silence clang-3.6 warnings

I was fooling around with clang-3.6 today (as opposed to 3.5, which is the stock "clang" shipped with Debian), and noticed two new warnings. The first one is a real bug, and the second is just a cleanup (though I agree with clang on it).

[1/2]: read_and_strip_branch: fix typo'd address-of operator [2/2]: do not check truth value of flex arrays

-Peff

When we are chomping newlines from the end of a strbuf, we must check "sb.len != 0" before accessing "sb.buf[sb.len - 1]". However, this code mistakenly checks "&sb.len", which is always true (it is a part of an auto struct, so the address is always non-zero). This could lead to us accessing memory outside the strbuf when we read an empty file.

---
This dates back to 8b87cfd (wt-status: move strbuf into read_and_strip_branch(), 2013-03-16), so it is not a bug that needs addressed during the -rc period.

This is the most minimal fix, but I kind of wonder if it should just be using strbuf_rtrim (or even strbuf_trim) in the first place.

 wt-status.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index b54eac5..29666d0 100644
@@ -1140,7 +1140,7 @@ static char *read_and_strip_branch(const char *path)
 	if (strbuf_read_file(&sb, git_path("%s", path), 0) <= 0)
 		goto got_nothing;
 
-	while (&sb.len && sb.buf[sb.len - 1] == '\n')
+	while (sb.len && sb.buf[sb.len - 1] == '\n')
 		strbuf_setlen(&sb, sb.len - 1);
 	if (!sb.len)
 		goto got_nothing;

2.3.0.rc1.287.g761fd19

Yeah. Or strbuf_chomp(), which does not exist ;-)

It is tempting to apply this directly to maint and merge up immediately, as there is no way this 1-byte change will break things (of course that is not necessarily true for random 1-byte changes, though).

It sometimes gets really hard to resist that temptation during the pre-release freeze period.

This is not the first time I've seen this chomp/trim distinction come up. However, the thing that has prevented me from writing strbuf_chomp is that the trim is almost always a more reasonable choice.

Take this instance. We are opening and reading a whole file. Surely we need to drop the final newline, which is not interesting. But we are not just doing that; we are dropping _all_ trailing newlines. So "foo\n\n" becomes "foo". But "foo\n \n" does not. That doesn't make much sense.

IOW, I would venture to say that chomping like this falls into one of two categories:

1. You want to clean up any extraneous cruft. Multiple lines, extra whitespace, etc.

2. You want to read one line, but don't want the trailing newline.

And strbuf_getline already handles case (2).

End mini-rant. :)

That's part of why I did the simplest fix instead of strbuf_rtrim. To tempt you. :)

-Peff

Thanks.

There is no point in checking "!ref->name" when ref is a "struct ref". The name field is a flex-array, and there always has a non-zero address. This is almost certainly not hurting anything, but it does cause clang-3.6 to complain.

---
Note that even if "ref" is NULL, this is still going to be "0+offsetof(name)". So short of unsigned pointer wrap-around, I do not think this check can ever do anything.

 remote-curl.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index dd63bc2..515ac9b 100644
@@ -760,7 +760,7 @@ static int fetch_git(struct discovery *heads,
 
 	for (i = 0; i < nr_heads; i++) {
 		struct ref *ref = to_fetch[i];
-		if (!ref->name || !*ref->name)
+		if (!*ref->name)
 			die("cannot fetch by sha1 over smart http");
 		packet_buf_write(&preamble, "%s %s\n",
 				 sha1_to_hex(ref->old_sha1), ref->name);

2.3.0.rc1.287.g761fd19