After running checksetup.pl I’ve got the message, that required module Email::MIME::Modifier is missing and script suggested me to install it with the following command:

/usr/bin/perl -MCPAN -e ‘install “Email::MIME::Modifier”‘

I did it several times, but nothing happened. After some digging I found that the real culprit is a missing dependency from Email::MIME::Modifier to Email::MessageId.

So, you have to run the following command:

/usr/bin/perl -MCPAN -e ‘install “Email::MessageID”‘

or I prefer to use ports:
mail/p5-Email-MessageID

And everything will be just fine.

It seems like due to Vista “smarts” with user directories, compatibility issues and general MS idiocy FlexBuilder have no idea where to pick up the FlexBuilder.ini file. It works perfectly well on XP, but not on Vista.
It also grabs all default JVM parameters twice for some reason. Kind of an Irish approach

Bug is submitted.

Here is the example:

public function Parent()
{
}
}

If you don’t specifically tell to compiler that class Parent is bindable, compiler will give you the infamous warning on the attempt to bind to any public member of the Parent class:

Data binding will not be able to detect assignments to "m_sText".

P.S. It is logical if you think about the meaning of the [Bindable] directive and and you know the fact that PropertyChange events is not fired for changes happened inside of the complex members.

I’ve got 2 sets of wheels - one set is BFGoodrich/AT - All-Terrains mounted on factory’s  alloy wheels and another set is BFGoodrich/MT -  Mud-Terrains mounted on steel Pro-Comp wheels.

Every time before a trip I change the wheels by myself, it takes the time, but gives a perfect opportunity to have a good look under the truck and check for any signs of rust or other problems.

Recently, it appeared that my exercising with weights finally gave the result. A little bit unusual though - when I last time changed the wheels, I over tightened the nuts, so the thread were nearly stripped off and I had to replace all the rear axle studs (which cost me a 3 digit figure).  Ouch!

I’ve been told by FerntreeGully Nissan that if I want to prevent this from happening in the future I have to use the torque wrench. It is also appeared that Patrol’s studs are very susceptible for the over & under tightening.

So, the whole idea of this post is this: Use the Torque, Luke!
Use the torque setting of 135 Nm or 100 ft-lbs for GU Patrol wheel nuts!

If x or y coordinate becomes negative number with fraction, the whole popup window disappears.  This is strange as positive numbers with fraction are valid. And what is most frustrative, DesignView sets them with fractional parts when you re-arrange elements. 

UPD: bug submitted

Well, I still have to live with a couple of SC1435s as they are still doing their job, not so good though. Each Dell server had problems with RAID on the 13th month after the shipment date (maybe it is hard coded in their chips?). I was a cunning fox second time, so I’ve bought additionally 3 year warranty for the next server. And this is already paid off, as one of the WD hard drives collapsed last Friday (for some reason these type of things are happening only on Friday arvos).

Well, first sign of problem was an mpt-status utility which reported RAID status DEGRADED, and HDD1 status FAILED, OUT_OF_SYNC. And after a couple of hours HDD1 changed its status do MISSING.

I thought that this problem is similar to the problem I had before, when RAID array just become unsynchronised for some reason, so I was looking for an utility which will allow me to start remotely RAID re-synchronisation process (rebuild). I was surprised to find that Dell provides only 2 types of utilities - for Windows and for RedHat Linux. Well, Linux is Linux, even if I don’t run RedHat and I hate it, there was a good chance that I will be able to run the executables under Gentoo.  First ominous sign was the size of the download - 44Mb, which is too much for an utility. When I unpacked it with rpm2targz I found that most of the contents is Java crap which uses GUI to handle the RAID.

I know that I am weird, but none of my servers are running Windows (M$ or X, does not matter). My concept is that server is something which could be managed via the smartphone & ssh. GUI is not an option for servers. They are sitting in the hot hell of data centre and every single piece of processor & memory resources should be doing the primary task - running Apache/database/you name it, but not the Java+X+KDE for occasional maintenance.

Well, my attempt to use Dell’s utility failed. I looked up the controller’s chip (LSI Logic 1068) and went on their site for help.  The idiocy is spread there as well. They had a CLI tool for RAID management, but is was for the superseded MegaRAID, not for SAS 5 i/R anymore. They have this RedHat/Java rpm to manage controllers. Damn!

I googled for a while and finally found sane people at IBM who use command line tool called cfggen to manage SAS 5 i/R RAIDs on their servers.
NB HP servers also have it, not online though.

Well, at that point of time I already faced a fact that the hard drive went hanky-panky and never came back, so cfggen could only display the RAID and HDD0 status and all attempts to start rebuild failed, as there was nothing to rebuild.

There was another option - to use BMC facility of the server. Because Dell servers come without documentation (well, there were unpacking instructions packed inside of the box), I had no idea that BMC is accessible via shared network interface, and there is no need to use DRAC (Dell Remote Access Card), which is actually unavailable for SC1435 due to the space limitations. So I even didn’t bother to setup BMC in BIOS.
(On the controrary, HP servers have iLO, which is implemented more thoughtfully - separate interface which could use a separate or shared network and they are shipped with tons of documentations in PDFs.)

So, on the next business day, i.e. Tuesday (remember, that it happened on Friday arvo? technician from Dell changed the hard drive and RAID was back in business in a couple of hours.

So, what is the lesson I’ve learnt?

• Use Nagios plugin check_mpt
• cfggen from IBM website/ HP server utilities disk is a must for lucky Dell customers
• Out-Of-Band access is something which should not be ignored.
• think 3 times next time before buying Dell servers

I worked with all mainstream source control systems, but Subversion is the one which still puzzles me after a year of use.  That’s why I prefer to use TortoiseSVN, which seems to me really great thing (not the Subversion itself though). It’s a pity that it exists only in Windows version.

Anyhow, I was trying to understand why when I’ve got source tree checked out under my username, TortoiseSVN checks in modifications using root account, instead of my username. Subclipse does not have this problem at all.

Then I found the answer in the TortoiseSVN FAQ:

Why is there no ‘author’ shown in the logs when I commit changes via svn+ssh?
Since SSH completely takes care of the authentication process, Subversion won’t even see the author who does the commit. So to tell Subversion an author you have to specify the author in the URL itself. E.g. svn+ssh://username@server.com. You should do that when you check out your working copy.

When I tried it, TortoiseSVN used my username. Perfect! But Subclipse went nuts. It started to think that the repostory URL is username@ssh+svn://host/repository or similar. I tried both PureJava & SVNKit implementations with no luck. Nuts!

Here is more to read.

Seems that we have at least one sane Senator though.

And further reading

My previous fix has not covered all possible scenarios of the property assignment order. If you set dataProvider property it kicks out selectedItem/selectedIndex values. Here is attempt to make Combobox more conscious about its data.

Recently, I’ve bumped into the the strange thing: if you change directly the value:

this won’t generate propertyChangeEvent and all you bound variables won’t be notified about the change. So, instead when you need to notify others about the change, you have to do this:

This will cause the setter to be executed and propertyChangeEvent will be dispatched.

a) Flex Builder hangs (freezes) at the startup while loading the project which referes to that libs/cool_dir directory.

b) In the log file we saw the following:
Caused by: org.eclipse.core.internal.resources.ResourceException: The resource tree is locked for modifications.

After 4 hours of the tambourine dancing around Mac we have figured out that the cause of the problem was my libs/cool_directory. As soon as you remove it, everything gets back to normal. And only Flex Builder for MacOs is prone to this problem.

Damn it!  The number of hours I’ve already spent in the last 2 years troubleshooting Mac’s problems is now probably equal to the unforgettable hours I’ve spent installing SCSI board drivers for WinNT3.51/4.0 on SMP servers.  Oh…. sweet unpretentious Windows…

NB On the way, I’ve been really pissed off  by MacOS again - if you download zip file in Safari, it automatically unpacks it. And if folder in the archive is org.eclipse.core smart Safari-MacOs-SteveJobs-whatever unpacks it into the org/eclipse/core.  This is so sweet, caring and mindful.
I’ve been told that there is a setting somewhere in Safari to stop doing this, but we’ve got a presumption of innocence, right? Why should arrogant, ill-mannered and unsophisticated bloke like me know, that there is an option somewhere in the MacOs which says “let’s be stupid, OK/Agree?”

Anyway, enough about this bloody MacOs.  Allons! Revenons à nos moutons! Qu’en fut-il?

package

{

    import mx.controls.DataGrid;

    import mx.controls.listClasses.IListItemRenderer;



    public class EasySelectionDataGrid extends DataGrid

    {

        //this flag simulates holding a Ctrl key

        private var _holdCtrl:Boolean = false;



        public function EasySelectionDataGrid()

        {

            super();

        }



        public function set holdCtrl(value:Boolean):void

        {

            _holdCtrl = value;

        }



        public function get holdCtrl():Boolean

        {

            return _holdCtrl;

        }



        override protected function selectItem(item:IListItemRenderer, 

                              shiftKey:Boolean, 

                              ctrlKey:Boolean, 

                              transition:Boolean=true):Boolean

        {

            //override values if necessary

            return super.selectItem(item, shiftKey, 

                              _holdCtrl ? true: ctrlKey, 

                              transition);

        }

    }

}

Everything in the projects was pointing to the new sdk 3.1.0, instead of 3.0.0, but RSL libraries in the bin-debug folder were still from the old SDK (framework_3.0.0.477.swf, rpc_3.0.0.477.swf, datavisualization_3.0.0.477.swf along with corresponding SWZs)

After some head scratching I’ve opened up the properties of the main application,
Flex Builder path => Library path and looked into the settings of the framework libraries -bingo! It was still using the old settings.

So, all you have to do is to remove old RSLs and add new ones (both swf & swz), so it will look like that:

Single diff on 1000 lines files takes around 50 seconds. Multiply it by 10-20 per day and this is a stolen time from you.

It is time to change this. Here is the place to vote for this bug:

http://bugs.adobe.com/jira/browse/FB-12492

Sweet Adobe, please have a mercy on us all and give us back the
simple, no-gimmick diff, which just works.

Cheers!

Data binding will not be able to detect assignments to "SOME_STATIC_VARIABLE".

I don’t like compiler errors and don’t like compiler warnings as well. So, to avoid this warning you can do the following:

Instead of:

you can do this:

Voilà!

Problem 1:
setting ComboBox.selectedItem = null does not do anything. Logically, it should be equivalent to setting selectedIndex=-1. But it does not work.
There is Flex bug registered and it is marked as Fixed in Flex 3 (Released), which is a bullshit. It is still there.

Problem 2:
The whole Flash/Flex framework is event-driven thing. But sometimes developers forget about it. Unfortunately, including the framework designers.
Somehow, someone thought that dataProvider property will be always set first, and selectedItem second. This is wrong. Here is an example how it could be in a real world:

A multi-state Panel has Combobox. The Panel is cached (after creation it is saved to be reused in later calls).

Here is the sequence of events:
1. Already cached Panel is opened up by PopUpManager
2. Panel is initialised with new data
3. selectedItem property is getting initialised first (by data binding)
4. Panel performs the state change
5. Panel in new state initialises ComboBox by setting the dataProvider (by databinding)

Because there are no new dataChange events coming after this point, selectedItem won’t be set.

Touché!

Here is the overridden class which addresses both problems.

On the downside FAR editor has a couple of problems which bothered me a lot:
- Internal editor which is quite powerful, sometimes starts using Windows line endings (CRLF) and because I am working in a mixed environment I had to run the dos2unix utility on the routine basis to convert all the files into the UNIX format.

- FAR has a mysterious feature of changing file permissions or file sharing by adding user “None” with special permissions which were preventing files from to be executed. I.e. you’ve download ZIP file using FAR’s FTP browser, unpack it, try to run EXE file, Windows says:

. So you have to Explorer->Security settings and kill this None user.

After my recent upgrade from XP to Vista these problems became even worse.
So, I tried to start using the Midnight Commander, which I’ve been using for ages in Unix environment.
But under Cygwin/Vista it behaves quite differently.

Problems I’ve met so far are:

- no mouse support under Vista.
I’ve even tried to use Cygwin’s SSH port with PuTTY and then run MC - no way to have mouse working in Cygwin window.

- I am unable to browse/open any .zip files.
MC says:

.
What the heck? Somehow it opens the TAR files, but only by F4(Edit) command. If you try to open TAR file it writes similar bullshit about running a binary file.

-And the last thing which drives me mad - I am unable to save some of the MC’s config files, such as .mc/bindings. Every time it accesses it, it sets up the sharing, adding this infamous user None with special permissions and this file becomes read-only for MC.

PS While writing this post, I have found that you can get a Midnight Commander’s subshell working under Cygwin - you have to explicitly enable the subshell support via parameter (even if it is a default):

To overcome this I used to a trick: right-click to get into the Flash Player menu -> click “Play” to shove the debugging session. Sometimes it worked, sometimes it was crashing in the guts of the SystemManager. So I had to restart the whole thing all over again.

It is appeared that the culprit for this freezing is an Anti-Phishing filter.
As soon as added my sites to the Trusted domain and in the IE settings disabled the anti-phishing filter it immediately stopped doing this nasty thing.

It is appeared that the culprit was a Tortoise SVN - it is already built on subversion 1.5, while Subclipse is using 1.4.x.

The problem appears everytime when you do anything using newer client, like Tortoise SVN.  Newer client  leaves a different footprint, so Subclipse is getting upset with that fact.

To overcome the problem I’ve added subversione client into the Cygwin port - it is still some relatively old version (1.4.6-3). Then I checked out the project via cygwin version of svn.

It also helps when only a part of the source tree is damaged by a newer client, so you can simply overwrite these bits from an tree checked out using older version.  Just be sure to copy modified files to some safe location first.

UPD  The final solution is to upgdade to Subclipse >=1.4.2 as it is using Subversion 1.5.x. The problem will be completely fixed.

The full installation instructions could be found here:
http://subclipse.tigris.org/install.html

My first bug submitted to Adobe was that one: the last comment on the page

Guess what happened in the next Flex2 hotfix? This bug was fixed. In the shortest period of time.
Was it mentioned in the Release Notes? No, it wasn’t.
As the result my app started to crash on List selection as if you process selected cells you have to rely on the order of indices.

ComboBox selection behavior was fixed in Flex3. Perfect!
Was it mentioned in the Release Notes? Nope. Crash, crash and one more crash!
Who cares?

Do you need to mention everything you fix in the Release Notes?
No. She’ll be right!

PS This is how losers do - they put every bug fixed in the Release Notes

This is so boring to read! It is much better to read some overexcited evangelist’s diary about new cool features.

[UPD] Looks like this thing bothers some other people as well.

I’ve submitted this bug in January 2008 and after 2 FlashPlayer releases there is no sign of a progress on this issue. FP 10 beta does not show any signs of the addressed problem either.

https://bugs.adobe.com/jira/browse/FP-67
https://bugs.adobe.com/jira/browse/SDK-14483

Adobe’s attitude is well described by Screenivas

Version 1.0 release : We are releasing a great product which would help our users in a great way by boosting their productivity. We would fix all serious bugs soon.

Customer: Great. Let me use it.

Version 2.0 release : We have lot of new features for you. In this release we have addressed all serious bugs (read it as bugs reported by great partners and volume customers) in the previous release.

Customer: But what about other serious bugs faced by me?

Company: We would fix all bugs soon.

Customer: They are only fixing bugs for the new features and few more bugs from volume customers.

Version 3.0 release : We have lot of new features. We have revelutionized the workflow. We have rewritten many parts of the code to boost performance. etc… Of-course now we are open-source !

Customer: Now it is Open-source ? Great! But what about bugs which were in 1.0 release?

Company: Oh! our customers have lived with these bugs for more than 2 years, now they have almost stopped complaining…..they can live with it! If they complain more, let us ask them to upgrade to the new version and follow a new workflow/API for the same feature which we have introduced. And anyway we are open-source let them grab the source and fix the bug themselves. We can’t spend our precious development time over fixing those old bugs.

Customer: Where I can find a developer and time to study this piece of open-source product and fix bugs for me? …?….?

There are 6385 unresolved or deferred bugs in the Flex bugs database. And they are now actively working on Flex4. This is so great! I could not wait for another version for Flex - this so exciting!

If you want to change the date using number of seconds it will cause a problem going over the dates when daylight savings change.

E.g.

private var mydate:Date = new Date(2008, 3, 4, 0, 0, 0, 0);

mydate = new Date (mydate.getTime() + 1000 * 60 * 60 * 24)

mydate = new Date(mydate);

If it goes over the date when daylight savings is changed, instead of adding 24 hours it will add 23 or 25 hours (depending on the time of the year - summer or winter)

So, to do perform Date manipulation correctly you have to use date field, i.e.:

This method correctly handles the daylight savings and it will keep initial 00:00:00 time.

PS https://bugs.adobe.com/jira/browse/SDK-14983

is not enough, you have to set

, otherwise it selects the first element. Despite

. Weird…

I’ll keep this blog mostly for whinging on Flex, Adobe and other stuff I am currently working with.

Also, it is good place to keep some technical notes & tips.

PS Technically speaking I am not a Pom but after 9 years in Australia I began to understand the roots of this phenomenon.

Cheers!