Read a post the other day that said we would soon reach Peak Code (see ROUGH TYPE Peak Code? post). In his post, Nick Carr discussed a NBER paper (see Robots Are Us: Some Economics of Human Replacement, purchase required). The paper implied we will shortly reach peak code because of the proliferation of software reuse and durability which will lead to less of a need for software engineers/coders.
Peak code refers to a maximum amount of code produced in a year that will be reached at some point, afterwards, code production will decline.
Software durability, hogwash
Let’s dispense with the foolish first – durability. Having been a software engineer, and managed/supervised massive (>1MLoC) engineering developments over my 30 years in the industry, code is anything but durable. Fragile yes, durable no.
Code fixes beget other bugs, often more substantial than the original. System performance is under constant stress, lest the competition take your market share. Enhancements are a never ending software curse.
Furthermore, hardware changes constantly, as components go obsolete, new processors come online, IO changes, etc. One might think new hardware would be easy to accommodate. But you would be sadly mistaken.
New processors typically come with added enhancements beyond speed or memory size that need to be coded for. New IO busses often require significant code “improvements” to use effectively. New hardware today is moving to more cores, which makes software optimization even more difficult.
On all the projects I was on, code counts never decreased. This was mostly due to enhancements, bug fixes, hardware changes and performance improvements.
Software’s essential difference is that it is unbounded by any physical reality. Yes it has to fit in memory, yes it must execute instructions, yes it performs IO with physical devices/memory/busses. But these are just transient limitations, not physical boundaries. They all go away or change after the next generation hardware comes out, every 18 months or so.
So software grows to accommodate any change, any fix, any enhancement that can be dreamed up by man, women or beast. Software is inherently, not durable and is subject to too many changes which most often leads to increased fragility, not durability.
Software reuse, maybe
I am on less firm footing here. Code reuse is wonderful for functionality that has been done before. If adequate documentation exists, if interfaces are understandable, if you don’t mind including all the tag-along software needed to reuse the code, then reuse is certainly viable.
But, reusing software component often requires integration work, adding or modifying code to work with the module. Yes there may be less code to generate and potentially, validate/test. But, you still have to use the new function somewhere.
And Linux, OpenStack, Hadoop, et al, are readily reusable for organizations that need OS, cloud services or big data. But these things don’t operate in a vacuum. Somebody needs to code a Linux application that views, adds, changes or deletes data somewhere. Somebody needs to write that cloud service offering which runs under OpenStack that services and moves data across the network. Somebody needs to code up MapReduce, MapR or Spark modules to take unstructured data and do something with it.
Yes there are open source applications, cloud services, and MapReduce packages for standardized activities. But these are the easy, already done parts and seldom suffice in and of themselves for what needs to be done next. Often, even using these as is requires some modifications to run on your data, your applications, and in your environment.
So, does software reuse diminish requirements for new coding, yes. Does software reuse eliminate the need for new code, definitely not.
Coding Automation, yes
Coding automation could diminish the need for new software engineers/coders. However, this would be equivalent to human level artificial intelligence and would eliminate the need for coders/software engineers, if and when it becomes available. But if anything this would lead to a proliferation of ever more (automated) code, not less. So it’s not peak code as much as peak coders. Hopefully, I won’t see this transpire anytime soon.
So as far as I’m concerned peak code is never going to happen and when peak coders does happen, if ever we will have worse problems to contend with (see my post on Existential Threats).
Photo Credit(s): PDX Creative Coders by Bill Automata
2 thoughts on “Peak code, absurd”
Ray, absurd for definite. One piece of feedback I took away from SFD7 and other events was the number of vendors saying “we wrote our own file system” – presumably not a trivial task. Code is like prose, you don’t (and can’t legally) copy other people’s work. This statement is like saying because we have so many books in the world, writers will simply use sections from each other’s work to save time.
Chris, couldn’t agree more.
Comments are closed.