|
wojciech.kapcia@tigase.net opened 4 years ago
|
|
We could change to a non-WARNING level. I'm surprised that it was thrown in here as |
|
|
Log excerpt: payload: |
|
|
After investigation it occurred that code was running OK until JIT (Just In Time) compiler kicked in. For some reason it decided that calling constructor which may throw exceptions is not important as returned value was not used. I do not know why it ignored that exceptions were not caught and could change how code will behave. I've modified code, to ensure that returned value is used and to return value from that method based on created item, to make sure that JIT will not try to optimize that anymore. I stil think that this issue was caused by JIT error as it altered flow of the program which it should not. |
|
|
I've fixed the database (removed invalid channel) and updated installation. I think that this should be solved now. |
|
|
Just a small, sidenode comment:
While it was a bit surprising behaviour, I don't think it's that unexpected or wrong - JIT in itself may remove "dead code", ie parts of code that doesn't yield result / case any side effects. That particular block of code wasn't returning nor modifying anything hence "doing nothing" - even you yourself added pre-emptively comment: "// this line is required as it validates it configuration is correct!" to avoid it's erroneous removal. I think that the problem boiled down to "controlling by exceptions", and because exceptions are for "exceptional situations" and not executing particular piece of could would/should not yield any exceptional results hence removal by JIT. Now, I previously suggested (if you deem it as bug) submit it as such to openjdk bugtrack, but considering above I wouldn't consider it as bug per se. I did a small research to see if there are some linterns/validators/etc but there is nothing that would explicitly warning in this case (PMD was the main candidate). What Idea already signals ("unused variable") seems like the closest thing. Bottomline, we'd have to be conscious about cases like that. |
|
|
I think this is now working properly. Let's wait for a while to confirm that. |
|
|
Seems to be working fine. |
| Type |
Bug
|
| Priority |
Normal
|
| Assignee | |
| Version |
tigase-server-8.2.0
|
| Spent time |
5h 45m
|
Issue Votes (0)
Watchers (0)
It seems that this is OK - we just don't support it and that's it. Adjust log level to non-WARNING?
I checked the code and I was wondering if in
tigase.mix.model.MixRepository#updateChannelConfigurationwe shouldn't return error packet to the client?