On Sat, Apr 26, 2025 at 06:14:52PM +0000, Johannes Thyssen Tishman wrote:
> While testing the symlink behavior from my previous email I noticed some
> leftover /tmp/got-importmsg files that were not cleaned after incomplete
> imports with empty commit messages. From what I understand, this happens
> because the logmsg_path pointer is freed before unlink() can be called
> even when the pointer is already being freed in cmd_import(). Is this
> correct? The diff below seems to fix this for me. Regression tests are
> all passing.

A general principle I like to follow is that on error, the function which
creates resources for its caller will not return those resources and clean
them up itself. So in this case I would prefer to duplicate the unlink() call
inside collect_import_msg() if an error occurred, because that's where the
bug is.