other clients don't provide streaming as far as i'm aware. they should only run into the redundant piece issue in what's called the "end game", where you double up requests to ensure you can finish quickly. using torrent.Reader you're essentially always in end game mode
@elgatito did you solve your hang? what does litter provide that spew didn't?
@elgatito have you tried the dev branch? i've made breaking changes renaming Config->ClientConfig because it's now recommended to call NewDefaultClientConfig
@anacrolix not yet (for both questions). On some Windows (not all), getting a hang when dumping ClientConfig with DefaultStorage = NewFileStorage
it's a struct, nothing special, so i'm puzzled :)
@anacrolix dev branch is ready to test?
give it a spin
@anacrolix https://github.com/tj/go-disk-buffer new packages arise. interesting to test with good bandwidth and hdd's
that is not exactly what we need, because package uses consequent writing
in our case it should be probably something like a buffer of
[N] chunks, which flushes on interval or N size
i think most of the overhead is actually in the file open and closing
that time I tried to use persistent filehandles it was even worse. system or Go is probably managing all that
if there was an easy answer i would have done it :P
sure
so if there's not, i do the simplest thing
libtorrent implements buffered writer
mmapping is a super simple solution
are you using that?
mmap?
it is available in Elementum, but people say it's not helping when they have poor write speeds
ok
got a profile for the poor write speeds?
pprof profile?
that should be the best one for it. or the /debug/perf when they have poor speeds
it should show a very high average or max
there 's #234, but you see, people are silent
# github.com/syncthing/syncthing/vendor/github.com/syncthing/notify
../../syncthing/syncthing/vendor/github.com/syncthing/notify/watcher_fsevents_cgo.go:10:10: fatal error: 'CoreServices/CoreServices.h' file not found
#include <CoreServices/CoreServices.h>
^
1 error generated.
that is when cross-compiling with cgo for iOS. I feel I will need to move upnp to a local package to avoid pulling all that huge library
it's probably something missing in the xgo build script, because darwin compiles fine (when iOS does not), but still makes more trouble than help :)
hm
what we need is a standalone upnp implementation
i have no time to implement it myself
i'm half way with upnp migration
separate repo outside /torrent ?
Bootstrapping ios-8.1/arm-7...
Compiling for ios-8.1/arm-7...
# github.com/mattn/go-sqlite3
sqlite3-binding.c:31577:6: warning: "gethostuuid() is disabled." [-W#warnings]
Cleaning up Go runtime for ios-8.1/arm-7...
Cleaning up build environment...
awesome. finished moving upnp, it's not that much, when looking closer
they mostly do own logging and override net.dialer to support upnp behind proxy
can you make a PR of it?
done
log messages now all come with
log.Printf(), so spamming a bit
ah, i was thinking maybe make it a standalone repo
elgatito/nat
:P
ha. not sure it makes sense since it's so small
lots of people wish upnp wasn't embedded in syncthing
we're not the first ones to link in
you can make an empty repo so I can pr there :)
lol
so what do you think?
would you prefer not to maintain it?
everyone would prefer not to maintain something :D
but i'm finishing with own repo
will repost the PR
updated the PR
awesome
now the ios builds are fine again
next version will ship for ios, xbox as well