Just found a shocking fact that Android API document on ConsumerIrManger.transmit method is wrong!

KitKat has realised its own Infrared blaster API, which is incompatible with legacy Samsung private API. So I was working on Android Infrared Library to make it adapted automatically on both Samsung private API and Kit Katofficial API.

After I finished the coding according the document, I found the app broke on my Galaxy Note 3 with Kit Kat. It works perfect when running on Jelly Bean.

And I figured out an issue that it takes longer time to transmit the same seqeunce when I upgraded API. (When IR blaster is working, the LED indicator on the phone turns blue. And I found the time of indicator turning blue is significant longer than before.) And my IRRecorder cannot recognize the sequence sent by my phone any longer.

After spent several hours, I figured out the reason. The pattern was encoded in a wrong way. But I’m pretty sure that I strictly followed the API document.

So I get a conculusion that the ConsumerIrManager implementation on Samsung Note 3 is different to what described in Android API document. However I’m not sure the reason is that the Android document is lying or Samsung implemented the driver in a wrong way.

Here is the technical details of the issue and its solution:

IR Command is trasmitted by turnning the IR blaster LED on and off for a certain period of time. So each IR command can be represented by a series of time periods, which indicates how long the led is on or off. The difference between Samsung API and Kit Kat APi is that how the time is mesured.

carrierFrequency The IR carrier frequency in Hertz.
pattern The alternating on/off pattern in microseconds to transmit.

According to the Android Developer Refernece), the time in KitKat is measured in the unit of microseconds.

But for Samsung, the time is mesured by the number of cycles. Take NEC encoding as example, the frequency is 38kHz. So the cycle time T ~= 26us. BIT_MARK is 21 cycles, the period of time is around 26us x 21 ~= 546us.

So ideally, regardless of lead-in and lead-out sequence, to send the code 0xA in NEC encoding, Samsung API needs 21 60 21 21 21 60 21 21; and Kit Kat API needs 560 1600 560 560 560 1600 560 560.

But accroding to my experience, the Android Developer Reference is wrong. Even in KitKat, the time sequence is also measure by number of cycles instead of the number of microseconds!

So to fix the issue, you need some mathmatical work. Here is the conversion formula:

1
2
3
4
5
6
7

n = t / T = t * f / 1000
n: the number of cycles
t: the time in microseconds
T: the cycle time in microseconds
f: the transmitting frequency in Hertz

The Arduino IDE 1.5.3 introduced some new features, such as support latest board Yun, or introduces new libraries and samples. But I found it is too buggy to use.

1. Compile code against Arduino Nano fails due to parameter mcu passed to avrdude is missing.

Reason:
The issue seems caused because the new IDE merged the menu items for Nano board. But for some reason, the configuration haven’t been updated accordingly.

Workaround:
Choose Arduino Duemilanove and Diecimila instead of Arduino Nano. Nano uses same chip as Duemilanove(ATmega328) and Diecimila(ATmega168), but uses a different PCB design. So the binary should be compatible.

2. Compile String(100, DEC) throws ambiguous matching error

Reason:
The issue is caused because the API signature updated to String(*, unsigned char), but constants are still declared as int.

Workaround:
Add force cast DEC, HEX, BIN to byte instead of int.

I’m a newbie to Rails, and my past few projects are all rails based, including MetaPaas, Recruiting On Rails and current SFP.
I was amazed by the convenience of Rails, and also hurt by its “smartness”.
The power of Rails was described in quite a lot of posts, so I wanna to share some failure experiences.
Actually I have felt into quite a number of pitfalls in Rails, and here is one of the most painful ones.

To explain the problem easier, I just simplify the scenario:
I have a model called “Candidate”, which holds a “Status” to store the status of the candidate, so I have the code like this:

1
2
3
4
5
6
7
8
9
10
11

class Candidate < ActiveRecord::Base
	has_one status
	# other definition
end
class Status < ActiveRecord::Base
	belongs_to candidate
	# other definition
end

For some reason, I change the relationship between Candidate and Status. It is changed from one-to-one to one-to-many.
So I changed the has_one to has_many:

1
2
3
4
5
6
7
8
9
10
11

class Candidate < ActiveRecord::Base
	has_many status
	# other definition
end
class Status < ActiveRecord::Base
	belongs_to candidate
	# other definition
end

I thought it is an easy modification, but the app fails to run, even I have done the database migration.
It said rails cannot find a constant named “Statu”!

After my first sight on this error message, I believed it is caused by a typo, I must mistyped “Status” as “Statu”.
So I full-text search the whole project for “Statu”, but I cannot find any.

This error message is quite weird to me, since I have no idea about where the word “Statu” come from!
After spent half an hour on pointless trying, I suddenly noticed that the word “status” is end with “s”, and according to Rails’ convention, rails must think “status” is the plural form of “statu”. So according to the convention again, it try to find a class named “Statu”.
And we should use the plural form noun as name for one-to-many field, since that holds an array rather than a single object.

So after changing status to statuses, the problem solved.

Convention based system is powerful, a lot magic just happened there. But also the magic things are hard to debug when some special case breaks the convention presumption.

I assigned a dedicated ssh key pair for github repos.
And I have associated the key pair with github correctly in ~/.ssh/config.
But each time when I try to access github repos via ssh, both read(such pull or fetch) or write(such as push), I will get a strange error:

key_read: uudecode [some SSH key code]
ssh-rsa [SSH key code]
failed

I tried a lot to fix the problem, and finally I solved the problem by delete the file ~/.ssh/known_hosts
I assume the problem might be caused that there is some invalid association cached in the file. So maybe you can solve the problem by removing the related entries instead of delete the whole file.