Use a single QNetworkAccessManager instance for all our HTTP jobs, and also add a simple diskcache to that qnam. Further ensure there is only a single qnam for the entire application using kns' http jobs, across all threads (lock when accessing the qnam). Without this, we are liable to end up creating and destroying a great many qnam instances, which certainly is something to try and avoid.
CCBUG: 379193
Using Discover, which creates a great many of these jobs, we have much less network traffic this way.
| Automatic diff as part of commit; lint not applicable. |
| Automatic diff as part of commit; unit tests not applicable. |
| src/core/jobs/httpworker.cpp | ||
|---|---|---|
| 33 | How about getting a class with these 3 attributes? Looks like we're juggling here... | |
| src/core/jobs/httpworker.cpp | ||
|---|---|---|
| 33 | You know, that sounds like a good plan. Updated diff incoming :) | |
Simpler logic, with the global qnam (etc) stored in a locally defined class, so we only have the one global static, and also allowing the qnam access to be more pleasantly locked.
| src/core/jobs/httpworker.cpp | ||
|---|---|---|
| 41 | 0.1% of the partition size is a rather arbitrary value, no? It could go from something very tiny to something really big... On my 470GB partition this would lead to a 470MB cache for knewstuff, that's maybe a bit much, given the average size for knewstuff stuff? Maybe a qMin() call with a maximum value would be useful? | |
| 47 | remove space before ')' | |
| 144 | remove spaces inside ( ... ) | |
| 151 | remove spaces inside ( ... ) Also note that you added debugging values to a qCInfo message which is user-visible even in non debug builds. I think it might be better to revert this or split it out into a qCDebug call. | |
| 164 | should probably be qCDebug() | |
| src/core/jobs/httpworker.cpp | ||
|---|---|---|
| 41 | Good point yes, it's supposed to be helpful, not take over the system. I'll set a max of 50 megs, should be fine for most uses. | |
A bit of naming cleanup, and a very silly miscalculation because things count in bytes rather than some arbitrary random multiple of them