You say it requires a luajit-launcher bump but you didn't include it?

@Frenzie I thought Koreader's downloading latest master luajit.
Howto change version?

You go into the dir of the submodule and run git pull. Then you can commit that change.

Guys, how do you build working apk for android?

Do you have to add any parameters to the build?

I use ./kodev release android (as it's in https://github.com/koreader/koreader#for-android-devices)
APK file is successfully created.

I'm able to install it, but such created application never works on my phone.
It just display black screen (with circle a.k.a "please wait") & freezes.

I'm building it on Ubuntu 16.04 LTS.

Sorry, that's my fault (but I mostly blame Android itself). There is no such thing as an Android build that works on my phone, so I can only test if it builds. :-(

If you manually use base from just prior to koreader/koreader-base#482 and use the Docker Ubuntu 14.04 box it should work but I think/hope @chrox is really close to a proper fix.

Is there a docker container I can download? - - - edit : ok. I thought I need special koreader setup on Ubuntu, but it seems you're saying I just need 14.04 Ubuntu version.
But I would prefer to have had it fixed.

Hopefully koreader/koreader-base#482 will be finished soon as it's a blocker.

There are Docker and Vagrant containers in https://github.com/koreader/virdevenv

You'll need to make sure base is at the right commit.

Help needed. There's a problem:

I/KOReader( 2067): DEBUG frontend/ui/elements/screen_fullscreen_menu_table.lua: variable disabled= true
E/luajit-launcher( 2067): Failed to run script: [string "android.lua"]:1354: attempt to concatenate local 'fullscreen' (a nil value)
V/threaded_app( 2067): android_app_destroy!

Koreader frontend/ui/elements/screen_fullscreen_menu_table.lua :

   local disabled = G_reader_settings:readSetting("disable_fullscreen") ~= false
   G_reader_settings:saveSetting("disable_fullscreen", not disabled)
   logger.dbg("frontend/ui/elements/screen_fullscreen_menu_table.lua: variable disabled=", disabled)
   android.setFullscreen(disabled)

Lua-jit launcher, android.lua lines 1353 and 1354:

android.setFullscreen = function(fullscreen)
android.LOGI("set fullscreen "..fullscreen)

I guess that variable ("disabled") exist in Koreader (and equals "true"), but then when it's transfered to Lua-jit launcher it's missing (it's "nil"). Right?
Do I need to somehow translate this Koreader (boolean) "true" to Lua-jit boolean ? Or is something else the problem?

There's no such thing as a "KOReader boolean". :-P There is of course a possibility that you saved it as a string at some point, but even if you did there's no difference between "true" (string) ~= false and true ~= false. The result would always be true (boolean). Btw, I'm not sure why you'd save a setting right after reading it?

For checking if a key is true or false we have convenience functions like isTrue() and isFalse().

Glancing at the code I notice it's declared as final https://github.com/koreader/android-luajit-launcher/blob/cddd8ab65b78c87ec295ba789f6e92b59e92b58c/src/org/koreader/launcher/MainActivity.java#L106

As "koreader boolean" I thought about converting boolean between lua and Java. But right, this part of Lua-jit is still lua code.

I used Greater:settings to read and store the full screen settings I thought it might be faster than calling android(java) code from lua to get that info. But maybe it's not?and should be replaced?

That Java "final" key word you highlighted just means once you passed the attribute to function it reference cannot be changed inside the function. In simple words function itself cannot change passed boolean. So shouldn't make difference unless there's something wrong with passing lua variable to Java final parameter..

Oh I get what you mean that I "save a setting right after reading it". I'll fix that but the main problem is still to be solved..

I played with Geom:new{ it seems to change the app resolution but not the starting point.

Comparision (hard coded values for tests):
local viewport = Geom:new{x=50, y=100, w=900, h=1600}
below:

with
local viewport = Geom:new{x=0, y=0, w=900, h=1600}
below:

The seems to have the same starting point.

I even tried with negative x & y values but then I had:
E luajit-launcher: Failed to run script: ffi/framebuffer.lua:137: bad viewport

Any ideas?

Sorry, I'm not sure what could be the matter. Have you tried it as part of the Device init instead of trying to set it through a function later? Here's a screenshot of EMULATE_READER_VIEWPORT="{x=100,w=950,y=50,h=790}" ./kodev run -w=1100 -h=1200

I even tried with negative x & y values

Yes, that's to be expected. :-)

The EMULATE_READER_VIEWPORT is used only in generic device.lua in init() method:

local viewport = os.getenv("EMULATE_READER_VIEWPORT")
    if viewport then
        self.viewport = require("ui/geometry"):new(loadstring("return " .. viewport)())

I tried simple :

local viewport = "{x=50, y=100, w=900, h=1600}"
local Device = require("device")
Device.viewport = require("ui/geometry"):new(loadstring("return " .. viewport)())

but it doesn't make any changes to viewport. The same result for code below:
android.viewport = require("ui/geometry"):new(loadstring("return " .. viewport)())

I believe the android.screen:setViewport(viewport) must be used.

But there might be something missing in current implementation?
android.lua has width and height in screen object:

    android.screen = {}
    android.screen.width, android.screen.height =

but I didnt' spot screen.x and y ...

I found setViewport method in the same file I I had exception on for negative x & y (ffi/framebuffer.lua:137)
There's also framebuffer_android.lua which extends above FB :
return require("ffi/framebuffer"):extend(framebuffer)

In framebuffer_android.lua init() method doesn't have x & y; only height & width:
self.bb = BB.new(android.screen.width, android.screen.height)
BB is local BB = require("ffi/blitbuffer")

Searching for vieport in BB gave me result:
function BB_mt.__index:viewport(x, y, w, h)
so here's x & y..

But maybe the data is lost when passed from a lua file to another?
I'm above my ability to track it:(

The EMULATE_READER_VIEWPORT is used only in generic device.lua in init() method:

Sure, but besides being slightly easier to test on the fly it makes no difference. It's the same as just writing local viewport = {whatever you stick in the environment variable}

In framebuffer_android.lua init() method doesn't have x & y:

Right, as shown in my screenshot the viewport affects where on the framebuffer the picture is drawn. Since the problem seems to be that the statusbar overlays KOReader, my thinking is that it doesn't matter if the framebuffer is still the full size. All you need to do is push the KOReader viewport down slightly. I have no idea why that doesn't seem to be working, but I don't think it's related to the particular line you mention.

Testing Android is tremendously annoying.

Anyway, after some general inconvenience I can confirm that it just sticks it in the top left. The value I used for testing was self.viewport = require("ui/geometry"):new{x=50, y=100, w=300, h=500}

PS

sudo apt install openjdk-9-jre-headless openjdk-9-jdk-headless
keytool -genkey -keyalg rsa -keysize 2048 -alias jdkcert2048 #from http://www.grok2.com/blog/2010/09/14/keytool-jarsigner-oddity-on-ubuntu-with-openjdk-version-6/ just fill out some nonsense and a password of 123456
# remove META-INF folder
jarsigner ko.apk jdkcert2048

It's been like a decade since I've done Java and signing jars wasn't required. What's that Android emulator thing?

Yes It's a pain.
Everytime I change code have to execute at least 2 commandline commends (1 more for logcat) + some manual actions (remove old Koreader). And cannot even run debugger, only viewing logs:(

Some time ago I asked for x86 build (#3046) so at least emulation would be faster.
Start time for arm emulator is around 3-5 min on my PC:/

But getting back to main problem:
I created a blocker for this PR #3148, It seems until resolved we cannot process with this feature.

ps: @Frenzie I use ./kodev release android IMHO seems easier that manually singing apk file.
BTW Java 9 is not released yet.

BTW Java 9 is not released yet.

Whatever it is, it's in the repos and I used it. :-P I doubt jarsigner changed much over the years; I simply got myself the stuff I needed with the latest version number.

IMHO seems easier that manually singing apk file.

Compilation tools are useful in general including on my laptop. Gigabytes of Android stuff are not at all (to me). I'd just as lief not have them on my desktop but there I have enough space to spare. :-)

@mwoz123 Do you know anything about Gradle? Google apparently wants everyone to switch to that moving forward, making build.xml incompatible with SDK 26.

@Frenzie I should have gradle as build tool in next work project..
For now I know it's all about json.

Android disabling/enabling fullscreen works.

Changing viewport has been extacted to #3148

Please review & merge.

Looking pretty good. 👍

@Frenzie I wasn't complaining about

local _, android = pcall(require, "android")

but now it does about :

frontend/ui/elements/screen_fullscreen_menu_table.lua:1:7: unused variable 'isAndroid'

can I get back to previous solution?

Then you'd still be upvalueing it, which is bad practice. (It might work but you shouldn't have to parse the code to know what a variable refers to.)

But one should expect it to complain about unused variables, because I suggested to use that variable:

local isAndroid, android = pcall(require, "android")
if not isAndroid then return end

But I suppose it'll do.