C Stringification: prepend __extension__ before the typedef with __int128

[hannesm]

This allows the generated C code to compile with gcc and -Wpedantic.

Suggested in #820 by @chjj

[JasonGross]

Thanks! Please also regenerate the C files (run make c-files or just make and let the build complete). Let's also add -Wpedantic to the cflags we test with by adding it at

fiat-crypto/Makefile

Line 498 in 7b25093

$(CC) -Wall -Wno-unused-function -Werror $(CFLAGS) -c $(ALL_C_FILES) $(EXTRA_C_FILES)

The current CI error is due to a packaging issue with my ppa which I hope will be fixed within an hour or so (but may take up to a couple of days).

[hannesm]

@JasonGross sure, done in 602069d (feel free to squash if that's preferred)

[JasonGross]

Thanks! This doesn't seem to work, though:

fiat-c/src/curve25519_64.c:24:31: error: ISO C does not support ‘__int128’ types [-Werror=pedantic]
 FIAT_EXTENSION typedef signed __int128 fiat_25519_int128;

[JasonGross]

Maybe you meant __GNUC__ rather than GNUC?

[hannesm]

indeed, sorry about that. I did not notice since I'm using on my development machine clang and there no such warning is issued, even with -Wpedantic. I fixed it in f3fc8ce (using __GNUC__).

[chjj]

It might be wise to have a different prefix for the extension macro in each file: fiat_p224_extension, fiat_p256_extension etc. Otherwise GCC will complain about macro redefinition if two files are included.

[hannesm]

sure, as you like. (running make c-files again and will push an update). I'm not entirely sure which C compilers fiat-crypto targets -- in both clang and gcc we can actually emit __extension__ unconditionally (they also both define __GNUC__)...

[JasonGross]

Thanks!