BUG: Local rerender on windows generates spurious diff in .ci_support/README

[h-vetinari]

Rerendering an already fully up-to-date feedstock on windows creates the following diff:

diff --git a/.ci_support/README b/.ci_support/README
index a47316b..69c5db6 100644
--- a/.ci_support/README
+++ b/.ci_support/README
@@ -1,6 +1,6 @@
-This file is automatically generated by conda-smithy. If any
-particular build configuration is expected, but it is not found,
-please make sure all dependencies are satisfiable. To add/modify any
-matrix elements, you should create/change conda-smithy's input
-recipe/conda_build_config.yaml and re-render the recipe, rather than
-editing these files directly.
+This file is automatically generated by conda-smithy. If any^M
+particular build configuration is expected, but it is not found,^M
+please make sure all dependencies are satisfiable. To add/modify any^M
+matrix elements, you should create/change conda-smithy's input^M
+recipe/conda_build_config.yaml and re-render the recipe, rather than^M
+editing these files directly.^M

IOW, all changes are that the line-endings get turned into \r\n instead of \n. This should normally not happen as each feedstock's .gitattributes has:

conda-smithy/conda_smithy/feedstock_content/.gitattributes

Line 1 in 9b727a8

* text=auto

which should use git's autocrlf handling that avoids just that sort of behaviour. So somehow, the rerender manages to regenerate the file with the wrong file endings and then git add it in a way that circumvents .gitattributes.

This has been happening for a while, and it's relatively easy to work around with git reset / git add . before committing changes (because then the .gitattributes takes effect again), but I got asked to open an issue for it in the context of #1821.