Jeremy Dillworth's Blog

PyCharm, Charming Indeed

2010-11-21T21:44:00.000-08:00

At my day job, I use Eclipse for writing Java. On other projects, well, it's been complicated. At home I prefer Python and Django over Java and JSF.

In the past, I've cycled between Vim, Eclipse and NetBeans.

Pydev does a great job of supporting Python in Eclipse, but Eclipse itself can be aggravating at times. Firstly, I'm using it on a Macbook, and it eats battery. Secondly, I've had mixed results setting up the mercurial plugin. It had been a while since I had tried (I seem to remember things basically working last time) but on my most recent attempt at Eclipse, I needed to sign-up for an account at javaforge.com to install the plugin. "Annoying," I thought, "but at least it's free." Well, after signing up, Eclipse wouldn't let me agree to the license. I suppose I could have kept fighting with it, but Eclipse starts to feel like paperwork if you have to work too much at configuring it (I hate paperwork). The dialogs for installing/removing/updating plug-ins don't seem like they're well thought-out.

NetBeans, has support for Mercurial that was either easy to setup or out-of-the-box (I can't remember needing to install it, but I may be wrong). I really like how NetBeans highlights uncommitted changes on the left of the editor without doing anything (no need to do a Team > Compare ... as in Eclipse). Very nice. NetBeans also has a much nicer plug-in installation process. I had an OK experience with NetBeans while dabbling with Groovy & Grails earlier this year (would have enjoyed it more if code-completion wasn't abysmally slow) and considered giving NetBeans a try for Python. Unfortunately, it seems that the project to support Python in NetBeans has stalled or died.

These experiences being par for the course, mostly I've been hacking away using Vim. I've learned to really enjoy Ctrl+N (complete word) when you've got your whole project open in buffers. Typing ":b part-of-some-filename<tab>" isn't a bad way to navigate. Setting up indent-based folds and using zM (max folds) and zR (no folds) is a workable substitute for Eclipse's outline. Battery life is fantastic, and my fingers love Vim (ddp, %s/blah/what/, :w).

Still, I missed having something underline my mistakes, having a nice outline of the code, and having side-by-side diffs like those I get with Eclipse at work (where CVS is used, hence side-stepping plug-ins).

I decided to give PyCharm a try today. I'm pretty impressed so far.

PyCharm has a "Power Save Mode", and after watching my battery life while working un-plugged for nearly an hour, it seems to work well.

CVS, Git, Mercurial and Subversion are supported out of the box. Uncommitted changes are highlighted on the left of the editor, just like NetBeans.

The editor doesn't seem to have block selection like Vim. I've yet to find a way to change the color scheme. I miss Vim's key bindings, but it's early.

Copying and pasting code from a web2py project I'm rewriting using Django gave me a little trouble. PyCharm kept nagging me to import web2py symbols that I was busy replacing with the appropriate Django equivalents. It didn't actually interrupt me, I just kept seeing this little tool-tip giving me the key-combo to do an import (useful when I've needed it, but annoying in this case). I can probably switch this off.

Overall, I'd say JetBrains has a pretty good chance of getting some money from me by the time my trial runs out.

Cappuccino Book on the Way

2010-09-07T20:27:00.000-07:00

I was scanning some groups and noticed this thread on the objective-j list: http://bit.ly/c7E3wX.

Looks like the Pragmatic Programmers will be publishing a book on the Cappuccino framework! This is pretty cool. I'll have to get a copy when it comes out...

Fixing a GORM/HQL Group By Clause

2010-04-15T09:45:00.000-07:00

I was just trying to use a "group by" clause with some Grails domain classes in HQL and ran across the following cryptic message:

"Not in aggregate function or group by clause."

My models looked a little like this:

class Account {
  String name
  static hasMany = [entries:Entry]
}

class Entry {
  Date entryDate
  double amount
  static belongsTo = [account:Account]
}

I wanted to get a balance of all the accounts and ran an HQL query something like this:

def results = Entry.executeQuery(
  'select ent.account, sum(ent.amount) ' +
  'from Entry as ent group by ent.account')

I found that the reason for the failure seems to be that Hibernate adds all the other fields of Account to the select clause of the SQL under the hood. Changing the HQL to this, solved my problem:

def results = Entry.executeQuery(
  'select ent.account.id, sum(ent.amount) ' +
  'from Entry as ent group by ent.account')

On Web2py and Having Templates

2010-02-11T21:17:00.000-08:00

I've been writing an app for some friends using Web2py.

I really like how the generic views let you try out your controllers without worrying about the details of what variables you will or won't pass to the view. Lots of diagnostics are given in the generic views, and you can see how your forms render. Paired with web2py's automatic database creation and modification, you can iterate really quickly with this framework.

I've worked with some ORMs and ran into limitations when you try and do things that would normally require a nested SELECT in SQL. Web2py's DAL (database abstraction layer) can do nested selects relatively easily. The DAL is much more relational than object oriented.

In any event, implementing new features in a web2py app is like eating potato chips! It's hard to stop.

I've toyed with Grails a bit and similarly enjoyed the way the scaffolding makes it easy to move fast. The generic views in web2py also help one avoid that "blank canvas" feel.

I've been itching to do some more work with Django, and I think I'm going to have to put together my own Django skeleton project like they did at the Washington Times (although I'm not sure if I need to go as deep as getting into virtualenv just yet, just some templates and config options I habitually set).

IE 8 Canvas Workaround

2009-10-10T20:02:00.000-07:00

Here's what I learned since my last post.

Putting an element with no background over the shape, stroke and fill elements used internally by excanvas does not stop them picking up mouse events through that element. This can cause undesirable behavior as your invisible element doesn't receive the mouse events you expect. The shape/stroke/fill elements seem to steal mouse events from the invisible element. This happens even if you give them a lower zOrder than the invisible element.

I found 2 ways to solve this:

Accept it and give the shape, stroke and fill elements mouseover/out/move handlers to catch the events that bleed through to them.
Give the invisible element a background-color and a very low opacity. Then no events bleed through to the shape, stroke & fill elements. I used an opacity of 0.01 and couldn't see the element.

I've decided to go with door number 2. I'd rather make a couple of CSS changes than have extra event handlers.

Unfortunately, your mileage may vary.

IE8 Clipping Issue?

2009-10-09T22:32:00.000-07:00

I just solved a tricky issue. I don't 100% understand what just happened.

I'm working on my own custom area chart using jQuery and Google's excanvas. As the user moves their mouse over the chart, I wanted a partly transparent vertical line to trace the mouse movement across the chart. Later, I plan to add a little sign with the value.

To create my chart, I start with a div I call "container", then my custom jQuery plugin works as follows:

Add a canvas element and do the appropriate IE setup for excanvas to work
Draw a chart with canvas API calls
Set a mouseover handler on container which does the following:

Append 2 absolute-positioned divs to the container, so they go over the canvas. There's a "screen" div and a "tracker" div. The tracker div is the vertical line I want the user to see. It is positioned over the canvas. The screen div goes over the tracker div and the canvas. So, in Z-order from bottom to top, that's container to canvas to tracker to screen. The screen div has no background-color or image, so it's transparent.
I setup a mousemove handler on the screen div which moves the tracker div wherever the mouse goes
I set a mouseout handler on the screen div to remove tracker and screen divs

I found I needed the screen div because I originally had the container's mouseout event removing the tracker. Trouble is, since I was creating the tracker right under the mouse, the mouseout event for the container fired immediately because the tracker was between the mouse and the container. As a result, I got a loop of mouseout/over and the tracker would flash. I decided to put the screen div over everything to simplify matters by blocking mouse events.

I got it all working in Firefox and then tested in IE.

Odd thing was, when I got a mousemove event on the screen element, the event.pageX would be undefined sometimes (remember, I'm using jQuery, so it should be consistent). Also, my tracker div would go to the left (left:0px) on and off after I moved the mouse past a certain X coordinate. That doesn't make much sense from the code. The mousemove event must have been failing somehow.

Googling for various permutations of: "jQuery IE8 event.pageX undefined" yielded nothing.

I don't really understand what was happening, but I feel like it has something to do with the mouse events making it through the screen div. Testing confirmed this was happening, but I can't see why it would cause an issue.

Finally, I decided to give the screen a background color and a low opacity so I could see how the tracker moved. Suddenly everything started working!

I tried writing a minimal test page that I could post here to demonstrate the problem.

When I couldn't replicate it, I commented-out basically all the HTML on the page, and I commented out my fix. That replicated the problem on a mostly blank page. Then I did a "Save As" on the page and let IE save all the included JavaScript. When I opened the page from disk (instead of from the Django dev server) the issue went away.

I'm tempted to think it's a security issue, since it's disk vs. localhost. But since when do security settings change how events work?

Google Chrome Frame is starting to sound like a good idea.

This is going to bug me until I understand it.

UPDATE: I was able to produce a cut down test case. I'm still trying to isolate the cause of the tracker div's sporadic trips to the left. The problem goes away if the canvas is still present but I comment out all the calls to the 2D context. Maybe it's something to do with how Explorer canvas works?

UPDATE: Solved

SproutCore 1.0 Coming Soon

2009-10-07T15:43:00.000-07:00

I discovered SproutCore a while back and it didn't really seem all that interesting at the time.

The other day I was checking using Indeed's excellent jobtrends feature to watch the adoption of various JavaScript toolkits (jQuery vs Dojo vs GWT vs ExtJS vs Qooxdoo). I was mainly interested in the "heavier" frameworks which have support for lots of widgets but included jQuery for reference.

Thinking of the adoption rates of the above got me thinking about SproutCore again. So I went over to the project site and it seems that while I wasn't looking, the SproutCore guys have busily started moving towards a 1.0 release.

I spent a few hours playing with it. I had a little difficulty with some of the docs. Some of it looks out of date, other docs have 'coming soon' next to them. The docs site seemed to be down for much of the time I was playing.

Still, I feel like this looks really promising. I'm looking forward to trying to build something with it and see how things develop over the next few months.