Make the command for reloading code more robust #84
Conversation
|
Is it really safe to delete entries from sys.modules like that? I'm not entirely clear on what you're doing here or whether that part's necessary. Could you explain? |
|
This was in the old version already in Aenea as well. It's necessary to ensure that all the modules get unloaded before natlink reloads them. Otherwise, if you have some file that isn't a grammar file, it won't get reloaded. This is what allows files inside Aenea itself to be reloaded for example. |
|
As for is it safe, I suspect that there is some module that refers to something that isn't getting reloaded so it still has the old (dead) reference. Either that, or it's something to do with the ordering. |
|
Fair enough. This is no worse than back when we didn't support any reload at all -- worst case is you have to restart Dragon. |
Make the command for reloading code more robust
|
Incidentally, if you know a better way, it would be great to see it. Back when I developed this originally, I also tried several others which turned out to be even worse. |
|
afraid not. If I knew a better way, I'd have implemented it. I thought about trying trickery like this, but truth be told my opinion was (and is) that this setup is flaky enough without reload hackery. I'd rather just restart Dragon and know the issue I'm having isn't caused by reload hacks, even if it does lengthen the dev cycle a bit. I don't mind offering this command because it's strictly better than not having it. |
I've been using this version of the reload_code() function and have found that it causes #77 to occur less frequently. Now it only occurs when I add a new grammar file in the MacroSystem directory. It always occurs when that happens however. I'm still not sure what causes that particular issue...