fopencookie is not always available

[ffontaine]

Currently, tinycbor checks in Makefile.configure the availability of open_memstream in the toolchain to compile src/open_memstream.c. However, this file is also using fopencookie which is not always available as fopencookie is not a part of any standard (it's a GNU function and open_memstream is POSIX). If GLIBC is defined, the availability of this function should be checked.

Moreover, funopen is tested in Makefile.configure but never used in Makefile. Indeed, funopen is called depending on APPLE being defined, perhaps this test should be removed from Makefile.configure?

[thiagomacieira]

Just describe the problem you're having.

[ffontaine]

Tinycbor 0.3.2 with patch on unnamed union is failing on buildroot with bfin target. More info can be found here: https://patchwork.ozlabs.org/patch/659398/.

[ffontaine]

Here is more information: bfin target uses uclinux which has been configured without fopencookie, here is the stdio.h from uclinux:

#ifdef  __USE_GNU
#ifdef __UCLIBC_HAS_GLIBC_CUSTOM_STREAMS__
/* Create a new stream that refers to the given magic cookie,
   and uses the given functions for input and output.  */
extern FILE *fopencookie (void *__restrict __magic_cookie,
                          __const char *__restrict __modes,
                          _IO_cookie_io_functions_t __io_funcs) __THROW __wur;

/* Create a new stream that refers to a memory buffer.  */
extern FILE *fmemopen (void *__s, size_t __len, __const char *__modes)
  __THROW __wur;

/* Open a stream that writes into a malloc'd buffer that is expanded as
   necessary.  *BUFLOC and *SIZELOC are updated with the buffer's location
   and the number of characters written on fflush or fclose.  */
extern FILE *open_memstream (char **__restrict __bufloc,
                             size_t *__restrict __sizeloc) __THROW __wur;
#endif
#endif

I don't think it's possible to implement a version of open_memstream without fopencookie or funopen. So, I think we'll have to disable tinycbor for bfin target. You can close this issue.

[thiagomacieira]

Not really. You're right that the check was wrong. We checked for A in the Makefile, then for B in the source code. That should be fixed.

Also, you don't need to disable the entire library. You will only need to disable the part about converting to JSON. Though I can also accept a patch that degrades that functionality and returns an internal error or "not implemented" for when it would have been necessary.