Re: GDL2: prototypes and EOAttribute.m

[Matt Rice]

Originally START_SET/END_SET merely did some exception handling,

#define START_SET(supported) if ((supported)) { NS_DURING
#define END_SET(desc , args...) NS_HANDLER \
           fprintf(stderr, "EXCEPTION: %s %s %s\n", \
                   [[localException name] cString], \
                   [[localException reason] cString], \
                   [[[localException userInfo] description] cString]); \
           unresolved (desc , ## args); \
           NS_ENDHANDLER } \
           else unsupported (desc , ## args)


Current versions of these macros declare local variables:
#define START_SET(setName) \ { \
...
NS_DURING \
 NSAutoreleasePool *_setPool = [NSAutoreleasePool new]; \
...

#define END_SET(setName) \
} \
[_setPool release]; \

This is at least my guess as to which variable is the deallocated instance,
If we want to regain the old behavior, this should probably add the
'_setPool', to a _setPoolStack or some such (declared outside of the
macro).

However, removing the nested sets seems acceptable, it just means that
abandoning a set will abandon more tests than before, which seems like
an improvement over the status quo/deallocated instance errors, so i'm
not really going to oppose it, if you want to go this route, or anyone
pushes back against nested sets.


On Wed, Jun 7, 2017 at 2:28 PM, Mark Clements <mark.clements@ki.se> wrote:
> Next up: the gdl2 testsuite was failing for some files with nested
> TEST_SET's, with errors for "message sent to deallocated instance". Is
> it obvious why the nesting does this?
>
> If we remove the nested sets, then we resolve the "deallocated instance"
> errors. Proposed patch attached.
>
> -- Mark
>
> On 06/06/2017 05:34 PM, Mark Clements wrote:
>> See attached for an improved patch on quoting. This gives similar tests
>> successes for both quoted and unquoted external names.
>>
>> -- Mark
>>
>> On 06/05/2017 10:00 PM, Mark Clements wrote:
>>> Please see attached for a patch to fix quoting of column names and
>>> Postgresql sequences. I have also included a few formatting changes.
>>>
>>> Kindly, Mark.
>>>
>>>
>