Add exact line coverage report (#2057)

The output is intended to be fed into coverage.py, the Python code
coverage tool, for example:
```
(echo -n "!coverage.py: This is a private format, don't read it directly!"; cat coverage.json) > .coverage
coverage html
```

Author: ddfisher

mypy/main.py

@@ -219,6 +219,8 @@ def process_options(args: List[str]) -> Tuple[List[BuildSource], Options]:
                               dest='special-opts:xslt_txt_report')
     report_group.add_argument('--linecount-report', metavar='DIR',
                               dest='special-opts:linecount_report')
+    report_group.add_argument('--linecoverage-report', metavar='DIR',
+                              dest='special-opts:linecoverage_report')
 
     code_group = parser.add_argument_group(title='How to specify the code to type check')
     code_group.add_argument('-m', '--module', action='append', metavar='MODULE',

mypy/report.py

@@ -2,11 +2,12 @@
 
 from abc import ABCMeta, abstractmethod
 import cgi
+import json
 import os
 import shutil
 import tokenize
 
-from typing import Callable, Dict, List, Tuple, cast
+from typing import Callable, Dict, List, Optional, Tuple, cast
 
 from mypy.nodes import MypyFile, Node, FuncDef
 from mypy import stats
@@ -105,6 +106,115 @@ def on_finish(self) -> None:
 reporter_classes['linecount'] = LineCountReporter
 
 
+class LineCoverageVisitor(TraverserVisitor):
+    def __init__(self, source: List[str]) -> None:
+        self.source = source
+
+        # For each line of source, we maintain a pair of
+        #  * the indentation level of the surrounding function
+        #    (-1 if not inside a function), and
+        #  * whether the surrounding function is typed.
+        # Initially, everything is covered at indentation level -1.
+        self.lines_covered = [(-1, True) for l in source]
+
+    # The Python AST has position information for the starts of
+    # elements, but not for their ends. Fortunately the
+    # indentation-based syntax makes it pretty easy to find where a
+    # block ends without doing any real parsing.
+
+    # TODO: Handle line continuations (explicit and implicit) and
+    # multi-line string literals. (But at least line continuations
+    # are normally more indented than their surrounding block anyways,
+    # by PEP 8.)
+
+    def indentation_level(self, line_number: int) -> Optional[int]:
+        """Return the indentation of a line of the source (specified by
+        zero-indexed line number). Returns None for blank lines or comments."""
+        line = self.source[line_number]
+        indent = 0
+        for char in list(line):
+            if char == ' ':
+                indent += 1
+            elif char == '\t':
+                indent = 8 * ((indent + 8) // 8)
+            elif char == '#':
+                # Line is a comment; ignore it
+                return None
+            elif char == '\n':
+                # Line is entirely whitespace; ignore it
+                return None
+            # TODO line continuation (\)
+            else:
+                # Found a non-whitespace character
+                return indent
+        # Line is entirely whitespace, and at end of file
+        # with no trailing newline; ignore it
+        return None
+
+    def visit_func_def(self, defn: FuncDef) -> None:
+        start_line = defn.get_line() - 1
+        start_indent = self.indentation_level(start_line)
+        cur_line = start_line + 1
+        end_line = cur_line
+        # After this loop, function body will be lines [start_line, end_line)
+        while cur_line < len(self.source):
+            cur_indent = self.indentation_level(cur_line)
+            if cur_indent is None:
+                # Consume the line, but don't mark it as belonging to the function yet.
+                cur_line += 1
+            elif cur_indent > start_indent:
+                # A non-blank line that belongs to the function.
+                cur_line += 1
+                end_line = cur_line
+            else:
+                # We reached a line outside the function definition.
+                break
+
+        is_typed = defn.type is not None
+        for line in range(start_line, end_line):
+            old_indent, _ = self.lines_covered[line]
+            assert start_indent > old_indent
+            self.lines_covered[line] = (start_indent, is_typed)
+
+        # Visit the body, in case there are nested functions
+        super().visit_func_def(defn)
+
+
+class LineCoverageReporter(AbstractReporter):
+    """Exact line coverage reporter.
+
+    This reporter writes a JSON dictionary with one field 'lines' to
+    the file 'coverage.json' in the specified report directory. The
+    value of that field is a dictionary which associates to each
+    source file's absolute pathname the list of line numbers that
+    belong to typed functions in that file.
+    """
+    def __init__(self, reports: Reports, output_dir: str) -> None:
+        super().__init__(reports, output_dir)
+        self.lines_covered = {}  # type: Dict[str, List[int]]
+
+        stats.ensure_dir_exists(output_dir)
+
+    def on_file(self, tree: MypyFile, type_map: Dict[Node, Type]) -> None:
+        tree_source = open(tree.path).readlines()
+
+        coverage_visitor = LineCoverageVisitor(tree_source)
+        tree.accept(coverage_visitor)
+
+        covered_lines = []
+        for line_number, (_, typed) in enumerate(coverage_visitor.lines_covered):
+            if typed:
+                covered_lines.append(line_number + 1)
+
+        self.lines_covered[os.path.abspath(tree.path)] = covered_lines
+
+    def on_finish(self) -> None:
+        with open(os.path.join(self.output_dir, 'coverage.json'), 'w') as f:
+            json.dump({'lines': self.lines_covered}, f)
+
+reporter_classes['linecoverage'] = LineCoverageReporter
+
+
 class OldHtmlReporter(AbstractReporter):
     """Old HTML reporter.