On 09/04/2013 09:12 PM, Simo Sorce wrote:
On Wed, 2013-09-04 at 17:06 +0200, Sumit Bose wrote:
> On Tue, Sep 03, 2013 at 02:07:29PM -0400, Simo Sorce wrote:
>> On Tue, 2013-09-03 at 15:25 +0200, Pavel Březina wrote:
>>> On 09/03/2013 01:27 PM, Sumit Bose wrote:
>>>> Hi,
>>>>
>>>> while looking at expand_ccname_template() becasue of shadowing rewind()
>>>> I realized that there a some issues with some of the new krb5.conf
>>>> templates. This patch fixes them and adds some tests to avoid similar
>>>> issues in the future.
>>>>
>>>> There is one change in behaviour. If the name in the %{} braces does not
>>>> match any of the known krb5.conf templates for UNIX the new code returns
>>>> an error while the old just returned something, which in most case will
>>>> not be the original input. Please tell me if you prefer the original
>>>> input in this case so that I can change the patch accordingly.
>>>
>>> I think we should expand the parameters we can, i.e. uid and euid and
>>> leave everything else intact (i.e. don't even check ${LIBDIR} etc.). Ad
>>> absurdum: for what we know, users can ran custom build that supports
>>> other variables...
>>
>> True, I was looking at this code yesterday again as I am hadling 2071
>> and I was thinking the same, we should just ignore and leave unchanged
>> %{xyz} patterns we do not understand, so that if libkrb5 adds something
>> the admin can use we do not have to explicitly support it.
>>
>>> But if you don't agree, I'll ack this patch.
>>
>> No, let's nack, and if you all do not mind I can come up with an
>> alternative patch that does what I describe above.
>
> Whatever you prefer, but I can do the needed changes as well. If you do
> them, please do not forget to add tests for the cases which are
> currently not covered.
I opened bug #2076 to track this issue and attached find a different
approach with testes against current master.
Simo.
Nack.
krb5_utils-tests.c: In function ‘test_krb5_style_expansion’:
krb5_utils-tests.c:679:9: error: unused variable ‘ret’
It looks good otherwise.
Can we also include Sumit's tests?