comm: report all I/O errors #9086
Conversation
|
GNU testsuite comparison: |
CodSpeed Performance ReportMerging #9086 will not alter performanceComparing Summary
|
ad54136 to
d9f4f19
Compare
|
GNU testsuite comparison: |
I/O errors were being silently suppressed. This also removes an unnecessary check to see if one of the inputs is a directory.
d9f4f19 to
6bc9f47
Compare
|
For Windows I decided to stick with the same behavior of Unix: report the error from the operating system. In the case of Windows, if the file is a directory, you get "permission denied". Not very user friendly, but I think this is the best approach, mostly because it's not racy. If the maintainers don't like that, we can add a metadata check just for Windows (not on Unix), maybe something like: try to open the file, if it fails with a "permission denied", check if it's a directory, and based on that report the most appropriate error message. Let me know what you prefer and I will update the PR accordingly. |
|
GNU testsuite comparison: |
If a read error happens after opening the file, that error is never reported and
commjust quits without printing any error message. This commit fixes that.This also removes the
is_dir()check, preventing a potential race condition where the path could be swapped with a directory in between theis_dir()check and theopen()call. On Unix, this correctly results in the "is a directory" error message to be printed (this is detected atread()time).