Re: cpluspluscheck vs ICU

Hi,

On 2022-03-23 08:56:17 -0700, Andres Freund wrote:
> On 2022-03-23 08:19:38 -0400, Andrew Dunstan wrote:
> > On 3/22/22 22:23, Andres Freund wrote:
> > That only helps when running the CI/cfbot setup. Fixing it for other
> > (manual or buildfarm) users would be nice. Luckily crake isn't building
> > with ICU.
>
> Oh, I agree we need to fix it properly. I just don't yet know how to - see the
> list of alternatives upthread. Seems no reason to hold up preventing further
> problems via CI / cfbot though.

I just hit this once more - and I figured out a fairly easy fix:

We just need a
  #ifndef U_DEFAULT_SHOW_DRAFT
  #define U_DEFAULT_SHOW_DRAFT 0
  #endif
before including unicode/ucol.h.

At first I was looking at
  #define U_SHOW_CPLUSPLUS_API 0
and
  #define U_HIDE_INTERNAL_API 1
which both work, but they are documented to be internal.


The reason for the #ifndef is that pg_locale.h might be included by .c files
that already included ICU headers, which then otherwise would cause macro
redefinition warnings. E.g. in formatting.c.

Alternatively we could emit U_DEFAULT_SHOW_DRAFT 0 into pg_config.h to avoid
that issue.


The only other thing I see is to do something like:

#ifdef USE_ICU
#ifdef __cplusplus
/* close extern "C", otherwise we'll get errors from within ICU */
}
#endif /* __cplusplus */

#include <unicode/ucol.h>

#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */

#endif /* USE_ICU */

which seems mighty ugly.


Regards,

Andres