On Wed, Jun 30, 2021 at 03:47:19PM +0000, Jacob Champion wrote:
> On Wed, 2021-06-30 at 11:03 -0400, Tom Lane wrote:
>> It does not look to me like json_errdetail can sensibly be used in
>> frontend, since it returns palloc'd strings in some paths and
>> constants in others. There'd be no way to avoid a memory leak
>> in a frontend usage. So I think the dependency on psprintf there
>> is not really a problem, but maybe we should make the entire function
>> "#ifndef FRONTEND" to clarify the intended usage and avoid building
>> useless code into clients.
That sounds sensible from here. One thing to be aware of is
json_parse_manifest() in pg_verifybackup that uses it, but we could
just replace the error by a plain "failed to parse manifest"".
Backup manifests are generated by the backend, so failures should not
happen there anyway.
> FWIW this is one of the fixes (patch 0002) in the JSON-for-libpq thread
> [1]. It ensures that all returned error strings are freeable by the
> caller. That in turn was the impetus for the asprintf port suggestion.
Yes.
> But until/unless that is changed, an #ifndef seems like a good way to
> prevent issues for the current code.
Sounds sensible to do that as well for 14 before the release. Any
thoughts about that?
--
Michael
