On 19/02/17 01:10, Luc Van Oostenryck wrote:
> On Sat, Feb 18, 2017 at 10:37:36PM +0000, Ramsay Jones wrote:
>>
>>
>> On 18/02/17 20:30, Luc Van Oostenryck wrote:
>>> The existing test is an indirect test, using a warning
>>> about context imbalance to show that some part of code
>>> was discarded.
>>>
>>> Now that we have the minimal tools to test the output of
>>> test-linearize, use them to replace the test by a direct one.
>>
>> Hmm, it may be a more direct test, but it is not clear
>> just what is being tested (or indeed how it is being tested).
>>
>>
>> After applying this patch, I edited validation/c99-for-loop.c
>> like so:
>>
>> $ git diff
>> diff --git a/validation/c99-for-loop.c b/validation/c99-for-loop.c
>> index 427fde2..6b24fa8 100644
>> --- a/validation/c99-for-loop.c
>> +++ b/validation/c99-for-loop.c
>> @@ -2,8 +2,9 @@ int c99(void);
>> int c99(void)
>> {
>> int r = -1;
>> + int i;
>>
>> - for (int i = 0; i < 10; i++) {
>> + for (i = 0; i < 10; i++) {
>> r = i;
>> }
>>
>> $
>>
>> This modified test still passes (indeed the output is identical).
>> :-P
>>
>> ATB,
>> Ramsay Jones
>
> Which is wonderful because it's exactly what it should be.
So why is it called c99-for-loop.c? ;-) [rhetorical question]
> In fact I would love to be able to do this: show that two versions
> of a program/function give exactly the same output, but the testsuite
> can't do that (yet).
OK
> Now, you're right that It's may be not very clear what is being tested.
Indeed! (So, the commit message could use some improvement, right?)
> I may make more sense once you replace it in the context of the patch
> it replace and the associated fix:
> - 0e91f878 ("validation: Check C99 for loop variables")
> - ed73fd32 ("linearize: Emit C99 declarations correctly")
OK, so that sheds some light on the issue ...
> If you try the test case on the tree with the patch ed73fd32 reverted,
> you will see that the two versions doesn't give anymore the same result:
> The C89 version (your) version will still give the right output
> but the C99 version (the one of the test case) will essentialy gives
> an empty output (the code used to ignore the C99-style declaration
> and patch ed73fd32 fixed that).
Yes, I can confirm this behaviour. If I 'git revert ed73fd32' then
rebuild, sure enough the c99 version is 'essentially empty', but the
c89 version is as before.
So, how could I tell from the commit message alone, that this is what
the test is doing?
ATB,
Ramsay Jones
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
