* refactor: simplify logic to load web_devicons
* refactor: make register() asynchronous
* fix: load lazily modules outside this plugin
* refactor: simplify logic to wait initialization
* refactor: use uv.hrtime() instead of os.clock()
* fix: avoid errors in calling plenary.log in async
* test: store elapsed time to check in tests
* test: fix module names
This becomes a problem only in Ubuntu because macOS and Windows does not
care cases in filenames.
* test: fix types and unused modules
* style: fix by stylua
* refactor: make recency / entry_maker loaded lazily
Now it uses realpath for registering and validating DB. This means, if you have entries that has filenames differing only for case, it can deal with them as they exist.
Before this, it has miscalculated scores for such cases. For example, in case you have `/path/to/foo.lua` and `/path/to/Foo.lua`, it registers entries for each file. Now it detects accurate filename for the specified one, and removes it in validation.
* test: separate logic for utils
* fix!: register realpath for consistency
* refactor: convert fs module from class
* refactor: move db initialization phase to start()
* fix: run database:start() truly asynchronously
* fix: call each functions with async wrapping
* refactor: add types for args in command
* fix: run register() synchronously
Because vim.api.nvim_* cannot be used in asynchronous functions.
* docs: add note for calling setup() twice
* fix: run non-fast logic on next tick
* chore: change the position of `?`
* fix: use config module to simplify code
* test: fix tests to use frecency.config
* test: remove macOS with nightly Neovim temporarily
* fix: get file lock forcibly
When Neovim ends without removing file locks, it will fail to open
frecency ever because file locks remain. This fix makes it remove file
locks when all attempts fail forcibly.
* test: avoid error in Windows
In Windows, unclosed file cannot be removed, so close it before
unlinking.
* feat: retry to unlink when it fails
* refactor: make logic for Database be abstract
* feat: add logic for DB by string.dump
* fix: run with async.void to run synchronously
* test: add tests for native feature
* feat!: sort candidates by path when score is same
This is needed because candidates from SQLite is sorted by id, but ones from native is sorted by path.
* chore: clean up types
* feat: add lock/unlock feature to access DB
* test: use async version of busted
And disable benchmark tests (fix later)
* test: add tests for file_lock
* chore: use more explicit names
* chore: use plenary.log instead
* fix: wait async functions definitely
* feat: add migrator
* chore: fix logging
* fix: detect emptiness of the table
* fix: deal with buffer with no names
* test: loosen the condition temporarily
* test: add tests for migrator
* fix: return true when the table is not empty
* feat: load sqlite lazily not to require in start
* chore: add logging to calculate time for fetching
* feat: add converter from native code to SQLite
* feat: warn when sqlite.lua is not available
* feat: add FrecencyMigrateDB to migrate DB
* docs: add note for native code logic
* test: ignore type bug
