Re: pgsql: Check for GiST index tuples that don't fit on a page.

On 10/04/2014 05:10 PM, Alvaro Herrera wrote:
> Peter Eisentraut wrote:
>> On 10/4/14 12:59 AM, Tom Lane wrote:
>>> Peter Eisentraut <peter_e(at)gmx(dot)net> writes:
>>>> We don't have portable support for %zu formats until 9.4. The
>>>> backpatches to 9.3 and earlier need to use %lu.
>>> Good point --- and something that seems like it'll be an ongoing gotcha.
>>> Have we got any buildfarm critters that could be configured to throw
>>> errors about this?
>> If this change had contained a test, it would have caught it.

True. I didn't add a test because the error message depends on BLCKSZ
and alignment. It also didn't seem like a very likely issue to reoccur
in the same form.

>> Alternatively, we'd need to catch compiler warnings on the build farm,
>> but if the installation is too old to recognize %zu, who knows what else
>> it'll warn (or not warn) about. Eventually, users of such systems tend
>> point these things out, but that could take a while.
>
> I compiled 9.0 yesterday and there are a lot of warnings with my current
> compiler; something like -Werror is right out, I think. Maybe we can
> save warnings separately and have the member turn yellow if there are
> any, or something like that?

Hmm. For this particular case, it would straightforward to add a step to
the buildfarm script to do "grep -r '%z' src". It might turn up false
positives, if there's a %z in comments or such, but it shouldn't be much
effort to maintain a list of exceptions to filter out the false positives.

- Heikki