[Question] git cat-file --batch cannot handle irregular path

Hi, all

When I use "git cat-file --batch" command, with the form <tree-ish>:path, here path is the working tree in my repository, but it is irregular, such as "法律\(Law\)/民法\(civil\ law$'\n'\)\", this kind of path is unintentionally created.   The process cannot correctly handle this kind of case:

$ git cat-file --batch d071ea971fb93ebc937517736c2a63d64ddbe0d9:法律(Law)/民法(civil law) d071ea971fb93ebc937517736c2a63d64ddbe0d9:法律(Law)/民法(civil law) missing

How can I deal with this case when irregular paths cannot be avoided?

Thanks in advance.

"lilinchao@oschina.cn" <lilinchao@oschina.cn> writes:

I may be misreading what you wrote, but if you mean that there is a LF in the pathname, I do not think cat-file --batch is prepared to handle it; as we can see in builtin/cat-cile.c:

static int batch_objects(struct batch_options *opt)
{
	...
	while (strbuf_getline(&input, stdin) != EOF) {
		if (data.split_on_whitespace) {
		...
		}

		batch_one_object(input.buf, &output, opt, &data);
	}

it always treats what strbuf_getline() returns as a single record.

If properly designed, it would have

 - supported "-z" to take NUL-delimited input, and

 - when "-z" is not in effect, after reading a single line of input,
   we would have called quote API to un-cquote the string, so that
   we can express LF and other special characters in the pathname.

but those who worked on "cat-file --batch" didn't.

On Mon, Apr 19, 2021 at 09:23:31PM -0700, Junio C Hamano wrote:

Yeah, this has been a long-standing pain.

It would be easy to add "-z" now, and probably worth doing.

It is too late to do the unquoting thing by default now. We could add an option to enable it, but once "-z" exists, it hardly seems worth it.

-Peff