mirror of
https://github.com/fadden/6502bench.git
synced 2024-12-01 22:50:35 +00:00
195c93a94a
Another chapter in the never-ending AppDomain security saga. If a computer goes to sleep while SourceGen is running with a project open, life gets confusing when the system wakes up. The keep-alive timer fires and a ping is sent to the remote AppDomain, successfully. At the same time, the lease expires on the remote side, and the objects are discarded (apparently without bothering to query the ILease object). This failure mode is 100% repeatable. Since we can't prevent sandbox objects from disappearing, we have to detect and recover from the problem. Fortunately we don't keep any necessary state on the plugin side, so we can just tear the whole thing down and recreate it. The various methods in ScriptManager now do a "health check" before making calls into the plugin AppDomain. If the ping attempt fails, the AppDomain is "rebooted" by destroying it and creating a new one, reloading all plugins that were in there before. The plugin binaries *should* still be in the PluginDllCache directory since the ping failure was due to objects being discarded, not AppDomain shutdown, and Windows doesn't let you mess with files that hold executable code. A new "reboot security sandbox" option has been added to the DEBUG menu to facilitate testing. The PluginManager's Ping() method gets called more often, but not to the extent that performance will be affected. This change also adds a finalizer to DisasmProject, since we're relying on it to shut down the ScriptManager, and it's relying on callers to invoke its cleanup function. The finalizer throws an assertion if the cleanup function doesn't get called. (Issue #82) |
||
|---|---|---|
| .. | ||
| DomainManager.cs | ||
| PluginDllCache.cs | ||
| ScriptManager.cs | ||
| Sponsor.cs | ||