Make GitPackCache include ObjectType #942
Conversation
Otherwise, we would have I/O operations for every request for a wrong objectType. By having it as part of the value we can use the cache and aquire the information that the objectType does not match from the cache.
georg-jung
force-pushed
the
fix/cache-tryopen-objecttype
branch
from
3e57658
to
820e843
Compare
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
Sorry it took so long for me to get to this, @georg-jung. |
|
So weird.... I've re-queued the build 3 times and the test process crashes every time. Yet several other builds today passed the first time, so there seems to be something uniquely about this PR that causes a test crash, but it sure seems safe enough. I'm still looking into it. |
|
Thanks for taking a look! Back then, my pipeline ran. I started a run on my ADO on this PR's branch here: https://dev.azure.com/georg-jung/Nerdbank.GitVersioning/_build/results?buildId=1871&view=results I also rebased the PR's contents in a second branch to be on-par with latest main and started a run on it too: https://dev.azure.com/georg-jung/Nerdbank.GitVersioning/_build/results?buildId=1872&view=results Both are still in progress as of writing. (Well the pipeline did/does fail, but that's because it can't publish packages to your CI feed; the tests ran successfully) |
|
All the test steps seem to work in my ADO, for the rebased version as well as the original one. Really strange. |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
See #876 (comment)
This adds caching of the determined objectType alongside the object's content. While the approach in d0aa014 seems a bit more straight forward than the one in fafa498, I think it is preferrable to have the contentType as part of the cached value instead of the cache key. Otherwise, repeated requests with a mismatching objectType will not profit from caching.
I thought about adding proper
TryGetObjectimplementations as part of this. This turned out to be quite hard to get right though. The topics aren't too closely related too so I think they work as seperate PRs. Because these cache improvements should be sufficient to make #876 work, I decided to split this up and postpone the properTryGetObjectimplementations.