On Fri, Jul 15 2022, Derrick Stolee via GitGitGadget wrote:
> From: Derrick Stolee <derrickstolee@xxxxxxxxxx>
> [...]
> - logerror("'%s': not in whitelist", path);
> + logerror("'%s': not in directory list", path);
> return NULL; /* Fallthrough. Deny by default */
> }
> if (strict_paths && (!ok_paths || !*ok_paths))
> - die("option --strict-paths requires a whitelist");
> + die("option --strict-paths requires a directory list");
>
> if (base_path && !is_directory(base_path))
> die("base-path '%s' does not exist or is not a directory",
The former here is worse than before, which relates to the latter.
Put yourself in the shoes of a user who gets this error message. You
then do "man git-daemon", search for "whitelist", and the only thing
you'll find are references to --strict-paths and --interpolated-path,
which puts you on the right path.
IOW I don't care about the term "whitelist" here, I care that it it has
(goes and counts...) 5 occurances in the manpage, directorty has 16.
And SYNOPSIS of "git-daemon" says "...[<directory...]", so that
hypothehtical user is probably wondering e.g. if the directory they
provided on argv is bad. It's not just 5 v.s. 16, it's that we've
introduced jargon we only used for that feature, but now we use a vague
general term.
So we need to tell the user *what directory*. That's obviously easy, we
should refer back to the option that brought us here.
So per [1] I really welcome any improvements to the docs, but I was also
expecting to find various issues related to the origin of this series
blindly replacing things :( Please look a bit more at the bigger picture
beyond the specific hunks being changed, I might have missed some
similar issues on this read-through...
1. https://lore.kernel.org/git/220715.86sfn2zlkm.gmgdl@xxxxxxxxxxxxxxxxxxx/
