The Android port is just about usable. Still, we have some fun todo items to improve it.
high-priority TODO
- ?Android app permission denial on startup
- S3 doesn't work (at least to Internet Archive: "connect: does not exist (connection refused)")
- Get app into Google Play and/or FDroid
TODO
Don't make app initially open terminal + webapp, but go to a page that allows opening the webapp or terminal.
Possibly, switch from running inside terminal app to real standalone app. See https://github.com/neurocyte/android-haskell-activity and https://github.com/neurocyte/foreign-jni.I have seen an assistant thread crash with an interrupted system call when the device went to sleep while it was running. Auto-detect and deal with that somehow.
- Make git stop complaining that "warning: no threads uspport, ignoring --threads"
- git does not support http remotes. To fix, need to port libcurl and allow git to link to it.
- getEnvironment is broken on Android https://github.com/neurocyte/ghc-android/issues/7 and a few places use it. I have some horrible workarounds in place.
- Get local pairing to work. network-multicast and network-info don't currently install.
- Get test suite to pass.
git cloneof a local repo fails on android for some reason. - Make app autostart on boot, optionally. http://stackoverflow.com/questions/1056570/how-to-autostart-an-android-application
- The app should be aware of power status, and avoid expensive background jobs when low on battery or run flat out when plugged in.
- The app should be aware of network status, and avoid expensive data transfers when not on wifi. This may need to be configurable.
- glacier and local pairing are not yet enabled for Android.
- The "Files" link doesn't start a file browser. Should be possible to do on Android via intents, I suppose?
- Adding removable drives would work, but the android app is not in the
appropriate group to write to them.
WRITE_MEDIA_STORAGEpermission needed. Added to AndroidManifest, but did not seem to be used. Googleing for it will find a workaround that needs a rooted device.
mountrequires root and you'll have still the 4gb limit for your image by FAT. some phones (e.g. galaxy nexus) already useext4for/sdcardthough.I played around a bit with GHC and Android today. It isn't really a result, but maybe useful for someone out there.
I have a Debian
chrootenvironment on my Android device (howto: http://www.saurik.com/id/10/). In the Debian box:well, that isn't really static. tell the linker to build a static binary (those are arguments to
ld):now, get this (quite big) binary into the normal android environment, using
adb,SSHDroidor whatever:looking in the source of
System.IOit seems like aniconvissue. So, there's still some dynamic dependency in there... sighThanks Bernard, that's really massively useful. It makes sense -- statically building with libc should work, the Android kernel is still Linux after all.
To get past the iconv problem, I think all you need is part of the
localespackage from your linux system installed on the Android. Probably just a few of the data files from /usr/share/i18n/charmaps//usr/share/i18n/does not exists on my Debian ARM systemhowever,
strace ./armin the debian chroot reveals that some files from/usr/lib/gconv/are loaded:full log: http://wien.tomnetworks.com/strace_arm. Unfortunately, I don't have
stracein the android userland for comparison.Just copying the related
gconvfiles didn't work. I don't have so much time at the moment, I'll investigate further in some days or so.At least, output using
error :: String -> adoes workJust so this does not get lost: For better or for worse, the vanilla Android devices stopped shipping with micro-SD support in 2011 (or 2010 if the Nexus S does not support them either; on sketchy GPRS so not googling around). Most higher-end Android devices ship with at least 8 GiB of on-board Flash storage, some even go up to 64 GiB.
IMHO, this would make it viable to first get git-annex working on Android without regard for FAT.
The obvious advantage is that porting should be easier and quicker.
The obvious downside is that this may mean revisiting some parts of the code later.
-- Richard
Actually, this is something that would be ideal for a poll:
Should FAT-based Android repos be implemented:
Also, as another data point, the FAT-based SD card can be mounted as USB storage by any computer an Android device is connected to whereas the EXT4-based root FS can only be accessed via MTP.
I use Terminal IDE on android, which is 'sort of a shell environment'.
It has bash, git, dropbear ssh, and vim. It is a bit limited, but it does feel like a linux shell.
I'd like to have my phone sync a bunch of files when I'm at home and when I get to work, have them synced to my work machine, deleted off my phone and when I get home, also deleted off my home machine.
This will allow incremental backups to a local folder at home and I can then "mule" my files, sneaker-net style, to work with zero user intervention. It only works if events can be triggered or I can combine a triggering app with a special android intent for both ends of my trip.
One of the points is:
This isn't good for situations where the device is plugged in and charging, but will only be so for a limited time (such as charging during a car journey) as the device would charge slower.
Could this behaviour be configurable too?
I tried to install the version for android 5 on my nexus 5 with the most recent firmware. The initial sync (git clone) succeeds, the git annex stuff fails with "remote origin does not have git-annex installed; setting annex-ignore".
I have the android 4 version working on my tablet for the same ssh remote, so I think the remote is OK.
I can provide full logs of the terminal session if needed (I didn't use the assistant front end).
BTW: this is great software, thanks you so very much! I hope to have it up and running on all my devices soon as a replacement for Google Drive. I happen to care about privacy.