Tv
With some debugging added, and running a tight loop, I've seen a `syscall.Write` to the fuse fd fail with `ENOENT`. Quick skim of kernel code says this happens when e.g....
Directories are supposed to get a separate opcode, `opFsyncdir`. https://github.com/bazillion/fuse/blob/00c7ac3b73bbaff171daab245982cac04456de72/fuse.go#L730 ``` --- FAIL: TestFsyncDir (0.12s) mounted.go:119: FUSE: ID=0x1 Init {MaxReadahead:0 Flags:InitBigWrites MaxWrite:131072} mounted.go:119: FUSE: ID=0x2 Open {Handle:1 Flags:0} serve_test.go:1309: fsyncDir...
Putting a `time.Sleep(100 * time.Millisecond)` after the mount subprocess returns seems to make many more tests pass, but I refuse to resort to racy time-based criteria. How do we know...
The test completes just fine, but for some reason the `cat` subprocess doesn't seem to be waken up from IO wait until after 60 seconds. This may be a FreeBSD...
This makes reliable persistency impossible. This may be a FreeBSD FUSE limitation. Needs a platform expert.
This was triggered either by a `fusermount -u` or `echo 1 >/sys/fs/fuse/connections/XXX/abort`, not sure which. ``` 2014/08/06 11:51:52 FUSE: short kernel write: written=-1/24 error="no such file or directory" stack= goroutine...
Given that the connection pool is inside Transport, to have multiple streams on the same connection one must use the same Transport. However, to make requests with different priorities, the...