Wednesday, September 21, 2016

Delphi Features I Avoid Using And Believe Need to be Gradually Eliminated from Codebases

My guest post from L.V. didn't seem to have enough Delphi specifics for one commenter, so I thought about it, and realized, that what L.V. is talking about is Practices (stuff people do), not features.

But there are features in Delphi that I think are either over-used, or used inappropriately, or used indiscriminately, or which should almost never be used, since better alternatives almost always will exist.  Time for that list. No humorous guest-posting persona for this post, sorry, just my straight opinions.

1. WITH statement

This one is hardly surprising to be in the list, as it's one of the most controversial features in the Delphi language. I believe that it is almost always better to use a local variable with a short name, creating an unambiguous and readable statement, instead of using a WITH.  A double with is much more confusing than a single WITH, but all WITH statements in application layer code should be eliminated, over time, as you perform other bug fix and feature work on a codebase.

2. DFM Inheritance

I don't mind having TApplicationForm  inherit non-visually from a TApplicationBaseForm that doesn't have a DFM associated with it, but I find that maintenance and ongoing development of forms making use of DFM inheritance is problematic.  There can be crazy issues, and it's very difficult to make changes to an upstream form and understand all the potential problems downstream. This is especially true when a set of form inheritances grows larger.     I have even forced non-visual-inheritance using an interposing class, and found that IDE stability, and ease of working with a codebase is improved.

3. Frames

The problems with frames and with DFM-inherited are overlapping, but Frames have the additional troubling property of being hard to make visually fit and look good.  You can't really assume that any change in the base frame control's original positions will be overridden or not, you just don't know. Trying to move anything around in a frame is an exercise in frustration.  I prefer to compose parts of complex UIs at runtime instead of at designtime.

4. Visual Binding

I have had nothing but trouble with Visual Binding.  It seems that putting complex webs of things into a design-time environment is not a net win for readability, clarity, and maintainability. I would rather read completely readable code, and not deal with bindings.  Probably there are some small uses for visual binding, but I have not found them. My philosophy is to avoid it. It's a cool feature, when it works.  But the end result is as much fun as a mega-form.

5. Untyped Parameters in User Procedures or Functions

Modern Pascal should be done with PByte rather than the old way of handling "void *" types (if you know C) in Pascal is the untyped var syntax. If possible, I prefer to use PByte which I consider much more modern way of working.  I believe the two are more or less equivalent in capabilities, and that Delphi still contains untyped var params for historical compatibility reasons, but unless I'm writing a TStream and must overload a method that already has this signature, I prefer not to introduce any more anachronisms like that in my code.

6. Classic Pascal File IO Procedures

Streams should have replaced the use of AssignFile, Reset, Rewrite, CloseFile.

7. Unnecessary Use of Pointer Types and Operations in Application Layer Code

In low level component code, with unit tests, perhaps sometimes, pointer types and operations will be used. To implement your own linked list of value types which are not already implicitly by reference, but in application layer (form, data module) code that most Delphi shops spend 90% of their time, introducing raw pointer operations into the application code is almost always going to make me require it to be changed, if I'm doing a code-review.   Delphi is a compiled "somewhat strongly typed" language, and I'm most happy with application layer code that does not peel away the safety that the type system gives me.

8. Use of ShortString Types with Length Delimiters, in or out of Records

Perhaps in the 1980s, a pascal file of a record type, with packed records made sense. These days, it's a defect in your code.  The problem is once such a pattern is in your code, it's very difficult to remove it.  So while an existing legacy application may contain a lot of code like that, I believe a "no more" rule has to be set up, and module by module, the unsafe and unportable stuff will have to be retired, replaced, or updated.  The amount of pain this kind of thing causes in real codebases that I have seen that used it, is hard to overstate.

9. Use of Assignable (Non Constant) Consts

A compiler flag {$J+} in Delphi allows constants to be overwritten. It should never be used.




Tuesday, September 13, 2016

Delphi Worst Practices, The Path to the Dark Side

Guest Post from L.V.




If you want to do the worst possible job at being a Delphi developer, and go from merely weak, to positively devastating, and you want to give your employer the greatest chance of failing completely, making your users hate your product, and going out of business, while exacting the maximum amount of pain and suffering on all around you, if you wish everyone to fear your approaching footsteps, and to be powerless to cross you, here are some startlingly effective worst practices to consider.

Many require very little effort from you, other than occasionally putting your foot down and insisting that certain things are sacred and can't be changed, or that everything is bad and must be changed immediately, no matter what the cost.   It is important that the team never sense that they have the collective ability to go around you, and reinstate optimizations that undo your careful work to make things worse.  A strict hierarchical authoritarian power structure is key to maintaining steady progress towards pessimization.

No matter how bad things are, you can always find a way to make things a little worse.   I can't claim to have invented any of these, and I believe all of these are extremely popular techniques in Delphi shops around the world, and so it seems there is great interest in doing as bad a job as possible.  If I can contribute something to the art, it will be in synthesizing all the techniques of all the pessimization masters who have come before.

Now that you have considered whether you want to go there or not, I will share my secrets.
Here is the path that leads to the dark side...

1. Ignore Lots of Exception Types in the Delphi IDE

The more exceptions you ignore, the less aware of your actual runtime behaviors you will be.  Encourage other developers to ignore exceptions.  Suppress the desire to know what is going on, and become as detached as possible from reality.   The optimum practice is to ignore only EAbort and exceptions similar to it, like the Indy disconnect exception.  So the pessimum practice is to disable break on exception forever, or to add a very large number of classes to the delphi Exception Ignore.  Also make very sure that you ignore access violations.

2. Raise lots of  Exceptions, even for things which you didn't need an exception for.

This one is great, because you will annoy all developers and get them to ignore certain exception types.  Old code that uses StrToInt that could have used StrToIntDef will eventually make users ignore all manner of exceptions.

3. Try...Ignore

This worst-practice (or anti-pattern) can cause you more grief than any other worst practice:

   try
      MaybeDoAllOrPartOfSomeThing;
   except
   end;

To be maximally evil, don't even write a comment. Make every reader guess why you felt that not even logging an exception, and not even trying to restrict your exception handling to a specific sane type of thing to catch and ignore (like EAbort).  Make them wonder what kind of  evil things lurk below there, and how much memory corruption is being silently hidden.  Dare them to remove this kludge of doom that you have imposed.

4. Make your debug builds unable to ever run with Range Checking on, Overflow checking on, even if a developer wants to use it for a while.


While it can be a best practice to ship your release builds with Range Checking, and Overflow Checking off, because the effects to your customer of some relatively benign thing blowing up on them in release, that you can't predict or prevent, it can be a remarkably effective worst practice to build a giant codebase where you don't bother to explicitly turn OFF range checking and overflow checking and I/O checking where it's KNOWN to be generating false positives.     In codebases where I can turn on Range Checking and Overflow checking in particular, in my Developer Machine Debug builds, I often find my effectiveness in finding bugs is increased many times.  Those who want to pessimize their entire team's work, will want to make using such powerful tools that can be used for good, out of reach.

Note that turning on Range Checking and Overflow checking in Release builds could be a form of pessimization, because it's hard to guarantee that they won't have unknown effects.  Most of all, changing these defaults to anything other than what you've always had them at, is injecting a massive amount of chaos, and good developers will often state that this should be avoided in release builds.   You might be able to inject this kind of random evil chaos without anyone noticing, if for example, you can arrange for builds to be done on your machine instead of on a build server.  

5.  Permit Privileged Behavior By Developers with God-Like Egos

Unlike self-organized Agile teams, where rules apply the same to everybody, make at least one person on your team a God Like Developer, who can do things that other developers are not allowed to do.   Ugly pointer hackery, and evil kludges are okay, if you're this guy, and totally unacceptable if it's anybody else.   To really fully pessimize your team and your codebase, let this guy randomly refactor anything he wants to without asking anybody else's permission.  These God-Like developers can review other people's code, but don't need their code reviewed, because they never make mistakes.


6. Don't Document Anything

This is one of the easiest ways to pessimize, it requires basically no effort from you, and all things having to do with software teams and processes, will generally tend to rot on their own.  It is consequently one of the most popular ways of pessimization.  Sometimes you will need to quote the Agile Manifesto or people will accuse you of having evil motives. Quoting the Agile Manifesto will get these people to shut up.

7. Argue About Indentation

By now things are bad, and significant developer attention will be focused on improving things, undoing your careful work of Pessimization. Instead of letting the team focus on fixing core engineering mistakes and technical debt, redirect the team to consider more carefully the effects of one indentation style over another, and various formatting issues, or comment block styles.

8. Magical Unicorn Build Process, and the Voldemort Build Process

I call these special non-reproducible builds "Magical Unicorn Builds" because it is entirely possible that the one PC where the builds occur is actually the only place in the universe where the code as it lives now on version control actually will build.  The secrets and accidents of the entire projects history live as non-annotated non-recorded bits of state on that PC. Contents of the Registry. Contents of various folders that contain component source code that is not kept in source control, and will naturally tend to be slightly different on various machines, and there will be no way to assure that known and controlled set of input data created a traceable end product.   Lists of Tools that are required for the product to build will not exist, we don't need no stinking documentation.  For bonus Pessimization points, the build should not be done via a build.cmd batch script or a CI tool like FinalBuilder, but should instead require a bunch of Arcane and Undocumented actions performed Manually by the High Priest of the Dark Art of Building the Product.  In such a build, we may in fact get all the way to the Voldemort Build.   The Voldemort build is a secret known only to one developer, who we will call Voldemort. Voldemort knows arcane and terrible things that would make you weep, which must never be written down, or shared at all.  Only Voldemort knows the ultimate price of his own power, and he is willing to take any action to protect his own interests.

If you do all of these things, you may be very near being as bad as it is possible to be, and may become a Dark Lord some day.  It will take some hard work, but I'm sure you can do it. Go get 'em, tiger.

Please share your own worst practices in the comment box.  Together, we can rule the Galaxy.