There’s a third possibility: they wanted a piece of software they could customize to meet their needs. Admittedly, for simple bugfixes and the like, Hex-Rays’ support is known for being quite responsive (as suits the small number of customers). There’s also a quite large (albeit poorly-documented and crash-prone) SDK, which can handle a wide variety of needs and has gained functionality over time. But if you want to add new functionality that can’t be implemented using the SDK? (That includes just about anything related to the decompiler.) You’re at the mercy of Ilfak’s priority list. Well, mere mortals are, at least; the NSA has enough money that it
might be able to set up some sort of special contract with Hex-Rays, but I suspect that’s easier said than done.
From that perspective, the ideal is what the NSA ended up with, a codebase whose development is fully in-house. Notably, though, second best would be to just have access to the source code of an existing tool, so you can at least make your own patches if necessary, even if you’re not in control of the codebase’s overall direction. Did the NSA ever seek that in IDA’s case, and could they have obtained it if they did? I don’t know the answer to either question… but source access certainly isn’t offered to typical customers. In general I’m surprised that “paid + source access for customers” isn’t a more popular model of software development.
From my perspective, which admittedly is very different from the NSA’s, I was never very interested in low-cost IDA competitors like Hopper or Binary Ninja, but I’m very excited about Ghidra. Why? Partly because it’s a more full-fledged competitor in terms of feature set, I admit – but the competitors I mentioned are bound to narrow the gap over time. Partly because of cost: I myself am at a point where I could justify the $600/y for Binary Ninja’s commercial edition, or even the order-of-magnitude-higher cost of the Hex-Rays decompilers, without wincing too badly. but I believe that reverse engineering should be accessible to beginners and amateurs. (Piracy is a partial solution, including in IDA’s case, but some people don’t like to do that.).
But the main reason I’m excited about Ghidra is that I have the source code. As a concrete example, I’ve spent a good amount of time reverse engineering software for the Nintendo Wii and Wii U. Both consoles have a main CPU based on the PowerPC architecture, but with a custom ISA extension for an extremely barebones version of SIMD. Well, both Hex-Rays and Ghidra support PowerPC decompilation (although that’s a relatively recent development), but unsurprisingly, neither of them have full support for that ISA extension. IDA actually does have built-in support for disassembling it, but AFAIK not for decompiling; Ghidra doesn’t seem to support it at all (but I may just need to configure it properly). What can I do? Well, in practice, nothing, because I don’t care about the Wii U anymore. But if Ghidra had been released a few years ago, I’m pretty sure I would have gone and implemented support for the extension myself; I haven’t looked at Ghidra’s source yet, but since it already supports other vector ISAs, it probably wouldn’t be that hard. With IDA, I was stuck. The SDK supports adding custom instruction sets for disassembly, but the decompiler SDK is so limited that supporting them there is either impossible or at least would be a huge hack.
And that’s just one of many customizations I‘ve wanted over the years. Some of them are probably easier said than implemented, but at least now I can put that to the test!