asn1c: segfault on reusing a structure after ASN_STRUCT_FREE_CONTENTS_ONLY().

I’m not sure if this is a bug or working as intended. I didn’t encounter this problem with v0.9.24, and am seeing it now with 0.9.27, but maybe I just got lucky before. I also switched from native-types to wide-types which may have contributed.

When trying to reuse a message structure after using and then “emptying” it, setting attributes using things like OCTET_STRING_fromBuf() can segfault. An example LDAP code fragment looks like this;

msg = calloc(1,sizeof(LDAPMessage_t))
msg->messageID = msgid;
res->protocolOp.present = LDAPMessage__protocolOp_PR_searchResEntry;
SearchResultEntry_t *searchResEntry = &res->protocolOp.choice.searchResEntry;
/* Populate the msg searchResEntry attributes here... */
...
/* Now empty the msg and reuse it for a searchResDone response. */
ASN_STRUCT_FREE_CONTENTS_ONLY(asn_DEF_LDAPMessage, msg);
msg->protocolOp.present = LDAPMessage__protocolOp_PR_searchResDone;
SearchResultDone_t *searchResDone = &res->protocolOp.choice.searchResDone;
/* And now this will segfault... */
OCTET_STRING_fromString(&searchResDone->matchedDN, basedn);

The problem is that OCTET_STRING_fromString() tries to free the existing matchedDN attribute before alloc/set the new one. However, the existing matchedDN attribute contains a garbage pointer left over from the previous searchResEntry message.

I realize this is pretty normal C behavior; freeing things doesn’t re-initialize/zero the pointers, and it’s up to the caller to do that. However, there appears to be some effort in things like OCTET_STRING_free() to leave the free’d structure in a valid initialized state, doing things like zeroing the pointer after free. Digging around other *_free() functions it looks like most others don’t do this.

I think the right thing here is to adhere to C conventions and have ASN_STRUCT_FREE_CONTENTS_ONLY() offer no guarantees that the structure is left in a valid initialized/zeroed state. In that case I think *_free() functions should never expend effort on things like zeroing pointers after free(), because that just masks that this cannot be relied on.