Hacker News Reader

Comment by tannhaeuser | Hacker News Reader

It's cheap to portray COBOL programmers as cowboys, but COBOL and the ecosystem around it is actually one of the most organized and fit-for-purpose programming languages around even if it isn't modern by any means. Time will tell if banks and insurances will fare better with J2EE-based systems (meaning classic Java-based enterprise stacks such as EJBs and portlets running under shudder Websfear, which count as new-fangled compared to COBOL in these circles). I could imagine classic J2EE maintenance is being neglected as career path for millenials, just as COBOL is/was for the generation before. J2EE exposes much more escape hatches into system programming and rope to hang yourself, and at the same time will be around for a long time to come, since Java has been the go-to language for almost all big-time eCommerce and FinTech projects at banks/insurances since 2000.

Edit: a-ok, "cowboys" is part of name of that guy's consulting business

View on HN

I don't know for COBOL, but for JEE, we've been able to migrate out of weblogic (to jboss) and then from oracle to postgresql (on code which is 10 years old). So even if JEE might be hard to maintain, it allows some flexibility in the base tools you choose. (my view of COBOL is that once you're into a vendor, it's pretty hard to leave to another one because COBOL is close to the metal, I may be totally wrong)

but in the long run, the problem is also the business logic. What disappears is the knowledge of the hacks that were applied at the business logic level. And this is much harder than to recover from the code. So I guess one has to make the difference between business logic maintenance and the actual COBOL maintenance.

But having rewrote some huge COBOL app in Java, I can say that Java is better structured, at a framework complexity cost.

Better structured, easier to debug, and the tooling exists to incrementally refactor in a relatively safe way. There are also migration paths to better worlds. IntelliJ can auto-rewrite Java into Kotlin, preserving comments and formatting along the way, and even detecting common Java idioms that have better Kotlin equivalents and rewriting them too. So, not only can you migrate to better app containers, database engines, upgrade your garbage collector and compilers etc, but also even switch to a much more modern language nearly pain free.

Regardless of whether Kotlin is considered a better world than Java, I'm skeptical that this sort of automatic translation adds significant value.

I've been using Intellij's Java to Kotlin conversion a bit lately, and the code it spits out is fairly decent Kotlin, and with a pass through on nullability and switching out Java types with native Kotlin types afterwards. It gets just as good as it would get if I hand translated my self, but it takes a lot less time.

I suspect that part of the reason that this works well, is that Java and Kotlin isn't that dissimilar. Doing automated conversion between Java and Scheme for example, would be a lot harder to get right.

Hard-coded business logic: yuck. IIRC there's proprietary software (SDKs / integrations) specifically for specifying, running, testing and enforcing business rules efficiently so that applications just present an API or respond to events rather than having to know everything about the business, and so not having to change N apps when a business rule changes / track down weird bugs when a business rule changes because another system didn't get a change also.

Some people prefer to "hardcode" business logic in Java instead of hardcoding it in XML or whatever business logic specification language the proprietary tools use. At least you can hire Java programmers fairly cheaply.

I would pick hardcoding for most projects. You often end up creating a new DSL for your business logic, but because most of us aren’t language designers our DSLs end up being worse than what ever language the project is written in. Almost every time you will fail to correctly anticipate the future needs and your DSL won’t be sufficient anyway.

The code is only hard if you can't release it safely.

That quite deepnds on the business. Some of our data scenario are super complex (thousands of datapoints, where one influence the others depending on how they happen in time, depending on manual intervention, things that happen here and there). Because of that, the usual tools to write your business logic fall short on the testing part. They're good at writing/maintaining simple rules, versioning them, but when a time dimension is crucial, then well, it's not that good anymore.

TSB's migration from legacy to Java based was a complete omnishambles.

There were many HN articles on that...was it last year?

Oh how I loathe WebSphere.

Yes, they're definitely from Texas: not the usual "cowboy coders" == "coders whom take too much initiative / don't follow a plan / directions."

J2EE usually means GC pauses (hell), XML (hell again) and dependency injection (hell also). Thankfully, hells tend to pay well if someone is willing to schlep the muck around quickly and reliably.

JVM (IBM's, Oracle's) and CLR runtimes are battle-tested and much more widely deployed to run civilization than most non-enterprise developers realize.

Also, I should probably look that guy up as I'm in North Texas at the moment.

Can you explain why you describe dependency injection as "hell"? I prefer working with DI/IoC rather than without it.

Heh, not the op but I'll bite.

I just had to go through an open source project that uses guice, in order to try to trace a particular code path, since the user group was less than helpful.

Trying to figure what was running was an absolute nightmare of nested magic spread out through a score of files. Meaning as soon as I had a clue, I'd lose it as I had to retrace back through all those files.

I'm sure it's clear as crystal to the people who wrote it, but DI can be obtuse as hell if you have to figure what's actually running on your own.

I understand dependency injection to mean 'pass dependencies as arguments', which seems pretty simple and reasonable.

e.g. instead of

    void foo() {
        SingletonLogger::logError("no cheese");
    }

you'd write

    void foo(ILogger logger) {
        logger.logError("no cheese");
    }

Now it's much easier to unit test `foo` with a mock logger object.

Is there more to it than that? Why are there big, complicated frameworks for it?

DI/IoC gets complicated, fast.

hand rolling DI, yep brilliant, generally quite easy to work your way through because it's easy to see where classes are being used.

Hand rolling can get complicated when your applications get BIG and you want to do fancy things like having modular code (whether you ever use the module somewhere else or not) or lazy loading or multiple scopes.

Then you use a framework and you find your dependencies take dependencies, your wireup code is auto-magical, you inject dependencies into private properties, you want multiple implementations of particular interfaces but want to ensure only the right one gets used in each circumstance.

Then multiple people start getting their fingers into the pie and the DI/IoC explodes into config files throughout the application, and the only way to understand what's being wired up without a map, is to "find usages", pin files, and cry

(I'm not bitter or anything ;) DI/IoC can work wonders for handling complexity, it can also increase your complexity if you aren't very careful)

That last sentence is the important part: "if you aren't very careful". I work in multiple large java/spring codebases and the rule we use is "Keep all configuration related to the same base type/interface in the same file". That way if you want to find out which logger is injected you only have to check 1 file.

Another easy way to find out what is injected at runtime is to attach a debugger and set a breakpoint.

:( with the particular project I've been looking through, trying to debug via running the entry point was a non-starter. The part of the system I needed to trace would only be able to run if I had 4 other services running (all started seperately from the same main file) and a full dataset loaded.

Once I finally discovered a hint about which files governed the behaviour I was investigating I was finally able to isolate the tests to begin the debug cycle.

Sometimes it's just complicated.

Imagine you want to simplify the code, and you hate passing values around, so you write a framework that allows you to write something like:

    @GetMeALogger(andCallIt=loggy)
    void foo() {
        loggy.logError("no cheese");
    }

Where @GetMeALogger does some magic at compile time to pipe in some kind of logger class based on various config options. Then to unit test it, you use different config than production.

Mainly because once you get more than a few arguments, it gets tedious to start newing them all up and all of their dependencies and on down the stack. So some people built frameworks to do it for you but each of these frameworks requires their own configuration/documentation and suddenly instead of 50 lines of news, you have annotations in twenty different files that effect the way your code runs without actually being code. Each step to get from A to Z seems perfectly logical but the end result is a difficult place to start.

It's only a problem if you let it become a problem. Just like with code style, naming and formatting, the DI configuration should be enforced during code review.

Spaghetti code would not pass code review, so why would you allow spaghetti configuration.

In the code bases I work in I know if I need to know what Logger is injected I need to look at the LoggerConfiguration class to see which Logger implementations are used in production, staging and dev.

And even if I don't know the name of the LoggerConfiguration file, I have an IDE that can get me from injected property to configuration in one click.

But, your experienced in your teams code. Now, let's say you and your team get fired except maybe some person on another continent they hired to keep it up. OK, 20 years from now. Your reincarnated and you just got hired to dig through the code and you've hard of java, but nobody uses that relic anymore.. Good luck.

That would be a fun text adventure....

Declaring "I need an ILogger" is the easy part. The complicated part is the configuration related to which ILogger implementation will be injected at runtime.

This can depend on several things like configuration properties, environment variables, libraries loaded, feature flags enabled, etc

edit: also, you don't always want to manually pass the argument to every function. At application startup you typically want to build a full graph of objects with all the dependencies "wired up". The big complicated frameworks (Spring for example) read in a bunch of configuration, resolve all dependencies, and then construct the graph for you.

> I'm sure it's clear as crystal to the people who wrote it

This - in general - is the problem with programming.

Reply on news.ycombinator.com