* Refactor tests into "groups", each with its own filter (not/only/requires) to select which providers are appropriate.
* Test driver code is now a lot more simple and clear.
* Add support for not(), only(), and requires() as a way to select/reject providers for a test.
* Add docs explaining how to add tests
* Logging messages are much cleaner now, especially when tests are skipped.
* -start and -end now refer to test groups, not individual tests. Log messages list the group numbers clearly.
* Add stringer for Capabilities
* Change the order of the tests so that simple tests are first
* Removed knownFailures from providers.json
* fmtjson providers.json
Co-authored-by: Tom Limoncelli <tlimoncelli@stackoverflow.com>
The test 49 was changing the fingerprint type of a SSHFP DNS RR,
but not the fingerprint hash. Changing the type of hash function
changes the hash result size (there's only two defined hash type of
different size in the RFC, SHA-1 and SHA-256).
This was failing with the OVH provider, because OVH is validating
the hash sizes of any created SSHFP RR and was finding an inconsistency.
Since there's already the test 50 which is both changing the type
and the fingerprint, the functionality is still covered and we
can remove test 49.
Commit 87ad01d added the very useful `get-zones` command, which
requires providers to implement a new method `GetZoneRecords`.
This changes make the OVH provider support this.
* Update Exoscale provider
* Update vendor folder for exoscale/egoscale v0.23.0
* Fix typos and SRV parsing
* Add . at the end of SRV records for exoscale
* Tests: ensure provider capabilities are checked
Adds test: `TestCapabilitiesAreFiltered`
We have a number of records and pseudo-records which in theory can only
be used with a given provider if that provider indicates support. In
practice, we've been missing the checks for that support and have been
passing the records down anyway. The advice comment in the
providers/capabilities.go file to edit `checkProviderCapabilities()` has
not been reliably followed.
We need an internal self-consistency test. The constants are not
directly exported or enumerable based solely on the package interfaces
at run-time, but with source access for a test suite, we can use the
`go/ast` and related interfaces to examine the code, extract all the
constants from a given package, figure out which ones we want to be
handled, and then insist that they're handled.
Before my recent work, we only checked:
ALIAS PTR SRV CAA TLSA
After this commit, we check:
ALIAS AUTODNSSEC CAA NAPTR PTR R53_ALIAS SSHFP SRV TLSA
I've added `AUTODNSSEC` as a new feature; `SSHFP` and `PTR` were caught
in other recent commits from me; implementing this test caused me to
have to add `NAPTR` and `R53_ALIAS`. I whitelist `CanUseTXTMulti` as a
special-case.
This should prevent regressions. We will probably want to post publicly
to warn people that if they're using SSHFP/PTR/NAPTR/R53_ALIAS then they
should check the feature matrix and if they don't see their provider
listed, to report is as "hey that actually works" so we can update the
provider flags. Bonus: our feature matrix will suddenly be more
accurate.
* Add comments/docs for capabilities authors
* fixup!
* fixup!
* The old algorithm was very complex for no good reason.
* The new algorithm is simply: Use yymmdd00 or (previous serial number +1) whichever is bigger.
* github.com/miekg/dns
* Greatly simplify the logic for handling serial numbers. Related code was all over the place. Now it is abstracted into one testable method makeSoa. This simplifies code in many other places.
* Update docs/_providers/bind.md: Edit old text. Add SOA description.
* SOA records are now treated like any other record internally. You still can't specify them in dnsconfig.js, but that's by design.
* The URL for issue 491 was wrong in many places
* BIND: Clarify GENERATE_ZONEFILE message
There's a philosophy issue here around what is the Bind output meant to
do. Since AFAIK we're not integrating into Bind's catalog zones or the
like, we're just targeting the zonefiles, we're not in a position to do
_anything_ relating to registrar options such as setting up DS glue.
So at one level, enabling AutoDNSSEC for Bind is a lie. But without
this, folks can't target a Bind zone as a secondary provider for their
domain, to get debug dumps of the zone output, because the checks for
"Can" block it. So I think this commit achieves a happy compromise: we
write a comment into the Bind zonefile, indicating that DNSSEC was
requested.
Actually: we add support for arbitrary zone comments to be written into
a zonefile via a slightly ugly "can be `nil`" parameter. We then write
in a generation timestamp comment, and if AutoDNSSEC was requested we
then write that in too.
* Add AUTODNSSEC, implement for DNSimple
There are two models for DNSSEC in DNS management: either dnscontrol
manages all the DNSSEC records and has to be invoked regularly for
re-signing, or the provider manages DNSSEC automatically and dnscontrol
is not involved beyond saying "yeah, do that".
This implements the latter, as a model, and for one provider.
Potentially we could tune/configure options for DNSSEC such as
algorithm, but DNSimple don't expose that API so I haven't implemented
it.
This minimal model should be something which maps into other providers
cleanly.
* Fix missing CanAutoDNSSEC on provider
* Validation fix for master broken
This is broken in master and causing Travis in my branch to fail. The
validation tool runs with `gofmt -s` to require "simplify", and so
rejects an ignored second bound variable to range iteration.
* Correct wire in the AUTODNSSEC validation step
* dnsimple: bug-fix SSHFP, add multi TXT support
The default logic for encoding SSHFP records was dropping the key and
hash algorithms and just posting the content, the `Can` check didn't
stop attempts to use SSHFP. So, implement SSHFP support.
DNSimple support multiple DNS strings in a TXT record, by representing
the payload as quoted strings already. This doesn't appear to be
documented, but it does actually work.
* Update docs support matrix too
* fix go fmt missing upstream
Tests failing on my branch for something broken upstream
* fix typo in error message
* Update to latest dnsimple-go
* Implement GetZoneRecords
* Better naming
* Return NS records in GetZoneRecords
* Be clearer with the comment.
As an employee I confirm this is exactly how this works. No guessing needed.
* Respect that Puncycode encoding can blow up
* Implement ListZones and the ZoneLister Interface
* Categorize DNSIMPLE
* Update docs with go generate
* vendor modules
* Don't store intermediary Zone data
* Add GetZoneRecords to DNSProvider interface
* dnscontrol now uses ufave/cli/v2
* NEW: get-zones.md
* HasRecordTypeName should be a method on models.Records not models.DomainConfig
* Implement BIND's GetZoneRecords
* new WriteZoneFile implemented
* go mod vendor
* Update docs to use get-zone instead of convertzone
* Add CanGetZone capability and update all providers.
* Get all zones for a provider at once (#626)
* implement GetZoneRecords for cloudflare
* munge cloudflare ttls
* Implement GetZoneRecords for cloudflare (#625)
Co-authored-by: Craig Peterson <192540+captncraig@users.noreply.github.com>
