Catherine
Catherine
Confirmed to be still an issue on 1.70.
> As a temporary workaround while we investigate this, I suggest checking out the new app-based split tunneling feature we added in 1.70. You can exclude specific apps from using...
I can confirm that excluding VoIP apps from Tailscale fixes the issue.
It's unclear to me how to provide the desired semantics (`st1.st_ino == st2.st_ino` iff `isSameObject(f1, f2)`), or, in fact, any reasonable semantics that doesn't make private assumptions about the exact...
On reflection, I think the intended behavior here is to maintain a mapping from `metadataHash` to "potential files" (i.e. both files corresponding to file descriptors, and files that could be...
On more reflection, since `metadataHash` changes if the file is modified, it can't be used to index into such a table, so we're back to `fd_filestat_get` can't be sound. How...
Thank you. I've been feeling not the best lately, and this has directly impacted my ability to reason on really intricate topics like this one. I'll keep this issue on...
> Allocations are very carefully managed and such, so doing something fancy with `std_{dev,ino}` when `metadata_hash` is zero is probably going to be difficult. In this case, given how p1...
I think a backstop solution that prevents the most egregious consequences (e.g. Clang considering every input file as the same) is to use a monotonically increasing global counter for filling...
What about `fstat`? I guess `st_dev` would be the fd and `st_ino` would be the hash of a filename, if any, opened off that fd.