But at its logical conclusion, this idea puts up some serious practical barriers. Even if you only use an OS that you’ve personally audited, and even if you’ve gotten past the potential Thompsonian compiler issues, you’re still left with a minefield of possibly-adversarial pieces of software that might interfere with your computing experience.
Have you audited your BIOS/UEFI firmware [0]? Your CPU’s microcode [1]? Your WiFi/Bluetooth adapter’s firmware [2]? Your HDD controller’s firmware [3]?
I acknowledge that I’m advancing something of a “slippery slope” argument here, but I still care about the question. Where do you draw the line? Obviously being able to audit all of it yourself perfectly would be awesome. But to me that just seems impossible. So how much is enough?
EDIT: beaten by astrodust!
[0] https://www.pcworld.com/article/2948092/security/hacking-tea...
[1] https://www.dcddcc.com/docs/2014_paper_microcode.pdf
You can't win. Make compromises that are reasonable.
When it's so easy to use a community-vetted Linux distribution, I think "reasonable" gets shunted down a layer.
It's not all or nothing. Every reasonable step counts.
i wonder if this sort of confidence makes you less-safe. something like this should've prevented heartbleed, in theory, right?
You and your friend down there seemingly have more faith that a closed source process, what, would have identified this bug before it was deployed? Why? That it would have been found by good actors first? Why? That it would have been fixed promptly? Why? Announced with disclosure of its existence and ramifications? Why?
It's not a money question for this stuff either. There are hundreds if not thousands of paid engineers for known good-actors all working to break and responsibly disclose security issues in the same software stack I use.
The only thing you know about the comings and goings of closed source software is when public disclosure happens (and it's later fixed). Why would I be more confident in that process? It's not good enough.
Open-source is not inherently better or worse, it's all a matter of motivation on the part of the developers.
Recompile on compilers B and C and compare results. Or on CPUs Y and Z.
In reality, there will almost certainly be differences (different compiling algos, different CPU instruction sets), but if you can pinpoint behavioural distinctions or encoding distinctions which are not explained by known compile or instruction-set distinctions, well, you've got something interesting to explore.
The point being that you don't have to descend the entire stack of turtles. Just compare different stacks of turtles.
As much as I appriciate the sentiment, I have to admit I personally gave up trusting OSS for the soul reason that I / someone might be able to read and understand it.
https://news.ycombinator.com/item?id=15251891
https://developer.apple.com/opensource/