xref: /llvm-project/clang-tools-extra/clangd/support/FileCache.h (revision 2c675be9b232c1d0b5c55cbcb196e71036c681ea)
1 //===--- FileCache.h - Revalidating cache of data from disk ------*- C++-*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #ifndef LLVM_CLANG_TOOLS_EXTRA_CLANGD_SUPPORT_FILECACHE_H
10 #define LLVM_CLANG_TOOLS_EXTRA_CLANGD_SUPPORT_FILECACHE_H
11 
12 #include "Path.h"
13 #include "ThreadsafeFS.h"
14 #include "llvm/Support/Chrono.h"
15 #include <mutex>
16 #include <optional>
17 
18 namespace clang {
19 namespace clangd {
20 
21 /// Base class for threadsafe cache of data read from a file on disk.
22 ///
23 /// We want configuration files to be "live" as much as possible.
24 /// Reading them every time is simplest, but caching solves a few problems:
25 ///  - reading and parsing is cheap but not free (and happens on hot paths)
26 ///  - we can ignore invalid data and use the old value (we may see truncated
27 ///    compile_commands.json from non-atomic writers)
28 ///  - we avoid reporting the same errors repeatedly
29 ///
30 /// We still read and parse the data synchronously on demand, but skip as much
31 /// work as possible:
32 ///  - if not enough wall-time has elapsed, assume the data is still up-to-date
33 ///  - if we stat the file and it has the same mtime + size, don't read it
34 ///  - obviously we only have to parse when we re-read the file
35 /// (Tracking OS change events is an alternative, but difficult to do portably.)
36 ///
37 /// Caches for particular data (e.g. compilation databases) should inherit and:
38 ///  - add mutable storage for the cached parsed data
39 ///  - add a public interface implemented on top of read()
40 class FileCache {
41 protected:
42   // Path must be absolute.
43   FileCache(PathRef Path);
44 
45   // Updates the cached value if needed, then provides threadsafe access to it.
46   //
47   // Specifically:
48   // - Parse() may be called (if the cache was not up-to-date)
49   //   The lock is held, so cache storage may be safely written.
50   //   Parse(None) means the file doesn't exist.
51   // - Read() will always be called, to provide access to the value.
52   //   The lock is again held, so the value can be copied or used.
53   //
54   // If the last Parse is newer than FreshTime, we don't check metadata.
55   //   - time_point::min() means we only do IO if we never read the file before
56   //   - time_point::max() means we always at least stat the file
57   //   - steady_clock::now() + seconds(1) means we accept 1 second of staleness
58   void read(const ThreadsafeFS &TFS,
59             std::chrono::steady_clock::time_point FreshTime,
60             llvm::function_ref<void(std::optional<llvm::StringRef>)> Parse,
61             llvm::function_ref<void()> Read) const;
62 
path()63   PathRef path() const { return Path; }
64 
65 private:
66   std::string Path;
67   // Members are mutable so read() can present a const interface.
68   // (It is threadsafe and approximates read-through to TFS).
69   mutable std::mutex Mu;
70   // Time when the cache was known valid (reflected disk state).
71   mutable std::chrono::steady_clock::time_point ValidTime;
72   // Filesystem metadata corresponding to the currently cached data.
73   mutable llvm::sys::TimePoint<> ModifiedTime;
74   mutable uint64_t Size;
75 };
76 
77 } // namespace clangd
78 } // namespace clang
79 
80 #endif
81