On 21.06.2023 23:12, Andrew Cooper wrote: > A different option would be to sort out the stdint types ahead of including > <asm/types.h>, which can either be done by introducing a <asm/stdint.h> or > upping the minimum compiler version for x86; a task which is massively > overdue.
As per my patch moving in this same direction, I strongly think this is the direction we want to take. (I don't see a reason to introduce asm/stdint.h for this - the logic you put in x86'es asm/types.h can well be put in xen/types.h as well, suitably limited by a slightly more involved #if than you have it now.) Personally I think using "mode" attributes for the fallbacks would be the more widely compatible way, but I'm not going to object to your approach avoiding them. It would feel more safe though if without using that attribute the fallback wasn't tied to __INT8_TYPE__ being defined (but instead directly to being on very old gcc). As to upping the minimal gcc version for x86: While I'm pretty sure I'm considered the hindering factor here, I continue to think that my "blocking" request isn't unreasonable at all: We want to establish clear criteria, by which we can then also go in the future. And we want to make clear what, if any, baseline requirements are acceptable to put up beyond consideration of just binutils and compilers (basic utilities, make, perl, python, etc). Jan
