Re: [PATCH v4 1/4] revision: improve commit_rewrite_person()

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 14, 2022, 21:02 UTC
Message-ID: <xmqq8rov4dc5.fsf@gitster.g>
In-Reply-To: <20220712160634.213956-2-siddharthasthana31@gmail.com>

Siddharth Asthana <siddharthasthana31@gmail.com> writes:

> +/*
> + * Returns the difference between the new and old length of the ident line.
> + */
> +static ssize_t rewrite_ident_line(const char* person, struct strbuf *buf, struct string_list *mailmap)

"const char *person". Asterisk sticks to the variable, not to the type.

Wrap such an overly long line.

> +static void commit_rewrite_person(struct strbuf *buf, const char **headers, struct string_list *mailmap)

Likewise.

The second parameter should be called "header", not "headers", because the way this array is used is primarily to access its individual members, i.e. header[i] and it is more grammatical to say that header[0] is the zero-th header, not headers[0].

Show 11 quoted lines

> +{
> +	size_t buf_offset = 0;
> +
> +	if (!mailmap)
> +		return;
> +
> +	for (;;) {
> +		const char *person, *line;
> +		size_t i, linelen;
> +
> +		line = buf->buf + buf_offset;

buf_offset is initialized to 0. The idea is to keep track of the position of the current line in buf->buf as the byte offset from the beginning. This makes it safe to let rewrite_ident_line() reallocate buf->buf in the loop. So "line" points at the beginning of the current line.

> +		linelen = strchrnul(line, '\n') - line + 1;

This "linelen" counts the LF (or NUL) termination.  Hence ...

> +		if (linelen <= 1)
> +			/* End of header */
> +			return;

... linelen==0 means we are at an empty line.

> +		buf_offset += linelen;

And by adding linelen to buf_offset, we prepare buf_offset to point at the beginning of the next line.

BUT.

What happens when the buffer has only headers, no body, without an empty line after the header? buf_offset will be pointing at the byte past the final NUL at the end of the buffer. The next round of the loop will point line at an invalid piece of memory.

I _think_ that the current generation of high-level tools like "git commit" and "git tag" may leave an blank line at the end even when they are told to create a message with no body, but this code should not depend on that. An object with no body and without a blank line after the header is a valid object (cf. fsck.c::verify_headers()).

> +		for (i = 0; headers[i]; i++)
> +			if (skip_prefix(line, headers[i], &person))
> +				buf_offset += rewrite_ident_line(person, buf, mailmap);
> +	}

As the commit_headers[] array the caller gives us does not have duplicates, as soon as we find a match, we should break out of the loop, i.e.

		for (i = 0; header[i]; i++) {
			if (skip_prefix(line, header[i], &person))
				buf_offset += rewrite_ident_line(person, buf, maimap);
			break;
		}

Show 10 quoted lines

> +}
> +
>  static int commit_match(struct commit *commit, struct rev_info *opt)
>  {
>  	int retval;
> @@ -3832,11 +3859,12 @@ static int commit_match(struct commit *commit, struct rev_info *opt)
>  		strbuf_addstr(&buf, message);
>  
>  	if (opt->grep_filter.header_list && opt->mailmap) {
> +		const char *commit_headers[] = { "author ", "committer ", NULL };

It is OK to call the array "commit_headers", because the way we use the identifier is only to refer to the collection of headers as a whole.

Show 9 quoted lines

>  		if (!buf.len)
>  			strbuf_addstr(&buf, message);
>  
> -		commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
> -		commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
> +		commit_rewrite_person(&buf, commit_headers, opt->mailmap);
>  	}
>  
>  	/* Append "fake" message parts as needed */

Thanks.