threads / discuss / 14518

git-remote SEGV on t5505 test.

Subject: git-remote SEGV on t5505 test.

## tl;dr

7 messages between Jul 18, 2008 and Jul 18, 2008.

replies: 6people: 2as markdown or json

SungHyun Nam· Jul 18, 2008, 01:46 UTC · lore
Hello,

The gdb backtrace said: (gdb) bt #0 0xff03455c in strlen () from /usr/lib/libc.so.1 #1 0xff087058 in _doprnt () from /usr/lib/libc.so.1 #2 0xff088c18 in printf () from /usr/lib/libc.so.1 #3 0x0007df20 in prune (argc=1556136, argv=0x139cd0) at builtin-remote.c:590

The codes in builtin-remote.c:590,
         printf(" * [%s] %s\n", dry_run ? "would prune" : "prune",
                skip_prefix(refname, "refs/remotes/"));

And the 'skip_prefix()' returns NULL in this case. (The old skip_prefix() never returns NULL).

It seems there is another place need to be checked:
   builtin-remote.c:464 : can pass NULL to the strdup().

Regards, namsh

Junio C Hamano· Jul 18, 2008, 04:25 UTC · re: SungHyun Nam · lore

Re: git-remote SEGV on t5505 test.

SungHyun Nam <namsh@posdata.co.kr> writes:
> And the 'skip_prefix()' returns NULL in this case.
> (The old skip_prefix() never returns NULL).
Thanks.  Something like this?
 builtin-remote.c |   51 +++++++++++++++++++--------------------------------
 1 files changed, 19 insertions(+), 32 deletions(-)
diff --git a/builtin-remote.c b/builtin-remote.c
index 1491354..db12668 100644
--- a/builtin-remote.c
+++ b/builtin-remote.c
@@ -147,6 +147,15 @@ struct branch_info {
 
 static struct path_list branch_list;
 
+static const char *abbrev_ref(const char *name, const char *prefix)
+{
+	const char *abbrev = skip_prefix(name, prefix);
+	if (abbrev)
+		return abbrev;
+	return name;
+}
+#define abbrev_branch(name) abbrev_ref((name), "refs/heads/")
+
 static int config_read_branches(const char *key, const char *value, void *cb)
 {
 	if (!prefixcmp(key, "branch.")) {
@@ -176,18 +185,12 @@ static int config_read_branches(const char *key, const char *value, void *cb)
 			info->remote = xstrdup(value);
 		} else {
 			char *space = strchr(value, ' ');
-			const char *ptr = skip_prefix(value, "refs/heads/");
-			if (ptr)
-				value = ptr;
+			value = abbrev_branch(value);
 			while (space) {
 				char *merge;
 				merge = xstrndup(value, space - value);
 				path_list_append(merge, &info->merge);
-				ptr = skip_prefix(space + 1, "refs/heads/");
-				if (ptr)
-					value = ptr;
-				else
-					value = space + 1;
+				value = abbrev_branch(space + 1);
 				space = strchr(value, ' ');
 			}
 			path_list_append(xstrdup(value), &info->merge);
@@ -219,12 +222,7 @@ static int handle_one_branch(const char *refname,
 	refspec.dst = (char *)refname;
 	if (!remote_find_tracking(states->remote, &refspec)) {
 		struct path_list_item *item;
-		const char *name, *ptr;
-		ptr = skip_prefix(refspec.src, "refs/heads/");
-		if (ptr)
-			name = ptr;
-		else
-			name = refspec.src;
+		const char *name = abbrev_branch(refspec.src);
 		/* symbolic refs pointing nowhere were handled already */
 		if ((flags & REF_ISSYMREF) ||
 				unsorted_path_list_has_path(&states->tracked,
@@ -253,7 +251,6 @@ static int get_ref_states(const struct ref *ref, struct ref_states *states)
 		struct path_list *target = &states->tracked;
 		unsigned char sha1[20];
 		void *util = NULL;
-		const char *ptr;
 
 		if (!ref->peer_ref || read_ref(ref->peer_ref->name, sha1))
 			target = &states->new;
@@ -262,10 +259,7 @@ static int get_ref_states(const struct ref *ref, struct ref_states *states)
 			if (hashcmp(sha1, ref->new_sha1))
 				util = &states;
 		}
-		ptr = skip_prefix(ref->name, "refs/heads/");
-		if (!ptr)
-			ptr = ref->name;
-		path_list_append(ptr, target)->util = util;
+		path_list_append(abbrev_branch(ref->name), target)->util = util;
 	}
 	free_refs(fetch_map);
 
@@ -460,10 +454,8 @@ static int append_ref_to_tracked_list(const char *refname,
 
 	memset(&refspec, 0, sizeof(refspec));
 	refspec.dst = (char *)refname;
-	if (!remote_find_tracking(states->remote, &refspec)) {
-		path_list_append(skip_prefix(refspec.src, "refs/heads/"),
-			&states->tracked);
-	}
+	if (!remote_find_tracking(states->remote, &refspec))
+		path_list_append(abbrev_branch(refspec.src), &states->tracked);
 
 	return 0;
 }
@@ -530,15 +522,10 @@ static int show(int argc, const char **argv)
 					"es" : "");
 			for (i = 0; i < states.remote->push_refspec_nr; i++) {
 				struct refspec *spec = states.remote->push + i;
-				const char *p = "", *q = "";
-				if (spec->src)
-					p = skip_prefix(spec->src, "refs/heads/");
-				if (spec->dst)
-					q = skip_prefix(spec->dst, "refs/heads/");
 				printf(" %s%s%s%s", spec->force ? "+" : "",
-					p ? p : spec->src,
-					spec->dst ? ":" : "",
-					q ? q : spec->dst);
+				       abbrev_branch(spec->src),
+				       spec->dst ? ":" : "",
+				       spec->dst ? abbrev_branch(spec->dst) : "");
 			}
 			printf("\n");
 		}
@@ -588,7 +575,7 @@ static int prune(int argc, const char **argv)
 				result |= delete_ref(refname, NULL);
 
 			printf(" * [%s] %s\n", dry_run ? "would prune" : "pruned",
-			       skip_prefix(refname, "refs/remotes/"));
+			       abbrev_ref(refname, "refs/remotes/"));
 		}
 
 		/* NEEDSWORK: free remote */
SungHyun Nam· Jul 18, 2008, 05:44 UTC · re: Junio C Hamano · lore

Re: git-remote SEGV on t5505 test.

Junio C Hamano wrote:
Show 6 quoted lines
> SungHyun Nam <namsh@posdata.co.kr> writes:
> 
>> And the 'skip_prefix()' returns NULL in this case.
>> (The old skip_prefix() never returns NULL).
> 
> Thanks.  Something like this?

With this patch, 'make test' passed all the tests successfully. I ran test with GIT_SKIP_TESTS="t90?? t91?? t92?? t94??".

Regards, namsh

P.S I skipped svn/cvs test because I don't need them (And no
     svn/cvs installed).  But, I skipped send-email test because it
     fails.  I thought I don't need to use send-email.
     And now I check what caused fail.  The problem was this test
     script generates 'fake.sendmail' which uses '/bin/sh'.
     Is it possible that we can use 'SHELL_PATH' here?  Or could
     you apply the patch below?
diff --git a/t/t9001-send-email.sh b/t/t9001-send-email.sh
index de5b980..0320aa1 100755
--- a/t/t9001-send-email.sh
+++ b/t/t9001-send-email.sh
@@ -16,7 +16,7 @@ test_expect_success \
      '(echo "#!/bin/sh"
        echo shift
        echo output=1
-      echo "while test -f commandline\$output; do
output=\$((\$output+1)); done"
+      echo "while test -f commandline\$output; do output=\`expr
\$output + 1\`; done"
        echo for a
        echo do
        echo "  echo \"!\$a!\""
Junio C Hamano· Jul 18, 2008, 06:18 UTC · re: SungHyun Nam · lore

Re* git-remote SEGV on t5505 test.

SungHyun Nam <namsh@posdata.co.kr> writes:
>     Is it possible that we can use 'SHELL_PATH' here?

It is not just possible but we really should. There are other test scripts that use hardcoded /bin/sh, but by setting SHELL_PATH the user is already telling us that what the vendor has in /bin/sh isn't adequately POSIX enough, and we really should try to honor that.

"git grep -n /bin/sh t/t*sh | grep -v ':1:#!'" would tell you which ones are suspect.

---
 t/t9001-send-email.sh |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/t/t9001-send-email.sh b/t/t9001-send-email.sh
index de5b980..1c857cf 100755
--- a/t/t9001-send-email.sh
+++ b/t/t9001-send-email.sh
@@ -13,7 +13,7 @@ test_expect_success \
 
 test_expect_success \
     'Setup helper tool' \
-    '(echo "#!/bin/sh"
+    '(echo "#!$SHELL_PATH"
       echo shift
       echo output=1
       echo "while test -f commandline\$output; do output=\$((\$output+1)); done"
@@ -138,7 +138,7 @@ test_expect_success 'Valid In-Reply-To when prompting' '
 '
 
 test_expect_success 'setup fake editor' '
-	(echo "#!/bin/sh" &&
+	(echo "#!$SHELL_PATH" &&
 	 echo "echo fake edit >>\"\$1\""
 	) >fake-editor &&
 	chmod +x fake-editor
@@ -235,7 +235,7 @@ test_expect_success 'sendemail.cc unset' '
 
 test_expect_success '--compose adds MIME for utf8 body' '
 	clean_fake_sendmail &&
-	(echo "#!/bin/sh" &&
+	(echo "#!$SHELL_PATH" &&
 	 echo "echo utf8 body: àéìöú >>\"\$1\""
 	) >fake-editor-utf8 &&
 	chmod +x fake-editor-utf8 &&
@@ -254,7 +254,7 @@ test_expect_success '--compose adds MIME for utf8 body' '
 
 test_expect_success '--compose respects user mime type' '
 	clean_fake_sendmail &&
-	(echo "#!/bin/sh" &&
+	(echo "#!$SHELL_PATH" &&
 	 echo "(echo MIME-Version: 1.0"
 	 echo " echo Content-Type: text/plain\\; charset=iso-8859-1"
 	 echo " echo Content-Transfer-Encoding: 8bit"
Junio C Hamano· Jul 18, 2008, 06:26 UTC · re: Junio C Hamano · lore

Re: Re* git-remote SEGV on t5505 test.

Junio C Hamano <gitster@pobox.com> writes:
Show 11 quoted lines
> SungHyun Nam <namsh@posdata.co.kr> writes:
>
>>     Is it possible that we can use 'SHELL_PATH' here?
>
> It is not just possible but we really should.  There are other test
> scripts that use hardcoded /bin/sh, but by setting SHELL_PATH the user is
> already telling us that what the vendor has in /bin/sh isn't adequately
> POSIX enough, and we really should try to honor that.
>
> "git grep -n /bin/sh t/t*sh | grep -v ':1:#!'" would tell you which ones
> are suspect.

SungHyun, I did not test this patch myself (all my shells grok $() command substitutions), so I won't be committing this until/unless I see a "tested on system X and works fine".

SungHyun Nam· Jul 18, 2008, 09:07 UTC · re: Junio C Hamano · lore

Re: Re* git-remote SEGV on t5505 test.

Junio C Hamano wrote:
Show 16 quoted lines
> Junio C Hamano <gitster@pobox.com> writes:
> 
>> SungHyun Nam <namsh@posdata.co.kr> writes:
>>
>>>     Is it possible that we can use 'SHELL_PATH' here?
>> It is not just possible but we really should.  There are other test
>> scripts that use hardcoded /bin/sh, but by setting SHELL_PATH the user is
>> already telling us that what the vendor has in /bin/sh isn't adequately
>> POSIX enough, and we really should try to honor that.
>>
>> "git grep -n /bin/sh t/t*sh | grep -v ':1:#!'" would tell you which ones
>> are suspect.
> 
> SungHyun, I did not test this patch myself (all my shells grok $() command
> substitutions), so I won't be committing this until/unless I see a "tested
> on system X and works fine".
I tested it on Solaris and works fine. ^^

$ uname -sro SunOS 5.9 Solaris

$ SHELL_PATH=/bin/bash bash ./t9001-send-email.sh
*   ok 1: prepare reference tree
*   ok 2: Setup helper tool
*   ok 3: Extract patches
*   ok 4: Send patches
*   ok 5: Verify commandline
*   ok 6: Show all headers
*   ok 7: reject long lines
*   ok 8: no patch was sent
*   ok 9: allow long lines with --no-validate
*   ok 10: Invalid In-Reply-To
*   ok 11: Valid In-Reply-To when prompting
*   ok 12: setup fake editor
*   ok 13: --compose works
*   ok 14: first message is compose text
*   ok 15: second message is patch
*   ok 16: sendemail.cc set
*   ok 17: sendemail.cc unset
*   ok 18: --compose adds MIME for utf8 body
*   ok 19: --compose respects user mime type
*   ok 20: --compose adds MIME for utf8 subject
* passed all 20 test(s)
Junio C Hamano· Jul 18, 2008, 09:13 UTC · re: SungHyun Nam · lore

Re: Re* git-remote SEGV on t5505 test.

SungHyun Nam <namsh@posdata.co.kr> writes:
Show 5 quoted lines
>> SungHyun, I did not test this patch myself (all my shells grok $() command
>> substitutions), so I won't be committing this until/unless I see a "tested
>> on system X and works fine".
>
> I tested it on Solaris and works fine. ^^
Thanks.

← back to recent threads