All 2 entries tagged Programming

Proper Planning and Preparation

There is an old, supposedly military adage, which reads

Proper Planning and Preparation Prevents Piss Poor Performance.

This applies to many things, including software development projects. There is also a popular idea known as Hofstadter's Law:

It always takes longer than you expect, even when you take into account Hofstadter's Law.

This is an amusing idea, but is sometimes allowed to apply to real life, which rapidly spirals out of control. There are many reasons a software project keeps taking longer the more you work on it, most of which are best avoided. This post is a brief introduction to how to plan, how to work, and how to keep things on track.

Software for Research

RSE is Research Software Engineering, which can mean many things, but ultimately means we are in the business of writing software used for doing research. Similarly, researchers-who-write-code should be aiming to for the least work which gets them reliable, verifiable, useable code. Note that usable includes being fast enough, and reliable includes doing enough testing.

This makes planning and scheduling essential, to make sure work gets done in time to produce research outputs, but also means it should be lightweight, simple and flexible.

Why Plans Fail

No plan survives contact with the enemy

(A well known paraphrase of a quote by H. von Moltke). Just like Hofstadter's law, one can add the corollary 'even if you include that in the plan'. Knowing the challenges ahead of time lets you plan around them, which gives the vast chance of something from the plan surviving. The following are the commonest sorts of issues to consider.

Shooting in the Dark

It's quite common to start a project not knowing exactly how to do it. This is fine, but has to be allowed for in the plan. For instance, allowing time to write a small test problem using a new library/tool before jumping into the main project, or allowing time at the end to rewrite/refactor (see below). Do not fall victim to the sunk costs fallacy and continue something unpromising just because you have already invested time and effort into it.

Scope Creep

Scope creep is probably the biggest problem many people face with research code. What starts as a simple script turns into a complex behemoth, and never seems to end. A vital part of planning, particularly if you are shooting in the dark, is to decide exactly what the code must do, and once that goal is reached, to stop. This is the perfect point to rejig things, debug well, document, optimise etc, before making a new plan to add all the things you now know you need. Endless rolling feature addition is risky.

Debug Deluge

It isn't uncommon to see a programmer writing pages of code before ever testing if it runs, or even compiles. This goes away with experience, but is often replaced by a related issue, of writing a bunch of stuff and then finding out that it doesn't quite work and has created a debugging nightmare. A better approach is to work in chunks, making each bit work, testing it, and only then integrating it into the whole. This way there's no deluge of debugging to be done, which can take a very long time; only small pieces at a time which are easier to plan for. This is related to the programming style called Test Driven Development (see below).

Premature Optimisation...

...is the root of all evil (Donald Knuth). Until your code works, there is nothing to optimise, as you'd just be doing the wrong thing faster. However, it is important to consider what the code needs to do in the end, and design and plan something that will be fast and efficient enough.

Refactor or Rewrite

Refactoring is rewriting pieces of a code so that the blocks (usually functions) give the same answers but are simpler, faster, more maintainable etc. Rewriting covers more substantial changes, such as changing core data structures or moving from hard-coded input values to a configuration file etc.

Getting Agile

If you have seen any of the Indiana Jones films, you may notice he applies something very like this. Rather than crafting a complex plan and trying to implement it in the face of changing circumstances, he adapts to things as they happen and is never more than a step, or a few steps ahead. This is basically the idea behind the 'Agile Methodology' of software development.

Since clients (or in our case, a research problem) change what they need over time, the idea is to plan only a few weeks ahead, adapting to changing needs as they arise. This is great for commercial development, which is all about happy customers, but does have a few pathologies for research software if done strictly. The main concerns are stopping yourself letting your scope creep out of control, and getting good enough time estimates to be sure you can complete a project, and that it will benefit you.

It's well worth reading a few articles about Agile to see what it's all about, and taking any bits you feel are useful for your own development.

Test Driven Developement (Lite)

'Test-driven-development' (TDD) means writing tests first, and considering a piece of code complete once it passes them, at which point it is done and work should stop. This has plenty of advantages, in particular that you avoid the pitfall of writing the test already knowing how the code works, which often leads to writing a test you know will pass.

The over-literal genie crops up all over the place and something very similar happens when you give clever people a set of rules. For a lot of scientific code, the natural test set is a handful of analytically soluble problems, for which exact answers are known. Using these as the test suite, the obvious, and strictly correct, TDD approach is just to hard-code their solutions. Every test passes, and the code is complete and usually completely useless too.

Rather than strict Test Driven Development, we'd recommend using writing tests for functions before or just after you finish them, making sure they work for all the parameter ranges etc that you need especially things like negative numbers, and then keeping the tests for when or if you change those functions. Rather than running everything every time you make changes you can just run the relevant tests and satisfy yourself that nothing is broken and then focus on integrating pieces together and making sure they all work in combination.

Planning for a Project

Everybody has their own way of planning, and really the only important thing is to do something so that you can estimate how long something will take and keep yourself honest along the way. For small things you only need to take an educated guess how many hours or weeks something will require. For more complex projects you may want to consider some kind of itemised planning process.

This file project_estimatorwexample.pdf is an example of such a planning worksheet, including an example based on my last project. This sheet tries to calculate a time requirement based on some common factors in research code. The multipliers and scalings are only a guess, and the last section adds an arbitrary buffer factor, but this would hopefully provide a starting point for making formal time estimates.

Last Word

This is quite an eclectic post, running through a few bits of useful knowledge that often get left out of programing introductions. The core point is simply

If you fail to plan, you are planning to fail!

Make a plan, stick to it as far as possible, but don't plan so far ahead that you're constantly wasting time re-forming your plans instead of getting work done!

Heather Ratcliffe : 20 Apr 2018 13:09 |  Tags: Code Programming |  Comments (0) |  Close comments |  Report a problem

Upcoming training opportunity

February 26-27, University of Leicester. Software Carpentry. A 2-day rapid introduction to basic programming and tools, primarily focussed on programming in Python.

Heather Ratcliffe : 01 Feb 2018 14:06 |  Tags: External Programming Training |  Comments (0) |  Close comments |  Report a problem