Add support for new client reset functionality #778
Conversation
realm-ci
force-pushed
the
tg/client-reset
branch
2 times, most recently
from
6c9917e to
3a1ab90
Compare
bmunkholm
changed the title
|
@tgoyne Is this ready or is anything missing? |
realm-ci
force-pushed
the
tg/client-reset
branch
from
c92f968 to
faf4cdf
Compare
|
Status: Works as intended for Cocoa needs. Likely needs additional changes to support other SDKs. @cmelchior |
tgoyne
force-pushed
the
tg/client-reset
branch
4 times, most recently
from
31d287f to
eeb51ee
Compare
tgoyne
changed the title
realm-ci
force-pushed
the
tg/client-reset
branch
from
0e9ec87 to
2791266
Compare
realm-ci
force-pushed
the
tg/client-reset
branch
from
2791266 to
9c09eb0
Compare
|
This is now up-to-date with latest master/core/sync. IIRC there were more things I wanted to test related to it, but nothing that would actually block merging this. |
| @@ -181,9 +183,7 @@ stage('unit-tests') { | |||
|
|
|||
| stage('publish') { | |||
| node('docker') { | |||
| publishReport('linux') | |||
There was a problem hiding this comment.
Building with coverage enabled greatly increases build time, and because we don't try to test any of the non-sync fallback code paths, the non-sync coverage reports don't tell us anything useful that isn't already covered by the sync-enabled coverage reports.
| std::map<std::string, std::string> custom_http_headers; | ||
|
|
||
| util::Optional<std::string> url_prefix = none; | ||
|
|
||
| // The name of the directory which Realms should be backed up to following | ||
| // a client reset | ||
| util::Optional<std::string> recovery_directory = none; | ||
| util::Optional<std::string> recovery_directory; | ||
| ClientResyncMode client_resync_mode = ClientResyncMode::Recover; |
There was a problem hiding this comment.
There is no doubt that this is the sensible default, but should it be considered a breaking change for the SDK's? In the strictest sense, I think it is, but for almost all practical purposes this is better.
In any case, the bindings can override this to get the default behavior they want.
There was a problem hiding this comment.
The argument was that we're turning an error scenario into a non-error rather than changing the behavior of correct things. The assumption is that in normal usage when running against a perfect server with infinite resource a client reset error will never occur, so client resync just lets us pretend that we're in that scenario more often than we actually are.
3.7 has a broken pthreads check that incorrectly concludes that everything needed is supplied by libc.
tgoyne
force-pushed
the
tg/client-reset
branch
from
9c09eb0 to
a458c34
Compare
Currently actually just async open.