On 18/02/18 22:55, Ramsay Jones wrote:
>
>
> On 16/02/18 14:55, Adam Dinwoodie wrote:
>> On 12 February 2018 at 08:08, Olga Telezhnaya wrote:
>>> Add some tests for new formatting atoms from ref-filter.
>>> Some of new atoms are supported automatically,
>>> some of them are expanded into empty string
>>> (because they are useless for some types of objects),
>>> some of them could be supported later in other patches.
>>
>> This commit appears to be introducing failures in t1006 on Cygwin.
>> Specifically, tests 15, 36, 58 and 89, all titled "Check %(refname)
>> gives empty output", are failing on the pu branch at 21937aad4, and
>> git bisect identifies this commit, 3c1571744 ("cat-file: tests for new
>> atoms added", 2018-02-12), as the culprit.
>
> Hi Adam, Olga,
>
> Sparse has been complaining about this 'ot/cat-batch-format' branch for
> a while, so I had already noted (apart from a symbol which should be
> marked as static) something that looked somewhat off - namely, the on
> stack initialisation of a 'struct ref_array_item' (builtin/cat-file.c,
> line 246, in function batch_object_write()).
>
> In particular, the 'struct ref_array_item' ends with a [FLEX_ARRAY] field
> for the refname, so it clearly isn't meant to be declared on the stack.
> The 'ref-filter' code uses a 'constructor' function called 'new_ref_array_\
> item() which, among others, takes a 'refname' parameter with which to
> initialise the refname FLEX_ARRAY field.
>
> So, on Linux, if you build with SANITIZE=address and then run that test:
BTW, I forgot to mention that, if you run that test without -i, then
more than tests #15,36,58 and 89 will fail. In addition, tests #19-20,
#40-41, #62-63 and #93-94 failed - I did not investigate these
failures at all.
ATB,
Ramsay Jones
