[8.6.0] Prevent InvalidPathException in rctx.download_and_extract

[bazel-io]

Fixes #28121

Closes #28128.

PiperOrigin-RevId: 852374394
Change-Id: I4390a6c0581cc955494bf3be549b0ca9efd48c7b

Commit 5578c00

[gemini-code-assist]

Code Review

This pull request aims to prevent InvalidPathException when downloading and extracting files by sanitizing the filename derived from the URL. The approach is to replace characters that are not filesystem-safe with an underscore. The core logic change is sound, and a new test case has been added to verify the fix for a filename containing a colon. However, the sanitization is incomplete and can still lead to InvalidPathException on Windows. I've left a comment with details and a suggestion to make the fix more robust.

[gemini-code-assist]

This sanitization is a good step, but it's incomplete and can still lead to InvalidPathException on Windows in a couple of cases:

1. Filenames ending with a dot: Windows does not allow filenames to end with a period. For example, if basename is myfile., it will pass through the current filter, but Path.getRelative("myfile.") will fail on Windows.
2. Reserved filenames: Windows reserves certain filenames, such as CON, PRN, AUX, NUL, and COM1-COM9, LPT1-LPT9. These names are invalid regardless of extension (e.g., CON.txt is invalid). If basename is CON.zip, it will not be sanitized and will cause an error on Windows.

I'd recommend enhancing the sanitization to handle these cases. Here's a suggestion to address the trailing dot issue and prevent an empty basename, which is the more straightforward of the two to fix:

    String sanitized = FS_UNSAFE_CHARS.replaceFrom(basename, '_');
    sanitized = CharMatcher.is('.').trimTrailingFrom(sanitized);
    if (sanitized.isEmpty()) {
      sanitized = "temp";
    }
    return output.getRelative(sanitized);