On Wed, Nov 27, 2019 at 10:52:02PM +0000, Ramsay Jones wrote:
>
> I decided to just test the 'luc/next' branch (commit 4a8aa8d1 cgcc: add
> support for riscv64). :-P
Hehe, quite wise :)
> I have only tested on 64-bit Linux and Cygwin (the sparse testsuite and
> running it over git), so far with no issues.
>
> I have also compared the output of:
> $ ./cgcc -dM -E - </dev/null | sort >sss
> $ gcc -dM -E - </dev/null | sort >ggg
> $ meld ggg sss
> on both Linux and cygwin (with similar results).
>
> I have ignored the 'float stuff', since the output of gcc and sparse
> is almost totally different! :(
Yes, the 'TYPE' and 'SIZEOF' predefines should be correct but
I ignore the remaining (which is generated by cgcc, not sparse
itself).
> The main difference, which is new, is the spelling of the 'type names'.
> e.g. __CHAR16_TYPE__ is given as 'short unsigned int' by gcc but
> 'unsigned short' by sparse. The following table shows the 'type name'
> differences:
>
> CHAR16_TYPE short unsigned int => unsigned short
> INT16_TYPE short int => short
> INT64_TYPE long int => long
> INTMAX_TYPE long int => long
> INTPTR_TYPE long int => long
> PTRDIFF_TYPE long int => long
> SIZE_TYPE long unsigned int => unsigned long
> UINT16_TYPE short unsigned int => unsigned short
> UINT64_TYPE long unsigned int => unsigned long
> UINTMAX_TYPE long unsigned int => unsigned long
> UINTPTR_TYPE long unsigned int => unsigned long
I was a bit surprised by the 'new' aspect as sparse itself
outputs these names since last December (IIRC) but yes
these were overwritten by cgcc until the patch that removed
cgcc's integer_types():
fba1931d2 ("cgcc: removed unneeded predefines for integers")
But, yes, they're different, sparse just uses show_typename()
for it. I've already wondered why GCC issues them like this
Well, I see that GCC's way inhibit something like:
INTMAX_TYPE double var;
so, maybe sparse should do the same for the predefines.
> sparse seems to '#define linux linux' rather than '#define linux 1'.
Funny, I wonder why. I never noticed.
I think that it should not be defined at all (and used also).
> sparse defines __LITTLE_ENDIAN__ but gcc does not.
Yes, indeed. Well, GCC defines it for some archs/OSes:
* on the *BSD
* on ppc64le
* probably on all platforms using big-endian by default
when -mlittle-endian is used.
> On cygwin, the results are similar to the above, with the addition
> of the following:
>
> WCHAR_TYPE short unsigned int => unsigned short
> Also, sparse defines __CYGWIN32__ when it shouldn't (this is on
> x86_64 cygwin, without -m32 etc).
Yes, I noticed a few days ago.
I tried to fix it but this part of cgcc is bitness agnostic.
Doing a #undef in sparse itself would be easier.
Thank you very much!
-- Luc
