GitLab

not really useful at this point anymore

It isn't actually used by the test, but by not setting any db, the default db would've been created (and not cleaned up, making distcheck fail for leaving files around)

note that this is mostly a way to at least get it working, we need to figure out a more permanent solution related to static linking (or stop supporting it in the first place)

The removal of memory_datasrc.cc in b10_auth_SOURCES was premature; zone loading code is still in auth (but memory_datasrc-specific)

Which is now no longer necessary

instead of fighting the API to get empty zones in, actually change the zone in the database and reload

- RAII InMemoryClient's createInstance
- remove dead code
- added more comments

and store it as a scoped ptr

it is no longer necessary to derive it from InMemoryClient, but it can be directly derived from DataSourceClient

See http://bind10.isc.org/ticket/1207#comment:11

The temporary name with the added P is no longer necessary now the original method has been removed.

Made the getZoneCount() method from inmemory client a general datasource client
call (which throws NotImplemented by default), as this does seem useful in other
 cases as well.

Now the work starts to pay off; removed the temporary pointer variable (getInMemoryClientP() still returns a pointer but now directly retrieved from the container), and any direct reference to isc::datasrc::InMemoryClient is removed from au
th_srv (but not from tests as of yet)

Instead of the direct shared_ptr.

Changes needed:
Moved loading of zones in config into the memory_datasrc implementation; at this moment the general API has no method for this (and it may not be necessary; datasources should handle their own config anyway).
This does have one drawback; through the dynamic loading mechanism we currently lose exception type information. So for now specific exceptions such as MasterLoadError end up as 'general' DataSourceErrors. Tests have been updated to reflect this.

Also, I made config handler slightly more lenient; it has hardcoded defaults and will now accept config parts without "class" or "zones"; using .spec -derived values is impractical here with the current config API, but in this case it shouldn't *really* matter; as IN is hardcoded in several places anyway atm.

not directly related to this branch, but the review comment on this point
for the previous change should actually apply to this.

The run_unittests should be linked only if it is compiled. Confirmed on
jabber this was a mistake.

The old version is a result of copy-paste and is wrong.

We have camelCase in many places, but variable names are not one of
those.

previous commit had a mangled tab due to a copy-paste error from the original diff

So that the sqlite3_ds.so is found correctly
This has been acked on jabber, and is tentatively waiting full resolution in ticket #1895

Checking various error modes, like unreadable database, wrong or missing
configuration.

It puts the configuration of the zone (which is loaded from SQLite3
database) and the database at the disposition of the test. Then it asks
it to load the zone and it expects it to load it.

it creates a temporary SQLite3 DB file from a shared template with a specified
stream of zone data.  this will help reduce the number of pre-generated DB
files in the repository (which will be subject to upgrading the schema).

in this test we are only interested in the specific config is valid or not,
so we didn't need the 'errors'. (actually I didn't notice there was this
version of validateConfig() previously).

- added one test case for checking the new "filetype" item is optional.
- also added a test in MemoryDatasrcConfigTest to see whether "filetype" is
  accepted.
- others should be trivial editorial cleanup.

also added a new set of unittests that verify the new spec syntax.

 * Case when no file is specified
 * Case where both origin and file are not specified

NameParserExceptions are now caught and AuthConfigError is thrown.

(except for some "intentionally broken" ones).  these were semi-automatically
convered by b10-dbutil.  "nodiffs" needed to be tweaked by hand after upgrade
for their test.  and all unit/system/lettuce tests now passed.

To make this test workable, I needed to change the returned flag
for the NXRRSET case in MockZoneFinder::find() from setting NSEC_SIGNED
unconditionally to setting any SIGNED flags.  this doesn't affect other
tests.

that way we don't have to expose mutable member variables of `Query`.
now the responsibility of the "inserter" is enlarged, also renamed it
to represent the new job more appropriately.