On 11/03/2015 02:32 AM, Paul Eggert wrote: > Pedro Alves wrote: >> Instead of trying to detect the right types, detect >> good-enough-pre-C++11 stdint.h and in such case make the substitute >> stdint.h just wrap the system's stdint.h with >> __STDC_CONSTANT_MACROS/__STDC_LIMIT_MACROS defined. > > Instead, how about just adding the necessary #defines for > __STDC_CONSTANT_MACROS/__STDC_LIMIT_MACROS to config.h if they are needed to > fix > the bug? Then we can leave stdint.h alone (perhaps not even build stdint.h > at all).
The downside could be that config.h's #defines are unconditional, instead of guarded against multiple definition with #ifndef, as they are currently in the replacement stdint.h. I was worried (a bit) that that may break packages that have build systems that pass something like -D__STDC_CONSTANT_MACROS on the command line. But because -DFOO defines FOO as 1, this doesn't error out: echo "#define FOO 1" | g++ -Werror -DFOO -E - Given that config.h defines symbols to 1, I guess that concern is moot. (But that speaks nothing of other compilers gnulib supports, naturally.) I'll update the patch in that direction. Thanks, Pedro Alves
