov
Subscribers: javed.absar, MaskRay, jkorous, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D50446
llvm-svn: 339320
Summary:
It can be removed from the cache before the first access anyway, so
building it can be a waste of time.
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: javed.absar, MaskRay, jkorous, arphaman, cfe-commits
Differential Revision: https://reviews.llvm.org/D49991
llvm-svn: 338378
Summary:
After r338256, clangd stopped reporting diagnostics if WantDiags::No request
is followed by a WantDiags::Yes request but the AST can be reused.
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: javed.absar, MaskRay, jkorous, arphaman, jfb, cfe-commits
Differential Revision: https://reviews.llvm.org/D50045
llvm-svn: 338361
Summary:
If the contents are the same, the update most likely comes from the
fact that compile commands were invalidated. In that case we want to
avoid rebuilds in case the compile commands are actually the same.
Reviewers: ioeric
Reviewed By: ioeric
Subscribers: simark, javed.absar, MaskRay, jkorous, arphaman, jfb, cfe-commits
Differential Revision: https://reviews.llvm.org/D49783
llvm-svn: 338012
Summary:
log() is split into four functions:
- elog()/log()/vlog() have different severity levels, allowing filtering
- dlog() is a lazy macro which uses LLVM_DEBUG - it logs to the logger, but
conditionally based on -debug-only flag and is omitted in release builds
All logging functions use formatv-style format strings now, e.g:
log("Could not resolve URI {0}: {1}", URI, Result.takeError());
Existing log sites have been split between elog/log/vlog by best guess.
This includes a workaround for passing Error to formatv that can be
simplified when D49170 or similar lands.
Subscribers: ilya-biryukov, javed.absar, ioeric, MaskRay, jkorous, cfe-commits
Differential Revision: https://reviews.llvm.org/D49008
llvm-svn: 336785
Summary:
To avoid doing extra work of processing headers in the preamble
mutilple times in parallel.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: javed.absar, ioeric, MaskRay, jkorous, cfe-commits
Differential Revision: https://reviews.llvm.org/D48940
llvm-svn: 336538
By recomputing CompilerInvocation instead of copying it.
The problem was caused by the fact that copies of CompilerInvocation
store references to the shared state (DiagnosticOptions) when copied,
causing data races when two different copies are destroyed from
different threads.
llvm-svn: 335836
Summary:
After this commit, clangd will only keep the last 3 accessed ASTs in
memory. Preambles for each of the opened files are still kept in
memory to make completion and AST rebuilds fast.
AST rebuilds are usually fast enough, but having the last ASTs in
memory still considerably improves latency of operations like
findDefinition and documeneHighlight, which are often sent multiple
times a second when moving around the code. So keeping some of the last
accessed ASTs in memory seems like a reasonable tradeoff.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: malaperle, arphaman, klimek, javed.absar, ioeric, MaskRay, jkorous, cfe-commits
Differential Revision: https://reviews.llvm.org/D47063
llvm-svn: 333737
Summary:
This is more efficient and avoids data races when reading files that
come from the preamble. The staleness can occur when reading a file
from disk that changed after the preamble was built. This can lead to
crashes, e.g. when parsing comments.
We do not to rely on symbols from the main file anyway, since any info
that those provide can always be taken from the AST.
Reviewers: ioeric, sammccall
Reviewed By: ioeric
Subscribers: malaperle, klimek, javed.absar, MaskRay, jkorous, cfe-commits
Differential Revision: https://reviews.llvm.org/D47272
llvm-svn: 333196
Summary:
It was previously an easy way to concurrently access a mutable vfs,
which is a recipe for disaster.
Reviewers: sammccall
Reviewed By: sammccall
Subscribers: klimek, jkorous-apple, cfe-commits, ioeric
Differential Revision: https://reviews.llvm.org/D44463
llvm-svn: 327537
Summary:
The new implementation attaches notes to diagnostic message and shows
the original diagnostics in the message of the note.
Reviewers: hokein, ioeric, sammccall
Reviewed By: sammccall
Subscribers: klimek, mgorny, cfe-commits, jkorous-apple
Differential Revision: https://reviews.llvm.org/D44142
llvm-svn: 327282
Summary:
Don't actually start building ASTs for new revisions until either:
- 500ms have passed since the last revision, or
- we actually need the revision for something (or to unblock the queue)
In practice, this avoids the "first keystroke results in diagnostics" problem.
This is kind of awkward to test, and the test is pretty bad.
It can be observed nicely by capturing a trace, though.
Reviewers: hokein, ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D43648
llvm-svn: 326546
Summary:
Through the C++ API, we support for a given snapshot version:
- Yes: make sure we generate diagnostics for exactly this version
- Auto: generate eventually-consistent diagnostics for at least this version
- No: don't generate diagnostics for this version
Eventually auto should be debounced for better UX.
Through LSP, we force diagnostics for initial load (bypassing future debouncing)
and all updates follow the "auto" policy.
This is complicated to implement under the CancellationFlag design, so
rewrote that part to just inspect the queue instead.
It turns out we never pass None to the diagnostics callback, so remove Optional
from the signature. The questionable behavior of not invoking the callback at
all if CppFile::rebuild fails is not changed.
Reviewers: ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D43518
llvm-svn: 325774
Summary:
This has a bit of a blast radius, but I think there's enough value in "forcing"
us to give names to these async tasks for debugging. Guessing about what
multithreaded code is doing is so unfun...
The "file" param attached to the tasks may seem to be redundant with the thread
names, but note that thread names are truncated to 15 chars on linux!
We'll be lucky to get the whole basename...
Reviewers: ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D43388
llvm-svn: 325480
Summary:
LSP has asynchronous semantics, being able to block on an async operation
completing is unneccesary and leads to tighter coupling with the threading.
In practice only tests depend on this, so we add a general-purpose "block until
idle" function to the scheduler which will work for all operations.
To get this working, fix a latent condition-variable bug in ASTWorker, and make
AsyncTaskRunner const-correct.
Reviewers: ilya-biryukov
Subscribers: klimek, jkorous-apple, ioeric, cfe-commits
Differential Revision: https://reviews.llvm.org/D43127
llvm-svn: 324990
Initially submitted as r324356 and reverted in r324386.
This change additionally contains a fix to crashes of the buildbots.
The source of the crash was undefined behaviour caused by
std::future<> whose std::promise<> was destroyed without calling
set_value().
llvm-svn: 324575
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:
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:
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
Kadir Cetinkaya