RFR: 8295795: hsdis does not build with binutils 2.39+ [v3]

[Andrew Haley]

On Mon, 25 Sep 2023 12:01:41 GMT, Robbin Ehn <rehn at openjdk.org> wrote:

> > Maybe, but would it really be easier? I don't see that it would be _significantly_ easier, and it would be much more riscv community friendly to work on the shared disassembler.
> 
> The build would be easier, no deps at all and could be shipped as part of the release since we control license. (license not an issue with capstone) 

Building hsdis in a shippable form is key for me. As I see it, binutils is out of the question because its licence is incompatible, (IANAL) LLVM is too large, and only Capstone fits. 

> And testing as mentioned.

Well, yeah, but then if we had our own disassembler we'd have the cost too.

 > Just want to mention that RV _can_ have same OP code for multiple instructions (vendor specific),

Ouch ouch ouch no! That is the way of disaster. 

> binutils and llvm have the infrastructure to handle this, a quick peek at capstone made me think this might be an issue. Thus task may not be as trivial 'add missing' instructions. 

Yeah. These mostly look like kernel magic, though.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/15138#issuecomment-1735165891