The permissions problem reminded me of an issue where CI builds in Cygwin would intermittently fail, complaining about not being able to read a file. However, j

The permissions problem reminded me of an issue where CI builds in Cygwin would intermittently fail, complaining about not being able to read a file. However, just rerunning the job would always succeed. It turned out that although just opening the file for reading worked fine, access(path, R_OK) returned false. The build system only called access() to cover some special case, which didn't occur on the rerun (because of caching?).