Advogato blog for saju

6 Mar 2008

Thu, 6 Mar 2008 13:32:27 GMT

Maybe I am not understanding this properly but java's threadlocal storage seems funky. Unlike in other non-java threading environments one cannot introspect a thread's local storage and retrieve the objects needed. To retrieve a thread local value one needs a reference to a ThreadLocal object which was used to create this TLS. This seems a bit funky; to retrieve a TLS value at a random point in that thread I need a reference to a global ThreadLocal object; so if I have to use a global object to grab TLS values I can just make my TLS values globals !

The only difference b/w globals and TLS is that the value of the TLS object can be unique per thread. This is cool but the inability to introspect a thread's TLS values without holding a global reference doesn't make sense. This stops me from using TLS where I am unable to add/modify globals for some reason (say I am writing callbacks for an established framework).

Ofcourse I probably didn't grok this right... maybe...

4 Mar 2008

Tue, 4 Mar 2008 05:02:05 GMT

I am not a Java person and have less than a week's experience working with J2EE, maybe my brain is not wired correctly to understand Java but I have significant programming experience in multiple languages and I have never seen such a proliferation of frameworks and abuse of xml in the name of declarative programming. Most of the latest Java based frameworks have a large amount of documentation explaining how they are different from the other frameworks and how using their framework apparently makes it easier for the programmer to concentrate on business logic and forget about the "grunt" work and loose ends. This clearly seems like a bunch of authors writing a lot of code to create a fwmk/api based on an idea which was simply not significantly different from the previous fmwk to stand on it's own merit.

Also why do most new java fmwks/api's state one of their goals as "making programming easier" ? Surely the J2EE programmer is smart enough to know that he should close files he has opened in his program ? Surely the answer to a bad programmer is not to create a brand new framework that closes opened files via code injection and declarative xml programming. If there is a problem with a existing framework or an api - maybe everyone should just try fixing it ?

I definitely don't have enough exposure to understand all the problems of Java or J2EE and the need for so many frameworks so you could definitely say I not qualified to comment on something that I obviously don't understand well enough. In my defense; neither do I have any significant experience in say Python, Lisp or Perl but I can already see that they have got their story right. Maybe after you have spent enough hours hacking stuff at ungodly hours, you just develop a sense of what is right and what is not, and Java and it's framework hell just does not seem right

22 Feb 2008

Fri, 22 Feb 2008 11:40:13 GMT

I saw Spiderman-3 yesterday. It should have been named Mary Jane - 1. I was still recovering from the shock of watching Transformers, when I had to endure all that crappy teenage angst of spidey3.

Current mood - Disturbed

16 Feb 2008

Sat, 16 Feb 2008 15:56:41 GMT

Just got a shiny new Dell E248WFP monitor. Awesome good. Hooked up my Macbook Pro to the monitor and everything works great. Well almost everything. "Mirror mode" forces the monitor resolution to drop to the macbook's resolution.

I will get myself a nice USB mouse and keyboard tommw and run my macbookpro in "clampdown mode". That should make me happy mmmm..

15 Feb 2008

Fri, 15 Feb 2008 15:39:31 GMT

javax.net.ssl.SSLEngine adventures

I recently had to write a server in java that can speak SSL/TLS. The actual application protocol is handled by classes upstream to the SSL/TL layer. For scalability I decided to use Java NIO - which implied using the javax.net.ssl.SSLEngine.

The SSLEngine api is pretty cool though could possibly be a bit better designed. Essentially the application needs to manage 4 buffers of data, 2 of which are read by and 2 of which are written to by the SSLEngine. Sounds easy, infact it is, if it were not for the Handshake and renegotiation. Calling an api on the SSLEngine can cause it to ask you to call another api which may in turn ask you to call the original ... and so on.

I first wrote the SSL layer using recursion but with upto 5 exit points and special code to take care of stack unwinding and having to manipulate global buffers this was deemed too risky. So I re implemented the SSL engine via state transition. Since I was already playing with NIO and had to use a thread pool to handle concurrent clients, I decided to make the SSL processing asynchronous. Essentially I broke down the SSL processing (including breaking down the handshake and app layer protocol parsing) into various states, the currently executing thread would finish executing 1 state and hand over the "session" to the thread pool where another free thread will finish the next state. Ofcourse I have coded in optimization where an implementation of a state can decide whether to transition over a thread boundary or force the current thread to execute the next state.

Long story short. It works andit fits very nicely into the server's threading model.

4 Jan 2008

Fri, 4 Jan 2008 11:57:43 GMT

My VLC player is tuned to Gothic and Darkwave radio stations while I hack a working AES implementation in Actionscript 3. Coincidence or what ?

Dear ECMA people please please provide us mortals with byte addressing

1 Jan 2008

Tue, 1 Jan 2008 15:00:29 GMT

Joel Spolsky's "Travel Survival" tips.

Tip #2 : Fly first class.

Right. That's when you stop reading.

1 Jan 2008

Tue, 1 Jan 2008 05:09:54 GMT

The Leopard and The Tata Indicom Plug and Surf thinggy; a story.

Well actually not much of a story. I got my hands on a "Tata Indicom Plug and Surf thinggy". This is really a Sungil cdma usb modem using Qualcomm's mobile station modem (MSM) chipset. Darwin groks the msm chipset alright.

So I plugged in the modem into the USB port, pulled up "System Preferences", selected "Networking", filled in the account details - which should be same for all tata subscribers i.e dial #777, creds are internet/internet. Then I went to the "Advanced" menu section and changed the "Vendor" and "Modem" from "Apple" and "Apple Modem v34" to "Generic" and "Dialup Device". Hitting "Connect" should now get you online.

ps: i am posting this over the "tata indicom thinggy"

2 Aug 2007

Thu, 2 Aug 2007 12:47:09 GMT

framework HELL

and what's with the java community and flashy library names ?

4 Jun 2007

Mon, 4 Jun 2007 13:47:41 GMT

So I tried running Leenux on my shiny new dell d620 laptop. Since my linux skills suck and specifically since the 4th FC4 CD was fubar, I ditched the whole "put linux on my laptop" experiment and turned to FreeBSD.

fbsd 6.2 is lovely !. I used the 24 MB "boot" only iso image to get a install kernel on my box and performed the ports/packages install over the net. Very nice !

After spending a day configuring stuff I finally have the following working (or almost working..)

Video @1200x800 - thanks to 915resolution - which was in "ports" yeah
Sound ! - had to get the snd_hda.ko driver.
Skype !!!! - yeah baby - (linux abi ofcourse :D)
xmms for my lovely mp3's
mplayer for all those pron mpegs.
firefox with flash7 (linux binaries). The flash 9 crashes :( and there are no native flash binaries...
Java ! - yeah both the linux and native jdk
wired NIC - bge
USB - works like a charm
Wireless - if_wpi - maybe works maybe not - don't know - have to test yet - saw a bunch of mem align warnings on console though
fvwm2 -- mmm
acpi - well atleast I have it configured in my custom kernel

So - freebsd is going to stay on my new laptop from now on. Feeling happy!