Decrease cost of ipynb code path when unneeded (#3748)

IPython is a very expensive import, like, at least 300ms. I'd also
venture that it's much more common than tokenize-rt, which is like 30ms.
I work in a repo where I use black, have IPython installed and there
happen to be a couple notebooks (that we don't want formatted). I know I
can force exclude ipynb, but this change doesn't really have a cost.

diff --git a/CHANGES.md b/CHANGES.md
index a9d4d9d63c96f1dcadab6c5d4ec3b617bef071f7..6fa0e4b7cc0d8e0299538a055b7098ef6e0779f6 100644 (file)
--- a/CHANGES.md
+++ b/CHANGES.md
@@ -47,6 +47,8 @@
 
 <!-- Changes that improve Black's performance. -->
 
+- Avoid importing `IPython` in a case where we wouldn't need it (#3748)
+
 ### Output
 
 <!-- Changes to Black's terminal output and error messages -->

diff --git a/src/black/handle_ipynb_magics.py b/src/black/handle_ipynb_magics.py
index 9e1af757c32a70d4357a0792ae163fc3898c2ffb..2b6b920921195bbbb7ec18741b14ca8ce621097b 100644 (file)
--- a/src/black/handle_ipynb_magics.py
+++ b/src/black/handle_ipynb_magics.py
@@ -58,8 +58,13 @@ class Replacement:
 @lru_cache()
 def jupyter_dependencies_are_installed(*, verbose: bool, quiet: bool) -> bool:
     try:
-        import IPython  # noqa:F401
+        # isort: off
+        # tokenize_rt is less commonly installed than IPython
+        # and IPython is expensive to import
         import tokenize_rt  # noqa:F401
+        import IPython  # noqa:F401
+
+        # isort: on
     except ModuleNotFoundError:
         if verbose or not quiet:
             msg = (