Re: [Valgrind-developers] [PATCH v3] Fixes PLAT_amd64_win64 support

[Paul F. <pj...@wa...>]

On 12/12/2024 12:18, 罗勇刚(Yonggang Luo) wrote:
> 
> 
> On Thu, Dec 12, 2024 at 7:24 PM Mark Wielaard <ma...@kl...> wrote:
> >
> > On Thu, Dec 12, 2024 at 03:36:26PM +0800, 罗勇刚(Yonggang Luo) wrote:
> > > Any reviews? I think this can be get merged
> >
> > There were a couple of questions which I am not sure I saw an answer
> > to. Could you explain how you tested?  Why aren't you creating a new
> > section for win64?  You are adding it as a variant of 64bit
> This is already the section for win64/mingw
> 
> > linux/freebsd, but then need to change the data type of some, why is
> sizeof(unsigned long int) === 8*8 on linux/freebsd, but
> sizeof(unsigned long int) === 4*8 on PLAT_amd64_win64(mingw/msvc)
> mingw
> 
> > that OK for the existing platforms?  The patch Bart added for win32
> > used uintptr, which seems a better data type than long long.
> uintptr is fine, but I am not sure every compiler have ` uintptr_t`, that depends on header file
> 
> `unsigned long long it` is always safe for every compiler(without no need the header)

No it isn't.

All variations of "long long" were added with C99.

https://en.cppreference.com/w/c/language/arithmetic_types

I would be very surprised if there aren't people still using the Valgrind headers and compiling with c90. I think that the type used should either be compatible with the oldest compilers (along the lines Mark’s suggestion) or else there should be a separate section for win32 non-__GNUC__.

I'll have a look at Dr Memory. I think that it uses the same mechanism for client requests and also supports Windows.

A+

Paul