Add dynamic learning from accepted corrections

Record user-accepted corrections in a shelve-backed store and replay
them on future matching mistakes, bypassing rule evaluation entirely.

Two-level matching: exact full-command lookup, then word-level
reconstruction that generalises to different arguments (e.g. learning
'git psuh origin main' also fixes 'git psuh origin dev').

tests/entrypoints/test_fix_command_learned.py

@@ -0,0 +1,131 @@
+import pytest
+from mock import Mock, patch, call
+from thefuck.entrypoints.fix_command import fix_command
+from thefuck.types import CorrectedCommand
+
+
+@pytest.fixture
+def mock_learned(monkeypatch):
+    state = {"correction": None, "recordings": []}
+
+    def fake_get_correction(script):
+        return state["correction"]
+
+    def fake_record(original, corrected):
+        state["recordings"].append((original, corrected))
+
+    monkeypatch.setattr(
+        "thefuck.entrypoints.fix_command.get_correction", fake_get_correction
+    )
+    monkeypatch.setattr("thefuck.entrypoints.fix_command.record", fake_record)
+    return state
+
+
+@pytest.fixture
+def known_args():
+    return Mock(
+        force_command="git psuh origin main", yes=False, debug=False, repeat=False
+    )
+
+
+class TestLearnedAutoApply(object):
+    def test_auto_applies_learned_correction(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        mock_learned["correction"] = "git push origin main"
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.get_corrected_commands", lambda _: iter([])
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.select_command", lambda _: None
+        )
+
+        with patch("thefuck.types.CorrectedCommand.run") as mock_run, patch(
+            "thefuck.logs.show_corrected_command"
+        ):
+            fix_command(known_args)
+            mock_run.assert_called_once()
+
+    def test_learned_skips_rule_matching(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        mock_learned["correction"] = "git push origin main"
+        get_corrected = Mock()
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.get_corrected_commands", get_corrected
+        )
+
+        with patch("thefuck.types.CorrectedCommand.run"), patch(
+            "thefuck.logs.show_corrected_command"
+        ):
+            fix_command(known_args)
+            get_corrected.assert_not_called()
+
+    def test_shows_corrected_command_on_auto_apply(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        mock_learned["correction"] = "git push origin main"
+
+        with patch("thefuck.types.CorrectedCommand.run"), patch(
+            "thefuck.logs.show_corrected_command"
+        ) as mock_show:
+            fix_command(known_args)
+            assert mock_show.call_count == 1
+            shown_cmd = mock_show.call_args[0][0]
+            assert shown_cmd.script == "git push origin main"
+
+
+class TestRecordOnSelection(object):
+    def test_records_user_selection(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        selected = CorrectedCommand(
+            script="git push origin main", side_effect=None, priority=100
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.get_corrected_commands",
+            lambda _: iter([selected]),
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.select_command", lambda _: selected
+        )
+
+        with patch("thefuck.types.CorrectedCommand.run"):
+            fix_command(known_args)
+            assert mock_learned["recordings"] == [
+                ("git psuh origin main", "git push origin main")
+            ]
+
+    def test_does_not_record_on_abort(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.get_corrected_commands", lambda _: iter([])
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.select_command", lambda _: None
+        )
+
+        with pytest.raises(SystemExit):
+            fix_command(known_args)
+        assert mock_learned["recordings"] == []
+
+
+class TestFallthrough(object):
+    def test_falls_through_to_rules_when_no_learned(
+        self, mock_learned, known_args, settings, monkeypatch
+    ):
+        selected = CorrectedCommand(
+            script="git push origin main", side_effect=None, priority=100
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.get_corrected_commands",
+            lambda _: iter([selected]),
+        )
+        monkeypatch.setattr(
+            "thefuck.entrypoints.fix_command.select_command", lambda _: selected
+        )
+
+        with patch("thefuck.types.CorrectedCommand.run") as mock_run:
+            fix_command(known_args)
+            mock_run.assert_called_once()

tests/test_learned.py

@@ -0,0 +1,139 @@
+import pytest
+from thefuck.learned import LearnedCorrections
+
+
+@pytest.fixture
+def learned(tmp_path):
+    lc = LearnedCorrections()
+    lc._db = {}
+    return lc
+
+
+class TestRecord(object):
+    def test_noop_when_scripts_identical(self, learned):
+        learned.record("git push", "git push")
+        assert len(learned.db) == 0
+
+    def test_stores_full_command_mapping(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        assert (
+            learned.db["cmd:git psuh origin main"]["corrected"]
+            == "git push origin main"
+        )
+
+    def test_increments_count_on_repeat(self, learned):
+        learned.record("git psuh", "git push")
+        learned.record("git psuh", "git push")
+        assert learned.db["cmd:git psuh"]["count"] == 2
+
+    def test_updates_timestamp(self, learned):
+        learned.record("git psuh", "git push")
+        first_ts = learned.db["cmd:git psuh"]["timestamp"]
+        learned.record("git psuh", "git push")
+        assert learned.db["cmd:git psuh"]["timestamp"] >= first_ts
+
+    def test_stores_command_word_replacement(self, learned):
+        learned.record("pyhton script.py", "python script.py")
+        assert learned.db["word:pyhton"]["replacement"] == "python"
+
+    def test_stores_subcommand_replacement(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        assert learned.db["part:git:psuh"]["replacement"] == "push"
+
+    def test_stores_multiple_word_diffs(self, learned):
+        learned.record("gti comit -m msg", "git commit -m msg")
+        assert learned.db["word:gti"]["replacement"] == "git"
+        assert learned.db["part:git:comit"]["replacement"] == "commit"
+
+    def test_no_word_level_when_lengths_differ(self, learned):
+        learned.record("git push", "git push --set-upstream origin main")
+        assert "cmd:git push" in learned.db
+        assert not any(
+            k.startswith("word:") or k.startswith("part:") for k in learned.db
+        )
+
+    def test_updates_correction_on_changed_choice(self, learned):
+        learned.record("apt install vim", "sudo apt install vim")
+        learned.record("apt install vim", "apt-get install vim")
+        assert learned.db["cmd:apt install vim"]["corrected"] == "apt-get install vim"
+        assert learned.db["cmd:apt install vim"]["count"] == 2
+
+
+class TestGetCorrection(object):
+    def test_exact_full_command_match(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        assert learned.get_correction("git psuh origin main") == "git push origin main"
+
+    def test_generalises_subcommand_to_different_args(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        assert learned.get_correction("git psuh origin dev") == "git push origin dev"
+
+    def test_generalises_command_word(self, learned):
+        learned.record("pyhton script.py", "python script.py")
+        assert learned.get_correction("pyhton other.py") == "python other.py"
+
+    def test_combined_command_and_subcommand(self, learned):
+        learned.record("gti comit -m msg", "git commit -m msg")
+        assert learned.get_correction('gti comit -m "other"') == 'git commit -m "other"'
+
+    def test_returns_none_when_no_match(self, learned):
+        assert learned.get_correction("totally unknown cmd") is None
+
+    def test_returns_none_for_empty_script(self, learned):
+        assert learned.get_correction("") is None
+
+    def test_prefers_full_command_over_word_level(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        learned.db["cmd:git psuh origin main"]["corrected"] = (
+            "git push --force origin main"
+        )
+        assert (
+            learned.get_correction("git psuh origin main")
+            == "git push --force origin main"
+        )
+
+    def test_word_level_only_replaces_known_tokens(self, learned):
+        learned.record("git psuh origin main", "git push origin main")
+        result = learned.get_correction("git psuh origin dev")
+        assert result == "git push origin dev"
+
+    def test_cross_resolve_command_and_part(self, learned):
+        """When both cmd word and subcommand are typos, parts stored
+        under the corrected cmd name still resolve."""
+        learned.record("gti psuh origin main", "git push origin main")
+        assert learned.get_correction("gti psuh origin dev") == "git push origin dev"
+
+    def test_single_word_command(self, learned):
+        learned.record("sl", "ls")
+        assert learned.get_correction("sl") == "ls"
+
+
+class TestClear(object):
+    def test_removes_all_entries(self, learned):
+        learned.record("git psuh", "git push")
+        learned.record("pyhton x.py", "python x.py")
+        learned.clear()
+        assert len(learned.db) == 0
+
+    def test_no_matches_after_clear(self, learned):
+        learned.record("git psuh", "git push")
+        learned.clear()
+        assert learned.get_correction("git psuh") is None
+
+
+class TestRoundTrip(object):
+    def test_record_then_match(self, learned):
+        learned.record("docker bilud .", "docker build .")
+        assert learned.get_correction("docker bilud .") == "docker build ."
+
+    def test_record_then_generalise(self, learned):
+        learned.record("docker bilud -t foo .", "docker build -t foo .")
+        assert (
+            learned.get_correction("docker bilud -t bar .") == "docker build -t bar ."
+        )
+
+    def test_multiple_distinct_commands(self, learned):
+        learned.record("git psuh", "git push")
+        learned.record("pyhton x.py", "python x.py")
+        assert learned.get_correction("git psuh") == "git push"
+        assert learned.get_correction("pyhton y.py") == "python y.py"

thefuck/entrypoints/fix_command.py

@@ -6,6 +6,7 @@ from .. import logs, types, const
 from ..conf import settings
 from ..corrector import get_corrected_commands
 from ..exceptions import EmptyCommand
+from ..learned import get_correction, record
 from ..ui import select_command
 from ..utils import get_alias, get_all_executables
 
@@ -13,10 +14,10 @@ from ..utils import get_alias, get_all_executables
 def _get_raw_command(known_args):
     if known_args.force_command:
         return [known_args.force_command]
-    elif not os.environ.get('TF_HISTORY'):
+    elif not os.environ.get("TF_HISTORY"):
         return known_args.command
     else:
-        history = os.environ['TF_HISTORY'].split('\n')[::-1]
+        history = os.environ["TF_HISTORY"].split("\n")[::-1]
         alias = get_alias()
         executables = get_all_executables()
         for command in history:
@@ -29,20 +30,30 @@ def _get_raw_command(known_args):
 def fix_command(known_args):
     """Fixes previous command. Used when `thefuck` called without arguments."""
     settings.init(known_args)
-    with logs.debug_time('Total'):
+    with logs.debug_time("Total"):
-        logs.debug(u'Run with settings: {}'.format(pformat(settings)))
+        logs.debug("Run with settings: {}".format(pformat(settings)))
         raw_command = _get_raw_command(known_args)
 
         try:
             command = types.Command.from_raw_script(raw_command)
         except EmptyCommand:
-            logs.debug('Empty command, nothing to do')
+            logs.debug("Empty command, nothing to do")
+            return
+
+        learned_script = get_correction(command.script)
+        if learned_script:
+            learned_cmd = types.CorrectedCommand(
+                script=learned_script, side_effect=None, priority=0
+            )
+            logs.show_corrected_command(learned_cmd)
+            learned_cmd.run(command)
             return
 
         corrected_commands = get_corrected_commands(command)
         selected_command = select_command(corrected_commands)
 
         if selected_command:
+            record(command.script, selected_command.script)
             selected_command.run(command)
         else:
             sys.exit(1)

thefuck/learned.py

@@ -0,0 +1,157 @@
+import atexit
+import os
+import shelve
+import time
+
+from . import logs
+
+try:
+    import dbm
+
+    _shelve_open_error = (dbm.error,)
+except ImportError:
+    try:
+        import anydbm
+
+        _shelve_open_error = (anydbm.error,)
+    except ImportError:
+        _shelve_open_error = ()
+
+
+class LearnedCorrections(object):
+    def __init__(self):
+        self._db = None
+
+    def _init_db(self):
+        try:
+            self._setup_db()
+        except Exception:
+            logs.debug("Unable to init learned-corrections db")
+            self._db = {}
+
+    def _setup_db(self):
+        cache_dir = self._get_cache_dir()
+        cache_path = os.path.join(cache_dir, "thefuck_learned")
+        try:
+            self._db = shelve.open(cache_path)
+        except _shelve_open_error + (ImportError,):
+            logs.warn("Removing possibly out-dated learned-corrections db")
+            for suffix in ("", ".db", ".dir", ".bak", ".dat"):
+                path = cache_path + suffix
+                if os.path.exists(path):
+                    os.remove(path)
+            self._db = shelve.open(cache_path)
+        atexit.register(self._db.close)
+
+    @staticmethod
+    def _get_cache_dir():
+        cache_dir = os.getenv("XDG_CACHE_HOME", os.path.expanduser("~/.cache"))
+        try:
+            os.makedirs(cache_dir)
+        except OSError:
+            if not os.path.isdir(cache_dir):
+                raise
+        return cache_dir
+
+    @property
+    def db(self):
+        if self._db is None:
+            self._init_db()
+        return self._db
+
+    def record(self, original_script, corrected_script):
+        if original_script == corrected_script:
+            return
+
+        db = self.db
+        now = time.time()
+
+        original_parts = original_script.split()
+        corrected_parts = corrected_script.split()
+
+        full_key = "cmd:" + original_script
+        entry = db.get(full_key, {})
+        entry["corrected"] = corrected_script
+        entry["count"] = entry.get("count", 0) + 1
+        entry["timestamp"] = now
+        db[full_key] = entry
+
+        # Word-level diffs: only when token counts match, store each
+        # changed token keyed by position so lookups can generalise
+        # (e.g. learning "git psuh origin main" also fixes "git psuh origin dev")
+        if (
+            original_parts
+            and corrected_parts
+            and len(original_parts) == len(corrected_parts)
+        ):
+            for i, (orig_tok, corr_tok) in enumerate(
+                zip(original_parts, corrected_parts)
+            ):
+                if orig_tok == corr_tok:
+                    continue
+                if i == 0:
+                    key = "word:" + orig_tok
+                else:
+                    # Keyed under the corrected cmd name so "gti psuh"
+                    # resolves via word:gti→git then part:git:psuh→push
+                    key = "part:" + corrected_parts[0] + ":" + orig_tok
+                part_entry = db.get(key, {})
+                part_entry["replacement"] = corr_tok
+                part_entry["count"] = part_entry.get("count", 0) + 1
+                part_entry["timestamp"] = now
+                db[key] = part_entry
+
+        self._sync()
+
+    def get_correction(self, script):
+        db = self.db
+
+        full_key = "cmd:" + script
+        entry = db.get(full_key)
+        if entry:
+            return entry["corrected"]
+
+        parts = script.split()
+        if not parts:
+            return None
+
+        corrected_parts = list(parts)
+        found = False
+
+        word_entry = db.get("word:" + parts[0])
+        if word_entry:
+            corrected_parts[0] = word_entry["replacement"]
+            found = True
+
+        # Part lookups use the (possibly corrected) cmd name so that
+        # "gti psuh" resolves even though parts are stored under "git"
+        cmd_name = corrected_parts[0]
+        for i in range(1, len(parts)):
+            part_entry = db.get("part:" + cmd_name + ":" + parts[i])
+            if part_entry:
+                corrected_parts[i] = part_entry["replacement"]
+                found = True
+
+        if found:
+            return " ".join(corrected_parts)
+
+        return None
+
+    def clear(self):
+        db = self.db
+        for key in list(db.keys()):
+            del db[key]
+        self._sync()
+
+    def _sync(self):
+        try:
+            self.db.sync()
+        except AttributeError:
+            pass
+
+
+_learned = LearnedCorrections()
+
+record = _learned.record
+get_correction = _learned.get_correction
+clear = _learned.clear