Category: Software Development

Diving into Hibernate’s Query Cache behaviour

Hibernate is a very sophisticated OR-Mapper and as such has some overhead for certain usage patterns or raw queries. Through proper usage of caches (hibernates featured a L1, L2 cache and a query cache) you can get both performance and convenience if everything fits together. When trying to get more of our persistence layer we performed some tests with the query cache to be able to decide if it is worth using for us. We were puzzled by the behaviour in our test case: Despite everything configured properly we never had any cache hits into our query cache using the following query-sequence:

  1. Transaction start
  2. Execute query
  3. Update a table touched by query
  4. Execute query
  5. Execute query
  6. Transaction end

We would expect that step 5 would be a cache hit but in our case it was not. So we dived into the source of the used hibernate release (the 3.3.1 bundled with grails 1.3.5) and browsed the hibernate issue tracker. We found the issue and correlated it to the issues HHH-3339 and HHH-5210. Since the fix was simpler than upgrading grails to a new hibernate release we fixed the issue and replaced the jar in our environment. So far, so good, but in our test step 5 still refused to produce a cache hit. Using the debugger strangely enough provided us a cache hit when analyzing the state of the cache and everything. After some more brooding and some println()'s and sleep()‘s we found the reason for the observed behaviour in the UpdateTimestampsCache (yes, yet another cache!):

	public synchronized void preinvalidate(Serializable[] spaces) throws CacheException {
		//TODO: to handle concurrent writes correctly, this should return a Lock to the client
		Long ts = new Long( region.nextTimestamp() + region.getTimeout() );
		for ( int i=0; i
			if ( log.isDebugEnabled() ) {
				log.debug( "Pre-invalidating space [" + spaces[i] + "]" );
			}
			//put() has nowait semantics, is this really appropriate?
			//note that it needs to be async replication, never local or sync
			region.put( spaces[i], ts );
		}
		//TODO: return new Lock(ts);
	}

The innocently looking statement region.nextTimestamp() + region.getTimeout() essentially means that the query cache for a certain “region” (e.g. a table in simple cases) is “invalid” (read: disabled) for some “timeout” period or until the end of the transaction. This period defaults to 60 seconds (yes, one minute!) and renders the query cache useless within a transaction. For many use cases this may not be a problem but our write heavy application really suffers because it works on very few different tables and thus query caching has no effect. We are still discussing ways to leverage hibernates caches to improve the performance of our app.

An advent of unconditional quality code

A four-week experiment dealing with conditional statements and how to avoid or replace them. Starting at the first advent, the experiment runs until christmas. We invite you to join and share your experiences.

This blog entry invites you to an experiment in code. It’s an experiment that runs four weeks and can be performed secretly even at your workplace. It might improve the way you think about conditional statements in an object oriented programming language. You don’t need any special hardware or setup, just the will to change your coding style a bit each week.

The experiment

Beginning with this year’s advent (a season of the christian religion), you are asked to omit one type of conditional statement each week while programming your regular code. The omitted statements add up, so that you have to spare four different statements in the week before christmas. There is no relation to christmas (or religion) other than it’s a four week period at the end of the year, which is the perfect timeframe for the experiment. And you might buy yourself a little present for christmas if you succeeded at the experiment (idea: a new programming book).

The four stages

For every stage, you are asked to write your normal code without a specific statement. It is perfectly valid to use semantically equivalent code constructs to achieve the same goal. This experiment is even more successful if you are creative and diversified in your variations of the original statement. Remember that the stages add up. On the fourth stage, you are asked to use none of the statements mentioned below.

  • Stage 1 (first week): Don’t use “else”
  • Stage 2 (second week): Don’t use the conditional operator “?:”
  • Stage 3 (third week): Don’t use “switch”
  • Stage 4 (fourth week): Don’t use “if”

You are not asked to change existing code to conform to these restrictions, except you need to work on the lines that contain the prohibited statements. You should apply the rules to your new code rigorously, though.

Explanation of stage 1 (Don’t use “else”)

This rule bans all the different occurrences of the else-branch to your if-statements. It includes every “else if” or “elsif” your programming language might provide. The rationale behind the rule can be found in the Object Calisthenics, rule #2 by Jeff Bay. Here is an explanation of it by Being Cellfish.

Explanation of stage 2 (Don’t use the conditional operator “?:”)

Elvis is dead. Let this resemblance to his hairdo rest for a week, too. It contains a hidden else statement that is restricted since stage 1. Another rationale is that the conditional operator isn’t very easy to read/grasp if stretched out a long line.

Explanation of stage 3 (Don’t use “switch”)

A switch (or case, or select) statement is nothing but a big if-else cascade. It’s handy sometimes, but can be replaced by a lookup table (like a hashmap) virtually everytime . In Martin Fowler’s book “Refactoring”, the switch statement counts as its own code smell category. You should try to live without it for a week. If you need inspiration, try this article on how to avoid it.

Explanation of stage 4 (Don’t use “if”)

Yes, you didn’t misread. There is a whole campaign that tries to avoid the if-statement altogether. Read their website for inspiration on how to survive this week. Maybe you might make new friends with polymorphism and some other implicit conditional structures. Remember, this is a short week just before christmas. Try it, you might be surprised how easy it looks with hindsight.

Ready, steady, go!

This experiment starts with the first advent at Sunday, 28.11.2010. Every stage lasts for one week and adds up to the previous stages. The experiment ends at christmas.

Good luck! And if you’re done with it, drop us a comment with your experiences.

Bug hunting fun with std::sort

Small errors in custom comparison functions used with std::sort can lead to hard-to-find bugs.

The other day I came across a nice little C++-shoot-yourself-in-the-foot at one of our customers. Let’s see how fast you can spot the problem. The following code crashes with segmentation fault sometime, somewhere in the sort call (line 31).

#include <iostream>
#include <vector>
#include <boost/shared_ptr.hpp>
#include <boost/bind.hpp>

using namespace std;
using namespace boost;

enum SORT_ORDER
{
  SORT_ORDER_ASCENDING,
  SORT_ORDER_DESCENDING
};

bool compareValues(const std::string& valueLeft,
                   const std::string& valueRight,
                   SORT_ORDER order)
{
  const bool compareResult = (valueLeft < valueRight);
  if (order == SORT_ORDER_DESCENDING) {
    return !compareResult;
  }
  return compareResult;
}

int main(int argc, char *argv[])
{
  std::vector<std::string> strValues(300);
  std::fill(strValues.begin(), strValues.end(),
            "Hallo");
  std::sort(strValues.begin(), strValues.end(),
            bind(compareValues, _1, _2, SORT_ORDER_DESCENDING));
  return EXIT_SUCCESS;
}

Any ideas? The tricky thing about this bug is that the stacktrace output in the debugger gives absolutely no hint at all about its cause. And this is a simplified version of the real code which has to sort boost::shared_ptrs instead of strings. Believe me, you don’t want to see that stacktrace. Because of the use of boost::bind together with boost::shared_ptrs it looks, well, let’s say intimidating.

Still no idea?

I’ll give you a hint. If the SORT_ORDER is set to SORT_ORDER_ASCENDING everything is fine. …

Ok, the problem is that std::sort algorithm must be given a comparison function (object) that defines a strict weak ordering on the elements that are to be sorted. In other words the comparison function object must implement the ‘<‘ (less than) relationship on the elements.

Unfortunately, lines 20 to 22 break this ordering when SORT_ORDER_DESCENDING is given. The initial idea of this code was that, well, if compareResult gets returned on ascending sort order, lets just return the negation of it when the “negation” of acscending order is requested. This, of course, destroys the strict weak ordering requirement because whenever valueLeft == valueRight, the function returns true, meaning instead that valueLeft < valueRight. And this somehow wreaks havoc inside std::sort.

A better version of the function could be:

...
bool compareValues(const std::string& valueLeft,
                   const std::string& valueRight,
                   SORT_ORDER order)
{
  // solution: return false independent of sort order
  // whenever valueLeft == valueRight
  if (valueLeft == valueRight) {
    return false;
  }
  const bool compareResult = (valueLeft < valueRight);
  if (order == SORT_ORDER_DESCENDING) {
    return !compareResult;
  }
  return compareResult;
}
...

The really annoying thing about this whole issue is that std::sort just randomly crashes with a stack trace that shows nothing but some weird memory corruption going on. After the initial shock, this sends you down the complete wrong bug hunting road where you start looking for spots where memory could be overwritten or the like.

So beware of custom comparison functions or function objects. They might look innocent and easy, but they can give you lot’s of headaches.

Statement against public fields in Java

Every once in a while I talk to people about coding style and sooner or later there is discussion about public fields and getters/setters in Java. I would like to elaborate my opinion on this issue in addition to other quite well balanced articles to a broader audience.

First I want to differentiate properties of a class from other fields/member variables. Properties are fields, whose values are useful and important to clients of the class. We consciously decide to break encapsulation here and provide this data to our clients. The size of a collection may serve as a nice example. Fields on the other hand store state or dependencies our class needs to be fully operational. Datastructures like arrays, data access objects (DAOs) or some kind of notification service may serve as examples here.

The internal implementation of both, properties and fields, should never be exposed because this truely breaks encapsulation and takes away the freedom of the class implementor to change their implementation. At a later time you may decide to compute a value or read it from a database instead of storing it directly . On the other hand properties themselves may well be public and belong to the API of our class.

Now on to Java. There is no native property support in the Java language as it does not support the uniform access principle using language constructs. In other languages like Python, Ruby, Groovy or Scala you can change from direct field access to accessor methods without changing the clients, so it is no problem to expose fields (or more precisely properties) and thus make them public or protected. To gain the same degree of freedom in Java you have to emulate properties by using the getter/setter convention of Java Beans. You have to trade conciseness of public fields against this freedom and you really should do it. An IDE can generate the accessors and fold the methods away from your sight. The cost of getters/setters is really negligible.

Now we can derive the conclusions for Java programers. With each member variable you introduce you have to decide if it is a property or just some internal field. For properties you may provide getters and/or setters with appropriate visibility when needed. That means you should not provide accessor methods for all of your fields. In general you should never expose fields directly and all instance variables should be private. Not doing so will remove the freedom to change class internals without affecting the clients. Once a class with exposed internals is published as part of an API it is almost impossible to change internal design decisions.

Code Camp Experiences

Experiences gained when performing a two day code camp with a team.

Some weeks ago, I conducted a code camp with a team of twelve developers that build a software product together for years now. The team had already introduced sporadic code reviews (in the team vs. author review style), so the main emphasize of the meeting was to improve team coherence by writing code together while generally having some time off project. In this article, I describe what I had planned, what happened and what the effects are so far.

A plan for the code camp

The code camp was scheduled for two consecutive days when the whole team gathers in one room with one computer for each pair. We would switch pairs (and seats) after each iteration, with one iteration being 45 minutes coding time followed by 2-3 minutes presentation of the achievement to the team. With a recreation break of 20-30 minutes, this means one iteration every two hours.

Every iteration starts from scratch, without access to previous code fragments (see also the code retreat concept). This had several reasons: I wanted the iterations to be comparable. Some of the insights I wanted to share are dependent on directly assimilable experiences. The iterations should also be independent, without ballast from previous sessions.

On the first day, we started with a given code resembling a little puzzle game in Java Swing. The code worked, but had some bugs and was written in an awful manner. It was unknown code for the team with no emotional attachments. The assignment for each iteration was to refactor the code to something equivalently working, but much “better”. How this “better” is defined was up to the teams.

On the second day, we started with a blank editor and had the task to code the same little puzzle game (in Java Swing) we refactored the day before. Even with some practice, it was nearly impossible to finish within time, so concentration on the most important key aspects of the code was crucial. The main lesson here was to “create from scratch” rather than “fix the existing”.

What really happened

Day one

The camp started with the usual delay for setting up all the computers in a uniform manner. This couldn’t be prepared beforehands, as the computers were in use by another group. When we installed our software, we found the hotkey configuration of the whole system severely flawed (for example, Ctrl+1 was defined as “set keyboard layout to traditional chinese”).

To warm up for the first iteration, I presented the existing code and explained its structure in detail. The code was slightly too much to remember it all in one pass, so only a rough understanding remained. Every team had to examine the code again during their work.

After the warm up, the first iteration started, with everybody buzzing over the code. The 45 minutes went by really fast and the first presentations focussed on local improvements. No team had restructured the code in any meaningful way, but every solution was perceived as “better” than the original. One team failed to get their refactored code to work again.

The second iteration held the biggest surprise of the whole camp. The 45 minutes flew by and the presentations showed the difference. One team failed to work on the assignment, but every other team presented a solution that was far superior of the original code. Some teams restructured the code to an extend where the original structure wasn’t recognizable anymore. The distinction between the first and the second iteration was so great that everybody was baffled by what could be achieved in 45 minutes when you do it for the second time.

The third iteration added some interesting twists on the best solutions of the second iteration, but didn’t produce the massive boost in productivity and code quality. Everybody felt worn out afterwards, so we decided to close the coding part of this day.

While working with the Java Swing code, nobody on the team noticed the threading flaws in the code. When I pointed this out, I was asked to explain the mechanics of the Swing threading model. The team develops a web application and hasn’t much exposure to desktop application development, let alone with Java Swing. So we ended the day with a lecture about the EDT, the EventQueue and the SwingWorker.

The whole team strolled to a bar to share some beers afterwards.

Day two

After a short night, we gathered early in the morning to continue the coding part of the camp. I explained the task (develop the game from scratch) and off we went.

The first iteration yielded only very rudimentary results. One team started with an UML diagram of the application structure and had to stop after setting up the outline of every method in code. Most other teams started with the domain model and failed to attach the GUI part of the application. All solutions had similar concepts in mind, no team used test driven development or other “advanced” techniques.

As a result of the poor performance, we decided to change the rules. Instead of scrapping the whole code, every new team could take over the code base of any other team, as long as it wasn’t the own. In the second iteration, we completed the drafted solutions of the previous team. This didn’t work out, too. The teams were frustrated by their lack of results.

We changed the plan again and held a prolonged review discussion of the code camp instead of a third iteration. This was by far the better choice with hindsight.

The effects

The code camp was perceived very positively by the attendees. The main goal of the camp was not about coding, but about team coherence and team focus. We learnt a lot about the personal style and abilities of each team member because the code samples shown in the code review were directly comparable. And we revealed team problems we weren’t aware of yet, but some problems we thought would arise did not. This was a very healthy process, because some of these issues can be addressed directly now.

The side benefit of the camp, as stated by one programmer was the increased awareness that “throwing away your code and starting over isn’t as hurtful as I thought”.

Every attendee stated that they want another code camp soon.

Personal summary

The code camp greatly improved my sense for the team and for the individual team members. By sharing a common code base and performing the same tasks, I could directly see (and feel) their thoughts and abilities. The camp is a powerful way to get to really know a team. If you have to mentor a whole new team, consider performing a code camp to get in touch with them.

Developing Grails Apps – Some Dark Sides

Most of the time, developing Grails apps is a nice experience. But there are also dark sides. One of which is when bugs do appear or do not appear depending on how you started your app.

Usually, I try to avoid it but this time a Disclaimer is in order: This is not a Grails rant. Most of the time developing Grails projects is fast and smooth. Using Grails brought many advantages for us. But there are also dark sides…

My main criticism is that Grails abstractions are more than leaky! In every list of examples for the definition of the term Leaky Abstraction Grails should be top. As soon as you leave the tutorial/scaffolding/helloworld level you have to know a lot about the underlying stack. And with Hibernate and Spring neither of the words small, easy and lightweight do apply.

GORM, too, is only easy to use at first sight. The very informative blog series about GORM gotchas should absolutely become part of the user guide or the refence docs.

And there are those times where it gets really unpleasant. This is e.g. when a bug does appear in your grails application running in a servlet container (packaged in a .war)  but does not appear when the application is started from within the IDE. Our last one of those was a naming conflict in a .gsp file. The controller handed a model like this to the .gsp:

...
return [fieldValue: 'THE_VALUE', ...]

The model entry ‘fieldValue’ was used in the .gsp to set the value of a combo box. Unfortunately, ‘fieldValue’ is also the name of a built-in Grails tag …

Admittedly, ‘fieldValue’ was not the wisest choice of names and I would certainly expect to get scolded loudly by Grails for that – ideally with a nice descriptive exception. But what happend instead led to a loud scolding of Grails from us. And to some big question marks: What is the difference between executing Grails from the IDE and within a servlet container with respect to naming resolution? Why is there a difference, at all?

We had a hard time figuring out this one, not least because the error message was not very telling. And since this was not the first of those works-in-the-IDE-but-not-in-a-real-environment bugs there is always this slightly uneasy feeling…

As I said in the beginning, most of the time developing Grails applications is nice and shiny. I would not support their slogan, though. My personal search for the best web development tool is definitively not over.

How about your search?

Combine cobertura with the awesomeness of crap4j

Want the awesomeness of crap4j without running your tests twice in your build? Just combine it with your cobertura data using crapertura.

You may have heard of crap4j when it was still actively developed. Crap4j is a software metric that points you to “crappy” methods in your projects by combining cyclomatic complexity numbers with test coverage data. The rationale is that overly complex code can only be tamed by rigorous testing or it will quickly reduce to an unmaintainable mess – the feared “rotten code” or “crappy code”, as Alberto Savoia and Bob Evans, the creators of crap4j would put it. The crap4j metric soon became our most important number for every project. It’s highly significant, yet easy to grasp and mandates a healthy coding style.

Some enhancements to crap4j

Crap4j got even better when we developed our own custom enhancements to it, like the CrapMap or the crap4j hudson plugin. We have a tool that formats the crap4j data like cobertura’s report, too.

A minor imperfection

The only thing that always bugged me when using crap4j inside our continuous integration build cycle was that at least half the data was already gathered. Cobertura calculates the code coverage of our tests right before crap4j does the same again. Wouldn’t it be great if the result of the first analysis could be re-used for the crap metric to save effort and time?

Different types of coverage

Soon, I learnt that crap4j uses the “path coverage” to combine it with the complexity of a method. This is perfectly reasonable given that the complexity determines the number of different pathes through the method. Cobertura only determines the “line coverage” and “branch coverage”. As it stands, you can’t use the cobertura data for crap4j because they represent different approaches to measure coverage. That’s still true and probably will be for a long time. But the allurement of the shortcut approach was too high for me to resist. I just tried it out one day to see the real difference.

A different metric

So, here it is, our new metric, heavily inspired by crap4j. I just took the line and branch coverage for every method and multiplied them. If you happen to have a perfect coverage (1.0 on both numbers), it stays perfect. If you only have 75% coverage on both numbers, it will result in a “crapertura coverage” of 56,25%. Then I fed this new coverage data into crap4j and compared the result with the original data. Well, it works on my project.

Presenting crapertura

Encouraged by this result, I wrote a complete ant task that acts similar to the original crap4j ant task. You can nearly use it as a drop-in replacement, given that the cobertura XML report file is already present. Here is an example ant call:


<crapertura
coberturaReportFile="/path/to/cobertura/coverage.xml"
targetDirectory="/where/to/place/the/crap4j/report"
classesDirectory="/your/unarchived/project/class/files"
/>

It will output the usual crap4j report files to the given target directory. Please note that even if it looks like crap4j data, it’s a different metric and should be treated as such. Therefore, online comparison of numbers is disabled.

The whole project is published on github. Feel free to browse the code and compile it for yourself. If you want a binary release, you might grab the latest jar from our download server.

The complete usage guide can be found on the github page or inside the project. If you have questions or issues, please use the comment section here.

Conclusion

If crapertura is able to give you nearly the numbers that crap4j gave you is up to your project, really. Our test project contained over 20k methods, but very little crap. The difference between crap4j and crapertura was negligible. Both metrics basically identified the same methods as being crappy. Your mileage may vary, though. If that’s the case, let us know. If your experience is like ours, you’ve just saved some time in your build cycle without sacrificing quality.

GORM Gotchas: Validation and hasMany

Using validation on the end of hasMany associations yields unexpected results.

The excellent GORM Gotchas Series inspired me to write about a gotcha I found recently.
You have a domain class Container which contains elements:

class Container {
  static hasMany = [elements:Element]

  List<Element> elements
}

and the element has a constraint:

class Element {
  static belongsTo = [container:Container]

  static constraints = {
    email(email: true)
  }

  String email
}

When you try to save a container with more than one element that fails validation, only the first error appears:

Container c = new Container()
c.addToElements(new Element(email: "a"))
c.addToElements(new Element(email: "b"))
c.save()
assertEquals(2, c.errors.allErrors.size()) // fails, only one error is recorded!

The solution described in the docs coined with In some situations (unusual situations)) is to use a custom validator in the container class:

class Container {

  static constraints = {
      elements(validator: { List val, Container obj ->
          val.each {Element element ->
            if(! element.validate()) {
              element.errors.allErrors.each { error->
                obj.errors.rejectValue(
                      'elements',
                      error.getCode(),
                      error.getArguments(),
                      error.getDefaultMessage()
                )
              }
            }
          }
          return true
      })
  }

  static hasMany = [elements:Element]

  List<Element> elements
}

Responsive Qt GUIs – Threading with Qt

Qt4 used to have only primitive threading support. Starting with version 4.4 new classes and functions makes your threading life a lot easier. So in case you haven’t come around to look at those features, do it now!, it’s worth it.

If you have used Qt4 for some time now, specifically since pre 4.4 versions, you may or may not aware of the latest developments in the threading part of the library. This post shall be a reminder in case you didn’t follow the versions in detail or just didn’t get around to look closer and/or update.

In pre 4.4 versions, the only way to do threading was to use class QThread. Subclass QThread, implement the run method, and there you had your thread. This looks fine at first, but, taking the title of the post as example, it can get annoying very fast. Sometimes you have just few lines of code you want to keep away from the GUI thread because, e.g. they could potentially block on some communication socket. Subclassing QThread for every small little work package is not something you want to do, so I guess many users just wrote their own thread pool or the like.

Starting with version4.4. Qt gained two major threading features, for which, IMHO, the Qt people do not a very good job of advertising. The first is QThreadPool together with QRunnable. All Java programmers, which use java.lang.Runnable since the beginning, may have their laugh now, I’ll wait…

The second new threading feature is the QtConcurrent namespace (from the Qt documentation):

The QtConcurrent namespace provides high-level APIs that make it possible to write multi-threaded programs without using low-level threading primitives such as mutexes, read-write locks, wait conditions, or semaphores

Sounds great! What else?

QtConcurrent includes functional programming style APIs for parallel list prosessing, including a MapReduce and FilterReduce implementation for shared-memory (non-distributed) systems, and classes for managing asynchronous computations in GUI applications.

This is really great stuff. Functions like QtConcurrent::run together with QFuture<T> and QFutureWatcher<T> can simplify your threading code significantly. So, if you haven’t got around to look at those new classes by now, I can only advise you to do it immediately. Allocate a refactoring slot in your next Sprint to replace all those old QThread sub-classes by shiny new QRunnables or QtConcurrent functions. It’s worth it!

Let’s get back to the responsive GUIs example. In his Qt Quarterly article, Witold Wysota describes in detail every technical possibility to keep your GUI responsive. It’s a very good article which provides a lot of insights. He starts with manual event processing and mentions the QtConcurrent features only at the very end of the article. I suggest the following order of threading-solutions-to-consider:

  1. QtConcurrent functions
  2. QThreadPool + QRunnable
  3. rest

Stay responsive!