Correct hostent_addr h_addrtype typespec #5412
Conversation
|
|
varnerac
force-pushed
the
hostent_spec_fix
branch
2 times, most recently
from
56464be
to
4153cd2
Compare
varnerac
force-pushed
the
hostent_spec_fix
branch
from
4153cd2
to
dd3a664
Compare
| h_addrtype :: inet | inet6 | caa | cname | gid | hinfo | ns | mb | md | mg | ||
| | mf | minfo | mx | naptr | null | ptr | soa | spf | srv | txt | ||
| | uid | uinfo | unspec | uri | wks, %% host address type |
There was a problem hiding this comment.
I suggest this should be:
h_addrtype :: inet | inet6 | inet_res:rr_type(),and that inet_res:rr_type() should be exported.
This would be a bit incorrect since a and aaaa today cannot be in #hostent.h_addrtype, but future safer since it uses the type that the rest of inet_res uses for DNS record types, and future additions will be only to inet_res:rr_type().
There was a problem hiding this comment.
Would it perhaps make sense to have a inet_res:dns_sockaddr_record that could be exported and make the type h_addrtype :: inet | inet6 | inet_res:dns_sockaddr_record() and make inet_res:rr_type be
rr_type :: a | aaaa | dns_sockaddr_record() or for even more clarity
dns_ip_record:: a | aaaa and rr_type :: dns_ip_record() | dns_sockaddr_record() It could also be more clear to call rr_type dns_record_type , longer but much easier to understand!
There was a problem hiding this comment.
inet_res:rr_type() should probably rather be named inet_res:dns_rr_type() similar to inet_res:dns_name() and inet_res:dns_class()...
Other than that I think that subdividing rr_type() would be artificial since it simply specifies the flat set of RR types.
The name dns_sockaddr_record is misleading since:
- It is not a record - it is a record type or rather a Resource Record type, which within DNS is abbreviated RR, hence "RR Type". So "record" should be "resource_record_type" or rather an artificial subset thereof.
- It has very little to do with a socket address ("sockaddr"). Normally, when a
#hostentrecord is used, the fieldh_addrtypecontainsinetorinet6, which both are socket address types, a.k.a address families, so there is a connection between this field and a socket address type. But when the field is abused byinet_res:getbyname/2,3, it contains a DNS RR Type, and the content of a RR need not be an address - it can be a CNAME string, a TXT string and whatnot.
The problem is inet_res:getbyname/2,3 that returns the result as a #hostent record. That return format is only suitable for RR types a and aaaa that have the translations into socket address types inet and inet6. For all other RR types the #hostent record return value is unsuitable and is abused.
So inet_res:getbyname/2,3 creates the type problem and I see no good solution. The least bad I can think of is to type specify #hostent.h_addrtype to be the union of the address families that can be returned from gethosbyname(), i.e inet | inet6, and the DNS RR types, i.e inet_res:rr_type(). That two RR types are translated into socket address types by the function so they can never occur in the result, I think is a minor problem.
But, maybe we should change inet_res:rr_type() into inet_res:dns_rr_type() before exporting it...
There was a problem hiding this comment.
For the record, I see I misunderstood. We have discussed this internally and will make a different solution to try to make the best of it.
|
I have created an alternative PR #5803 that instead updates the |
…c_fix/PR-5412/OTP-17986 Correct typespec for inet_res:getbyname/2,3 (PR #5412, OTP-17986)
|
Fixed by PR #5803 |
Fixes a broken spec for the
hostenth_addrtypespec. It covers the results from this function head.