summary refs log tree commit diff
path: root/resolv
diff options
context:
space:
mode:
authorStefan Liebler <stli@linux.ibm.com>2019-07-02 14:07:44 +0200
committerStefan Liebler <stli@linux.ibm.com>2019-07-02 14:07:44 +0200
commitd1bc2cbbed9aea2017ef941f63c8786571da5b4f (patch)
treeeb77d636ea206639a5ea99ce0f37e223a954e163 /resolv
parent72edea80c1bb402f41d8bca58c835abf46dadd48 (diff)
downloadglibc-d1bc2cbbed9aea2017ef941f63c8786571da5b4f.tar.gz
glibc-d1bc2cbbed9aea2017ef941f63c8786571da5b4f.tar.xz
glibc-d1bc2cbbed9aea2017ef941f63c8786571da5b4f.zip
Fix build warnings in resolv/res_send.c
This patch fixes the gcc warnings seen with gcc 9 -march>=z13 on s390x:
res_send.c: In function ‘__res_context_send’:
res_send.c:498:7: error: ‘resplen’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
  498 |    if (n == 0 && (buf2 == NULL || *resplen2 == 0))
      |       ^

In this case send_vc is inlined into __res_context_send
and the maybe uninitialized resplen belongs to the one in send_vc.

In send_vc there is already a DIAG_IGNORE_NEEDS_COMMENT (5, "-Wmaybe-uninitialized")
and a comment which explains that this is a false-positive.
Note that resplen is used as return value.

This patch adds a further DIAG_IGNORE_NEEDS_COMMENT around the declaration of n
in __res_context_send and the comparison after n was set to the return value of send_vc.

ChangeLog:

	* resolv/res_send.c (__res_context_send): Disable maybe-uninitialized
	warning.
Diffstat (limited to 'resolv')
-rw-r--r--resolv/res_send.c17
1 files changed, 16 insertions, 1 deletions
diff --git a/resolv/res_send.c b/resolv/res_send.c
index eeeb8acf27..ed27f3abf8 100644
--- a/resolv/res_send.c
+++ b/resolv/res_send.c
@@ -401,7 +401,18 @@ __res_context_send (struct resolv_context *ctx,
 		    int *nansp2, int *resplen2, int *ansp2_malloced)
 {
 	struct __res_state *statp = ctx->resp;
-	int gotsomewhere, terrno, try, v_circuit, resplen, n;
+	int gotsomewhere, terrno, try, v_circuit, resplen;
+	/* On some architectures send_vc is inlined and the compiler might emit
+	   a warning indicating 'resplen' may be used uninitialized.  Note that
+	   the warning belongs to resplen in send_vc which is used as return
+	   value!  There the maybe-uninitialized warning is already ignored as
+	   it is a false-positive - see comment in send_vc.
+	   Here the variable n is set to the return value of send_vc.
+	   See below.  */
+	DIAG_PUSH_NEEDS_COMMENT;
+	DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized");
+	int n;
+	DIAG_POP_NEEDS_COMMENT;
 
 	if (statp->nscount == 0) {
 		__set_errno (ESRCH);
@@ -495,8 +506,12 @@ __res_context_send (struct resolv_context *ctx,
 				    ansp2_malloced);
 			if (n < 0)
 				return (-1);
+			/* See comment at the declaration of n.  */
+			DIAG_PUSH_NEEDS_COMMENT;
+			DIAG_IGNORE_NEEDS_COMMENT (9, "-Wmaybe-uninitialized");
 			if (n == 0 && (buf2 == NULL || *resplen2 == 0))
 				goto next_ns;
+			DIAG_POP_NEEDS_COMMENT;
 		} else {
 			/* Use datagrams. */
 			n = send_dg(statp, buf, buflen, buf2, buflen2,