zsh compinit: insecure directories, run compaudit for list.

If you encountered this message prompt:

zsh compinit: insecure directories, run compaudit for list.

Ignore insecure directories and continue [y] or abort compinit [n]? 

And you googled, binged, chatgpt'ed and tried everthing, and you still got it.  This post is for you.

Do this, put this line in (create or append to) this file: .zshenv

skip_global_compinit=1

Save it and restart shell, the message should be gone.

If you on Mac, you can easily solve this by:

compaudit  # find out insecure directories

sudo chown -R yourname /usr/local/share/zsh

However, if you are in multi-user Linux environment, you may not be able to do it, and even you can do it, your message may be gone, other users will get that message.

The reason for this message is that zsh sources /etc/zsh/zshenv,  inside zshenv it run compinit.   zsh source /etc/zsh/zshrc before .zshrc, that's why even if you put `ZSH_DISABLE_COMPFIX=true`, or change other configs in zshrc, the message will still be there.

By putting `skip_global_compinit=1` in .zshenv, it will /etc/zsh/zshrc to not run compinit.  

If you use oh-my-zsh, you still need to put:

`ZSH_DISABLE_COMPFIX=true`

at the beginning of your .zshrc file, otherwise it gives you warnings other than that message prompt.

apt update issue: Problem parsing dependency

If you encounter this problem when doing 'apt update'

E: Problem parsing dependency 21 of libc6-dev-s390x-cross:all=2.31-0ubuntu7cross1

E: Error occurred while processing libc6-dev-s390x-cross (NewVersion2)

E: Problem with MergeList /var/lib/apt/lists/us.archive.ubuntu.com_ubuntu_dists_focal_main_binary-amd64_Packages

E: The package lists or status file could not be parsed or opened.

Do these:

sudo rm -fr /var/cache/apt/archives/*

sudo dpkg --clear-avail

sudo rm -fr /var/lib/apt/lists/*

sudo apt update

Biden's 11/4 Early Morning Vote Jump in Michigan (Un-)Explained.

You've probably seen this graph and noticed the sudden jump of Biden's vote total in the early morning on 11/4 in Michigan.

You may have seen Josh Philip from EpochTime or the POTUS's videos on this jump.  (If you haven't seen them, that's because they are censored by BigTechs and main stream medias.)

In this post, I explain this jump and why it's fraudulent.

Let's focus on the "jump".

You can look at the data in your browser.  Open with Firefox:

https://static01.nyt.com/elections-assets/2020/data/api/2020-11-03/race-page/michigan/president.json

Click "data", "races", "0", then scroll down and find and click "timeseries", then scroll down and find 465 and 466, click to open both, click "vote_shares" in both, you will see:

The jump at 466 happened at 11/4 11:31:53Z.  The time is GMT, in eastern time, it's 6:31:53am.  Notice the vote dump before it has timestamp 11:31:48Z.  That's 5 seconds before.  Edison research aggregates data from different counties, 2 counties might happen to report their data 5 seconds apart.  So the 5 second time difference is not cause for concern, unlike some pundits have said.

The number of votes in this vote dump is 4724327-4574555=149772.  Using "vote_shares" provided, Biden got: 4724327*0.485-4574555*0.47=141258 votes

Trump got: 4724327*0.498-4574555*0.513=5968 votes.

The number of votes that're calculated are not accurate because "vote_shares" are rounded to 3 decimal places.  To get accurate number of votes, we turn to another source.  This is New York Time's "votes remaining" data snapshots.  The data can be found:

https://static01.nyt.com/elections-assets/2020/data/api/2020-11-03/votes-remaining-page/national/president.json

By compare different snapshots of this data, we can find out which county updated its vote total and exact vote splits.

The data from Wayne County is listed here:

Notice the difference between the rows at 10:43:33Z(blue) and 11:33:20Z(orange).  Votes increased by:

607656-457884=149772.  So we know the vote dump is from Wayne county.

The exact number of votes for Biden and Trump:

Biden: 408131-268229=139902 votes

Trump: 190951-181974=8977 votes

That's 139902/149772=93.4% for Biden, and 8977/149772=6.0% for Trump.

Now the question is: Are these votes from Detroit?

From Detroit's official election result website:  Out of 250138, Biden got 233908, Trump got 12654.  It seems obvious the vote dump is from Detroit because the percentages match.  But let's make sure.

From Wayne county election result website: Out of 874018, Biden got 597170, Trump got 264553.  So from NON-Detroit precincts: out of 623880 (874018-250138) votes, Biden got 363262 (597170-233908), Trump got 251899 (264553-12654).  Percentage wise, Biden got 363262/623880=58.2%, Trump got 251899/623880=40.4%.

So from vote distribution, the 149772 vote dump could NOT come from non-Detroit Wayne County precincts.  It could Only come from City of Detroit.

But it DIDN'T.

From precinct level data snapshots of New York Times, in the timeframe between 10:23Z and 11:38Z, City of Detroit's increase in vote total is 11067.  

(The latest version of this data can be found at: https://static01.nyt.com/elections-assets/2020/data/api/2020-11-03/precincts/MIGeneralConcatenator-latest.json , but you need different snapshots of this file and calculate the difference to make sense of the data.)

Up to 10:23Z, Detroit's total is 214644, and it's already included in the 457884 figure from Wayne county.  Detroit's final total is 250138, so the 149772 number can not come from City of Detroit.

Now we have a problem: We know these 149772 votes came from Wayne County.  They can only come from Detroit and they did not come from Detroit.   They seemed to come out of thin air.

No one have provide an explanation to the Biden Jump and no medias have disputed or "fact-check"ed the fact that it's suspicious and the 149772 votes were most likely fraudulent. 

Pennsylvania Election Night Data Irregularity

This is the plot of vote counts from Pennsylvania presidential race.  The data came from Edison Research, which was used by virtually all main stream medias for vote reporting.  You can get the data here.  Open it in Firefox, then click data->races->0->timeseries.  

You may notice something strange at the beginning part of the graph.  Vote count normally only goes up because it continues accumulating as time goes by.  But in this graph, there are spike-downs as well as spikes up. What's going on?

We will  expand this weird looking part and focus on the first 55 vote dumps from Pennsylvania.

The first vote dump (numbered 1 in the plot) out of Pennsylvania has 64535 votes with Biden's share 0.799 and Trump's 0.188.

If we do the calculation, Biden got 51563 and Trump got 12132.  Because vote shares are rounded to 3 decimal places, the vote counts Biden and Trump get are not accurate, even though the total vote is accurate.  But we can actually get the accurate vote count for each candidate.  The source is also Edison/New York Times.  NYTimes has timestamped snapshot vote reports for all precincts for certain counties of Pennsylvania.  For example the first report that has any non-zero votes is:

https://static01.nyt.com/elections-assets/2020/data/api/2020-11-03/precincts/PAGeneralConcatenator-2020-11-04T01:02:56.961Z.json

By calculating the difference between vote counts in different timestamped reports, we get a clear picture of vote increments for each candidate for each precinct of some counties at each timestamp.  So we know the exact vote split of the first vote dump - Biden:51555, Trump:12147, and they are all absentee votes from Allegheny county.  Here's sample of 2 precincts.

Next vote dump of 21818 is from Luzerne county.  The one after that is mostly from Centre county.

The vote dump numbered 2 in the graph is from Philadelphia. It has Biden 70573 and Trump 4279.  This first vote dump out of Philly is highly unusual.  It has by far highest Biden/Trump ratio of 16.5 among all vote dumps out of Philadelphia.  

In vote dump marked 3, out of 82711 votes, around 74000 were given to Biden, around 8500 was given to Trump.  At this point, Biden has 254000 votes, Trump has only 46000.  So Biden held a margin of 200 thousands over Trump.  It looks like Biden was given a huge head start.  But is it fraud?  Yes, it is.  Because they admitted in the vote dump #4.

In vote dump #4, there are 130027 votes, almost all of them are for Trump, Biden got virtually 0.  How can this possibly be a real vote dump?  It cannot.  I think someone was afraid of getting caught because the unrealistically huge Biden margin, so they bumped Trump's number up so it didn't look so suspicious.   But as we will see, this bump would be taken away later.

In the 3 vote dumps between #4 and #5, out of 121000 votes, Biden gained 96000, Trump gained 24000.   In the vote dump marked 6, out of 113941, Biden got 89000, Trump 24000.   A large component of this dump are votes from Montgomery county (72823 votes, Biden 64777, Trump 7756).

Biden's margin at this point is again over 200 thousands.

At #7, we have a crazy vote dump of 239084, out of which 200000 are for Biden, 40000 are for Trump.  Then after more than a dozen vote dumps, in #8, exactly 239084 votes were taken out.  I guess someone realized a huge Biden:Trump 5:1 vote dump was jus too unbelievable,  better taking it back then spread them out to smaller vote dumps so they are more credible.

In #9 vote dump, 114886 votes were taken out.  But only Trump's vote take a dive,  Biden's vote actually increased.  It seems they took out the #4 Trump bump.  At this point Biden's margin increased to be over 300 thousands.

Next in #10 vote dump, a total of 507047 votes were added.  Out of them, 337K are for Biden, 146K are for Trump.  This half million vote dump is absolutely fraud!  In these votes, 173K are from Chester county.  Out of 230 Chester precincts, 203 precincts reported 675 votes with Biden 421 and Trump 233.

To verify, open:

https://static01.nyt.com/elections-assets/2020/data/api/2020-11-03/precincts/PAGeneralConcatenator-2020-11-04T02:16:40.084Z.json

 in Firefox, click "Raw Data", then search for "chester" a few times.

Can 203 precincts have exactly same number of votes and exact same vote splits?  Impossible!

Both 421 and 233 are prime numbers.  What's the reason for prime numbers?  If you use prime number , you are guaranteed to NOT have a whole number ratio or numbers that have a common denominator (other than 1).    I guess the fraudulent vote tabulating system was programmed to generate random numbers that approach an overall ratio.  But to avoid suspicion, it try to generate numbers that have no apparent relationship.  Unfortunately in this case, the random number generator got stuck!

How many more  of these half million votes are fraudulent?  Most of them are.  Again realized this fraud vote dump was too apparent, in #11 at 2:22GMT or 9:22PM EST, they removed a total of 586189 votes!  Which votes did they remove?  I don't know.  But I do know that those fraudulent Chester county votes were not removed.  The next Chester vote update wouldn't come until 3 hours later at 5:27GMT (or 0:27 EST).

After #11 huge vote reduction, Biden still hold 200K margin over Trump.  From this point on, Trump was able to catch up and leap over.  The "glitch" that was caught on CNN where exactly 19958 votes were removed from Trump and added to Biden's total, happened at 4:08GMT (11:08pm EST).  As Trump continued gain momentum, at 2am EST in the morning he had a 700K margin over Biden.  Then came the big nation wide coordinated "halt", and vote dumps afterward that miraculously narrow the gap.  They have everyone's attention.

But I think just like the one in Maricopa county, the huge early Biden head-start in Pennsylvania need more attention.  The first 50 vote dumps out of Pennsylvania were complete mess.  Some of them got cleaned up,  but a good number of fraudulent votes are still there and in the final total.  They should be investigated and taken out.   The 421:233 Chester votes are still there.   The 75K Biden/Trump ratio 16.5 Philly votes are still there.  The 73K B/T 8:1 ratio Montgomery votes are still there.  The 30K 4:1 Centre votes are still there.  If even half of 200K early fraudulent Biden margin is taken out, Trump would win Pennsylvania (Biden's margin is 82K)

Healthy lifestyle and life expectancy

This is a summary of the Harvard study:

Healthy lifestyle and life expectancy free of cancer, cardiovascular disease, and type 2 diabetes: prospective cohort study https://www.bmj.com/content/368/bmj.l6669

In 3 sentences: 

At age 50, life expectancy free of cancer, cardiovascular disease, and diabetes was 23.7 (95% confidence interval 22.6 to 24.7), 26.4 (25.2 to 27.4), 29.1 (28.0 to 30.0), 31.8 (30.8 to 32.8), and 34.4 (33.1 to 35.5) years among women who adopted zero, one, two, three, and four or five low risk lifestyle factors, respectively. 

Life expectancy free of cancer, cardiovascular disease, and diabetes at age 50 was 23.5 (22.3 to 24.7), 24.8 (23.5 to 26.0), 26.7 (25.3 to 27.9), 28.4 (26.9 to 29.7), and 31.1 (29.5 to 32.5) years among men who adopted zero, one, two, three, and four or five low risk lifestyle factors, respectively. 

The percentage of life expectancy free of cancer, cardiovascular disease, and diabetes from total life expectancies was 74.8%, 77.6%, 80.1%, 82.2%, and 83.6% among women (75.3%, 75.8%, 76.8%, 77.9%, and 79.0% among men) who adopted zero, one, two, three, and four or five low risk lifestyle factors, respectively.

The 5 low risk lifestyle factors are:

1. high Alternate Healthy Eating Index - higher the better

2. no smoking - more smoking the worse

3. moderate activity - more vigorous the better

4. alcohol - none only better than heavy drinking, best is moderate drinking, surprise!

5. low BMI - lower the better until too low

I'm a little disappointed with these findings.  If you do everything right, you're only expected to live 31.1-23.5=7.6 years longer and only 4.2% more free of cancer, cvd, and diabetes.

Keep drinking!

Extracting video from blob url

First press F12, then select Network tab. reload page, then play video.  You should see a file called playlist.m3u8.  Right click it then copy address.

On a command line(Win, Mac, Linux), type youtube-dl followed by pasting the m3u8 link, then enter.

Of course you should already have youtube-dl installed.

If above doesn't work, you can download the m3u8 file. Look inside the m3u8 file, if it contains relative links to ts file for example: 

/somefolder/somesubfolder/file01.ts

You need to replace this relative link with absolute link by adding hostname, for example, if you downloaded the m3u8 file from https://example.com/, you need to insert 'https://example.com' before every ts lines, so the above will be like:

https://example.com/somefolder/somesubfolder/file01.ts

You need to do this to all lines that contain ".ts".

Save the file, then:

ffmpeg -protocol_whitelist file,http,https,tcp,tls -allowed_extensions ALL -i playlist.m3u8 -bsf:a aac_adtstoasc -c copy out.mp4

Disable ubuntu update manager, timer from command line

On command lines:

sudo systemctl disable apt-daily.timer
sudo systemctl disable apt-daily-upgrade.timer

then edit 2 files:

sudo vi /etc/apt/apt.conf.d/10periodic
sudo vi /etc/apt/apt.conf.d/20auto-upgrades

In both files, change "1" to "0".

Creating Time Lapsed video from dashcam videos

Time lapsed video are usually create from photos.  Here's how to create one from dashcam videos.
First move all videos into a single folder, then list files and redirect output into a text file.  On linux, it's `ls > files.txt`, on Windows, it's `dir /B > files.txt`.

Then edit this file.  This file should list videos in the order of timestamp, if not, adjust them manually. Remove any files that's not video or any videos that you don't want to be included in the final video.  Put 'file ' at the beginning of every line.  (In Vim, you can do this by :%s/^/file /g)

Then create a raw video by select one frame in every 30 frames from all the videos in the file:

ffmpeg -f concat -i files.txt -vf "select=not(mod(n\,30)),setpts=N/(FRAME_RATE*TB)" -vcodec rawvideo -pix_fmt yuv420p -an raw.yuv

Because this is uncompressed raw video, its size will be huge, probably tens of GigBytes. You probably don't want to play it on your computer, let alone uploaded to youtube.  To make it playable, we need to compress it to mp4 format:

ffmpeg -f rawvideo -pix_fmt yuv420p -s:v 1920x1080 -i raw.yuv -vcodec libx264 video.mp4

 You can see an example of the time lapsed video here:

Tornado vs Starlette in 2019

For the last 10 years I have been using mainly  Tornado as my web framework of choice.  And I mostly use it synchronously.  Only when dealing with uploading files did i use some asynchronous tornado features.   Most of my apps have database backends.  With database involved, it's not worth it to convert my code to asynchronous mode.  Tornado has been worked well for me I have no intention of stopping using it in the near future.  But I do want to keep my eyes open for new tool that enables me to build fast app fast.

In the last few years, there are several new async web frameworks.  One of them is Starlette, I decided to evaluate it.  The first step is to compare it basic performance with Tornado.

Starlette, using uvicorn, claims to be one of the fastest python web framework.  I copied the codes directly from its website:

from starlette.responses import PlainTextResponse

async def app(scope, receive, send):
    assert scope['type'] == 'http'
    response = PlainTextResponse('Hello, world!')
    await response(scope, receive, send)

I run it with "uvicorn example:app", then measure rps using 'ab':

ab -n 1000 -c 10 127.0.0.1:8000/

 The result is on average 4200 requests per second.

I then run basic Tornado app, again with code copied directly from tornado website:

import tornado.ioloop
import tornado.web

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("Hello, world")

def make_app():
    return tornado.web.Application([
        (r"/", MainHandler),
    ])

if __name__ == "__main__":
    app = make_app()
    app.listen(8888)
    tornado.ioloop.IOLoop.current().start()

 Using the same measurement, the ab result is a average 1900 requests/second.

So it looks like Starlette/uvicorn is more than 2 times faster than Tornado.   However, the code on Tornado website is actually not the best for production use.  Just replace "app.listen(8888)" the second to last line with following:

server = HTTPServer(app)
server.bind(8888)
server.start(0)

(and put "from tornado.httpserver import HTTPServer" at the beginning of the file),  its performance increases to 4700 requests/second, actually faster than Starlette.

The code change to Tornado make the app start 4 processes instead of 1.  Of course we can also do the similar for the Starlette app:

uvicorn --workers 4 example:app

The result now is 7500 requests/second, surpassing multi-process Tornado again.

These are superficial results that don't mean much.  But it did make me appreciate the works that got into Tornado that keep it performant in these years.

Result Summary:

• Tornado (default): 1900 rps
• Startlette/uvicorn (default): 4200 rps
• Tornado (4 workers): 4700 rps
• Starlette/uvicorn (4 workers): 7500 rps

DrJava Font Size Problem

I'm helping my son learning Java. One of the first things to do when learning Java is picking an IDE. I like DrJava for its simplicity and small size. However, when I first run DrJava, its font size is way too small on my 4K monitor.

There are 2 ways to fix this - the easy way and the hard way.

The easy way first.
Click Edit:Preferences, then click "Display Options", change "Look and Feel" to the one that ends with "Plastic3DLookandFeel".  Do this on Windows only, for Mac and Linux, default "Look and Feel" is fine.
Click "Font" under "Display Options", change all fonts to double the original size:

Press "OK", then close DrJava then open it again.  Now the fonts should be big enough to read.

(The changes you make are actually saved in a configuration file .drjava in your home directory,  you can edit the file directly for changes, but it's not recommended)

Now the hard way.
For the "easy way", we just changed DrJava's default configuration preferences.  The hard way is to compile a DrJava program with these changes already made in the source code so we don't have to change preferences.

Clone the DrJava repository on github: https://github.com/DrJavaAtRice/drjava

Try to compile it first:

cd drjava/drjava
ant jar

You must have ant already installed.  If it's compiled OK, try run it:
java -jar drjava.jar
(You can just double click drjava.jar file too)

Open src/edu/rice/cs/drjava/config/OptionConstants.java for editing:
Change "Monaco-12" to "Monaco-24"; change "Monospaced-12" to "Monospaced-24";
chanage "dialog-10" to "dialog-20"; change "dialog-12" to "dialog-24".

Save the file then recompile (ant jar).  Now DrJava is "pre-configured" with big fonts.

You might say this "hard way" is pointless.  Why on earth would anyone want to do this? I agree.  But just maybe one wants to provide a "pre-configured" copy of DrJava to his students.  I just show one way to do it.  Another way is to write a .drjava configuration file upon installation.  This is way more complex. Besides, I'm using a standalone jar file, there's no installation to speak of.

Bugzilla "Cannot determine local time zone" problem

I had this problem today and was pulling my hairs out trying to find a solution.
Below is the only fix that worked for me:
In 'lib/DateTime/TimeZone/Local.pm', find 'sub TimeZone {', right after it, put this next line:

return DateTime::TimeZone->new(name => 'Asia/Shanghai');

(Replace 'Asia/Shanghai' with your own timezone.)
Basically all the rest of this sub has no effect now.

Reload the error page, if it show another error says: "List::Util version 1.31 required...", open 'lib/List/AllUtils.pm', goto line 10, or find the line:

use List::Util 1.31 ();

Just delete '1.31'.

After that, it should work.  I'm using Bugzilla 4.4.9.

Cloning HDD to a smaller SSD for a Windows 8 Laptop

My win8 Dell Laptop has a 500GB HDD, I bought a 480GB SSD for hard disk replacement.  Unfortunately, the 500GB HDD's partitions span the whole disk, the exact layout copying wasn't possible.

I formatted the SSD to a GPT disk, then use various cloning softwares to do the cloning.  I tried Macrium Reflect Free, EaseUS Partition Master, Minitool Partition Wizard. None of them get me to Windows startup screen.

The last one I used is Minitool Partition Wizard. During boot, it said can't find OS.  So I tried changing different bios settings out of desperation and hoped one would work.  It did.

The setting is "disable virtualization".  I don't know why it worked, it doesn't make any sense, I'm just glad it did.

So if you have problem migrating HDD to SSD, this might be something you can try.

Re-sizing a Virtualbox disk image in Linux

One of my VirtualBox virtual machine had its disk full, I thought a simple resize should fix the problem. And I went:

vboxmanage modifyhd u140464-02-disk1.vmdk --resize 30000
0%...
Progress state: VBOX_E_NOT_SUPPORTED
VBoxManage: error: Resize hard disk operation for this format is not implemented yet!

What?!
Turned out "modifyhd" only support .vdi, not .vmdk. OK I will just use .vdi then:

vboxmanage clonehd u140464-02-disk1.vmdk u140464-02-disk1.vdi --format vdi
vboxmanage modifyhd u140464-02-disk1.vdi --resize 30000

Then detach vmdk disk from the VM and attach the vdi.
But after I boot it up, the disk was still full. The disk size wasn't changed. After some googling I found I need to install gparted and use it to grow the partition, and gparted has to be run at boot. Too much hassle.  All I want was some additional disk space. So I did it my way.
(Inside the VM)

sudo fdisk /dev/sda

then enter p, it showed:

   Device Boot      Start         End      Blocks   Id  System
/dev/sda1   *        2048      499711      248832   83  Linux
/dev/sda2          501758    33552383    16525313    5  Extended
/dev/sda5          501760    33552383    16525312   8e  Linux LVM

Enter n to create a new partition, press enter 2 times to accept default partition size. For some reason, there was a stray segment, so I have to enter n again to create a new partition. 2 partitions were created, after a p, my partition table looked like this:

   Device Boot      Start         End      Blocks   Id  System
/dev/sda1   *        2048      499711      248832   83  Linux
/dev/sda2          501758    33552383    16525313    5  Extended
/dev/sda3          499712      501757        1023   83  Linux
/dev/sda4        33552384    61439999    13943808   83  Linux
/dev/sda5          501760    33552383    16525312   8e  Linux LVM

sda3 is useless, so I deleted it by a 'd' followed by a '3'. sda4 is the one I will use. Enter 'w' to write to disk. Then reboot the VM.

After reboot, do a:

sudo mkfs.ext4 /dev/sda4

My /opt was not used, so I just mount sda4 as /opt

sudo mount /dev/sda4 /opt

Add "/dev/sda4 /opt ext4 defaults 0 2" to the end of /etc/fstab so it was mounted automatically at boot.

In the end, I should have just add another virtual disk without bothering resizing the current one.

fixing VPN problem in Ubuntu

I set up VPN in Ubuntu 14.04, but when I tried to turn it on in NetworkManager, I kept getting errors like "No suitable device found", "Could not find source connection" in /var/log/syslog.

The problem turned out to be that my ethernet connection was not managed by NetworkManager. "nmcli c" didn't show my wired connection.

The solution:
Edit /etc/NetworkManager/NetworkManager.conf, in section [ifupdown], change "managed=false" to "managed=true", then do a:

"sudo service network-manager restart"

Now both vpn and ethernet connections are managed by NetworkManager, turning on VPN just works.

Programming Minecraft with Python: Introducing JuicyRaspberryPie

I developed a Bukkit plugin called JuicyRaspberryPie, that let you add commands to Minecraft with Python programming.

It's largely based on RaspberryJuice plugin, I added a Python command server and also changed the Java code a little bit to let the plugin talk to the command server.

The github page has more detail.

Here's a video demo of the plugin. (Fast forward to 15 minutes for python examples, before that is just about how to download and install the plugin, which is pretty boring)

Solving windows installer error 2503 2502 problem

I got a new laptop and tried to install python on it. But it gave me 2503 2502 error.  I had installed python on other windows 8 machines and never had this problem.

2503/2502 problem is caused by insufficient permission, which should be fixed easily, just right-click and then "run as Administrator", right?  Wrong! There's no "Run as Administrator" option available for MSI (windows installer) file.

Now the solution: open explorer, find your way to C:\Windows\System32, locate cmd.exe, right click it, then "Run as Administrator".  (If that's too complicated, try this: press "windows+Q", type "cmd", right click "Command Prompt", then "run as administrator").

Now cd to the directory where download the MSI file, then type the filename and enter. Install as usual.

EasyDedupe: delete your duplicate photos, movies, or any files

I have a lots of duplicate photos and movies that's taken up lots of hard-drive space.  I'm tired of finding them one by one manually.  So I spent this weekend  writing a program that can automatically detect and delete the duplicates.  I was able to reclaim half terabyte of space by using this program.

The program is called EasyDedupe.  You can download it here.  After you download it, just unzip it (twice, if I don't double compress, the zip file can't be saved in Chrome), and run EasyDedupe.exe inside the unzipped folder.  Yes, it only runs on Windows.  But the program is written in Python, if you use Mac or Linux, just grab the source code and run it from the command line: "python EasyDedupe.py". To run the source, you need to have wxPython installed. (On Mac, also do a "export VERSIONER_PYTHON_PREFER_32_BIT=yes" to set python run in 32bit before running Python)

There are 2 ways to run it: one is comparing 2 folders and delete any files in the second folders that's duplicate of the ones in the first folder. The other is simply select one folder, the program will find duplicates and delete them.   The first way is useful for deleting duplicates that are on the different folders or disks.  The second way is useful for deleting duplicates within a single folder (and it sub folders).

The program is very much alpha.  Let me know if it crashes on you.  It works for me.  It shouldn't delete non-duplicates.  Of course there's absolutely no guarantee nor warranty.  If you want to be extra cautious, use "Dry Run" option at the bottom of the UI.  "Dry Run" will not actually delete any files.  You can then look at "log.txt" file to see if the detected duplicates really are duplicates.

Fix windows 7 explorer displaying square boxes

I had to re-install Windows 7 recently.  After installation, my Chinese file names are all square boxes in the windows explorer.
I searched online, some suggested deleting the file "C:\Windows\System32\FNTCACHE.DAT". This did not work for me.
The solution is actually this:
Open control panel-> Apperance and Personalization, select under "Fonts", "Change Font Settings".
Uncheck "Hide fonts based on language settings".   Reboot.

CombiMath My first IOS game

My first IOS game CombiMath is released. (App Store Link)

(Accompanying web site - http://combimath.com/, visit it from cellphone to take full advantage of it.)

It's a "calculate 24" game.  I made the same game for web 3 years ago using Google AppEngine (link).  For IOS, I tried  to make it using HTML5 wrapped in UIWebview so that I don't have to use Objective-C.  But I wasn't satisfied with the end result of that effort.   So I bite the bullet and start learning Obj-C.  Turned out it wasn't so bad.  Objective-C is quite easy.  I also used Cocos2d, a popular 2D IOS game engine, which simplified a lot of things.

I'm glad it's out of the door.  I need to add game center with leader-board and battle mode in the next release.

Video formatting for PowerPoint embedding

If you have problem embedding videos into your PowerPoint presentation, here's some help.

You need to convert your video to a format PowerPoint accepts. However even if you are able to embed your video and play it on your computer, it doesn't mean it will be playable on other computers. I tried quite a few combinations of video/audio format (h264, mp3, aac etc.) and found the least common denominator of video/audio format that can be embedded in PowerPoint and played on almost all windows computers is wmv2/wmav2. So here’s command for video conversion with ffmpeg:

ffmpeg -i video1.mov -vcodec wmv2 -acodec wmav2 -f asf video1.wmv

The resulting video quality is usually not very good, the images are often pixelated. To get better video quality, you need to specify bitrate:

ffmpeg -i video1.mov -b 1500k -vcodec wmv2 -acodec wmav2 -f asf video1.wmv

Here I specified 1500kbps bitrate. You can set higher bitrate. Of course it doesn’t make sense to go above the video’s original bitrate. You can find the a video’s original bitrate from ffmpeg output or on Windows, right click the video file and select “Properties” then “Details” tab. Try different bitrates and select a lower one while still maintaining an image quality. Often times, I want to keep my video (and therefore the size of PowerPoint) small by shrinking the video (e.g. 320x240) like this:

ffmpeg -i video2.mp4 -vcodec wmv2 -acodec wmav2 -f asf -s 320x240 video2.wmv

Small video has an marginal benefit besides small file size: because the image size is small, the effect of pixilation is markedly reduced. Even with smaller size, with several videos in the Powerpoint, the file can still be too large to be sent with email(gmail and yahoo mail have 25Meg limit). To solve this problem, I usually just embed small, low quality, incomplete “teaser” videos with notes and links to complete 1080p videos I (pre)uploaded on youtube. To cut a 30 second video, starts from 5 seconds at original video:

ffmpeg -i 027.MOV -ss 00:05 -t 30 -vcodec wmv2 -acodec wmav2 -f asf -s 480x270 027s.wmv

My original 1 minute 1920x1080 video is 180MB, the half-minute “teaser” video is only 1.1MB.