Simplify parse task to improve compatibility

Turns out the idea I was saving import time was a lie because
unfortunately I was accessing Black's internal APIs anyway. So
instead of giving myself a maintenance burden that has no good reason
to exist other than "don't touch what's not broken" let's just use a
still internal but much more high-level API from black.

This allows me to both clean up my mypyc-support patch (since it
involved changes to the blib2to3 driver blackbench's parse task
would crash) AND support black>=19.3b0 which is just awesome!

README.md

@@ -24,5 +24,4 @@ Helpful links:
 Blackbench: MIT.
 
 Targets based off real code maintain their original license. Please check the directory
-containing the target in question for a license file. There's also one task based off
-Black's code. Please also check the task templates directory for more information.
+containing the target in question for a license file.

docs/changelog.md

@@ -9,6 +9,7 @@ Date of release: *n/a*
 - Fixed the pyperf requirement which was originally pinning pyperf to 2.0.0 by accident.
 - Added pre-checks for both pyperf arguments and --format-config configuration so issues
   are caught earlier.
+- The parse task now supports installations of Black as old as 19.3b0!
 
 ## 21.7.dev2
 

docs/index.md

@@ -114,5 +114,4 @@ A breakdown of what's happening here:
 Blackbench: MIT.
 
 Targets based off real code maintain their original license. Please check the directory
-containing the target in question for a license file. There's also one task based off
-Black's code. Please also check the task templates directory for more information.
+containing the target in question for a license file.

docs/tasks_and_targets.md

@@ -67,8 +67,7 @@ implementation. This provides a benchmarking speedup and is also unavoidable for
 level tasks like `parse`. Due to this, each task imposes restrictions to what version of
 Black their benchmarks can be run under:
 
-- `fmt` and `fmt-fast`: >= 19.3b0
-- `parse`: >= 21.5b1
+- `fmt`, `fmt-fast`, and `parse`: >= 19.3b0
 
 ## Useful commands
 

src/blackbench/task-templates/parse-template.py

@@ -3,57 +3,19 @@
 
 import pyperf
 
-from blib2to3.pytree import Node, Leaf
-from blib2to3 import pygram, pytree
-from blib2to3.pgen2 import driver
-from blib2to3.pgen2.parse import ParseError
-from black.nodes import syms
-
-
-class InvalidInput(ValueError):
-    """Raised when input source code fails all parse attempts."""
+from blib2to3 import pygram
 
+# First try the relevant function post-refactor in 21.5b1, and fallback
+# to old function location.
+try:
+    from black.parsing import lib2to3_parse
+except ImportError:
+    from black import lib2to3_parse
 
 runner = pyperf.Runner()
-code = Path(r"{target}").read_text(encoding="utf8")
-pygram.initialize(tempfile.gettempdir())
-
-grammars = [
-    # Python 3.7+
-    pygram.python_grammar_no_print_statement_no_exec_statement_async_keywords,
-    # Python 3.0-3.6
-    pygram.python_grammar_no_print_statement_no_exec_statement,
-    # Python 2.7 with future print_function import
-    pygram.python_grammar_no_print_statement,
-    # Python 2.7
-    pygram.python_grammar,
-]
-
-
-def blib2to3_parse(src_txt: str) -> Node:
-    if not src_txt.endswith("\n"):
-        src_txt += "\n"
-
-    for grammar in grammars:
-        drv = driver.Driver(grammar, pytree.convert)
-        try:
-            result = drv.parse_string(src_txt, True)
-            break
-
-        except ParseError as pe:
-            lineno, column = pe.context[1]
-            lines = src_txt.splitlines()
-            try:
-                faulty_line = lines[lineno - 1]
-            except IndexError:
-                faulty_line = "<line number missing in source>"
-            exc = InvalidInput("Cannot parse: " + str(lineno) + ":" + str(column) + ":" + str(faulty_line))
-    else:
-        raise exc from None
-
-    if isinstance(result, Leaf):
-        result = Node(syms.file_input, [result])
-    return result
-
+code =  Path(r"{target}").read_text(encoding="utf8")
 
-runner.bench_func("{name}", blib2to3_parse, code)
+with tempfile.TemporaryDirectory(prefix="blackbench-parsing-") as path:
+    # Block the parser code from using any pre-existing cache.
+    pygram.initialize(path)
+    runner.bench_func("{name}", lib2to3_parse, code)

tests/smoke_tests.py

@@ -35,7 +35,7 @@ def test_provided_targets(tmp_result: Path, run_cmd, target: Target):
     # because their external usages makes maintenance harder :/
     code = target.path.read_text("utf8")
     try:
-        black.format_file_contents(code, fast=False, mode=black.Mode())
+        black.format_file_contents(code, fast=False, mode=black.FileMode())
     except black.NothingChanged:
         pass