Simply reading the code doesn't work


5 min read

I read an article about not trusting third party code at

Quite frankly, it struck me as an off the rails rant, I can't remotely see how it got into the CP news list, unless popularity is the only requirement.

I can think of many reasons why it is just not possible or practical to read all the code you ship, as obvious as that idea might sound (particularly to a non-programmer). Here are a few:

  • You need to use a math library that takes advantage of processor instructions like SSE or SIMD that have no representation in pretty much any high level programming language. Even if it is, it is almost guaranteed to be handled by an underlying C library that has the assembly included in it. Does anyone on your team actually know how to read assembly?
  • You use a spectacularly large library like Spring or JPA, which are megabytes of code. Who has time to read all that? Can you even read that code?
    • Do you even know what Proxy and MethodHandle are?
    • Have you any idea how runtime byte code generation is being used?
    • Do you actually understand how generics work at runtime - eg, do you know what generic information is available at runtime, or how to get it?
    • Do you know what kind of circular infinite loop can be caused by proxying a default interface method? Or how to solve it with the InvokeSpecial byte code?
  • Are you even qualified to find security issues in the first place? If you're an average dev:
    • You've spent 0 hours, 0 minutes, and 0 seconds assessing security of code
    • At best, you run some scanner as part of a Jenkins job
    • You probably don't even know why encrypting passwords is completely bloody pointless
    • You get the same tired old OWASP top 10 training at every company, but have never actually read the OWASP in its entirety, never mind write insecure code, and learn how to exploit it and fix it
    • Do you get that blindly following the crowd and treating security as a tickbox is the EXACT OPPOSITE of any kind of real security?
  • Who pays for this time spent reading all that code?
    • What about - god bloody forbid - an update to that underlying SIMD/SSE C/Assembly library or gifreakingnormous Java library?
    • Can you assess the update in isolation, or do you need to reexamine a larger section of code around it?
    • How do you determine the bounds of the larger section of code around it?
    • And who pays for reassessing every update?
  • How do you keep a consistent quality of security via code reading as people come and go?
  • Is any old random assortment of junior, intermediate, and senior devs amenable to determining security of code?
  • What about the elephant in the room called licensing? What about all the proprietary code that open source libs may use, but cannot alter or reverse engineer? And therefore, you can't read it!

I'm sure given the time, I could write 6 more pages of reasons why this just isn't practical. Does that mean we should shrug our shoulders, say meh, and keep doing what we're doing?

I'm not a great fan of cynicism. I think the best option for supply chain security is to embrace small programming libraries, that you can, in fact, actually read. Look for libraries that:

  • Are just "straightforward" programming (no runtime code generation, no embedding of base64 text, no weird pragmas to turn off compiler options, etc)
  • Provide bang for buck with a small set of functionality
  • Are small enough to be practical to actually read it

Try to use as few of such libraries as possible. Combine this with a language that removes any uncalled functions when generating the executable, which further limits your exposure to any security issues you might have missed.

Now the question becomes: can I actually ship a real world application that works that way? Well there are some factors to consider:

  • You don't HAVE to use the most complex libraries possible that have as much as code as possible.
  • You don't HAVE to have a super complex frig-ton of Javascript happening in the browser. You can - brace yourself - actually use server side rendering with almost no javascript. My bank does this, and it is faster than sites that use client rendering.
  • You don't HAVE to constantly masturbate at every shiny new toy that appears on the horizon, just because it exists.
  • You CAN - believe it not - just keep using the same stuff. Even with all that fancy computer crap, car mechanics still use wrenches, and cars are still four wheels, a couple of pedals, and a steering wheel.
  • A lot of "improvements" are really not:
    • How many hundreds of hours were spent developing Lombok so devs could avoid spending seconds using their IDE to generate updated accessor methods, toString, and hashCode?
    • What is so great about VSCode compared to Eclipse? I'm forced to use VS Code because everyone else uses it, so some extensions I need are now only available for VS Code. I fail to see the improvement here.
    • Every new language requires rewriting all libraries again (SVG, CSV, math, http, ad infifreakinitum)

I would argue that more than anything else, this bloody obsession with new languages and libraries makes supply chain security next to impossible to keep on top of. In other words, you'd basically have to convince the entire damned developer world they are all doing it wrong, and what they should be doing instead. The first part of that argument is probably pretty easy, good luck with the second half!

While I agree with the authors rant that we are doing a poor job of security, and are simply lying in the bed we created, I will most emphatically disagree that some simple childish answer like "Just read all the code. Duh." is the solution.

I'm personally not convinced this problem can be solved by writing more code - only by writing less code, and by using processes geared towards security, not yet more candy for developers to rot their teeth with.

I can't honestly say my hopes are particularly high this will actually happen. If it does, it will only be because it is foisted upon us by others.