On (09/02/16 11:54), Pavel Reichl wrote:
On 02/09/2016 08:58 AM, Lukas Slebodnik wrote:
>On (08/02/16 09:24), Pavel Reichl wrote:
>>On 02/05/2016 10:25 PM, Lukas Slebodnik wrote:
>>>On (05/02/16 19:34), Michal Židek wrote:
>>>>On 02/05/2016 06:32 PM, Lukas Slebodnik wrote:
>>>>>On (05/02/16 17:30), Michal Židek wrote:
>>>>>>On 02/05/2016 05:13 PM, Lukas Slebodnik wrote:
>>>>>>>On (05/02/16 16:56), Pavel Reichl wrote:
>>>>>>>>Hopefully the last one.
>>>>>>>
>>>>>>>>From 58b06b8fee18c4242e301ecf71f92a77f6ba6e7b Mon Sep 17
00:00:00 2001
>>>>>>>>From: Pavel Reichl <preichl(a)redhat.com>
>>>>>>>>Date: Fri, 22 Jan 2016 08:34:14 -0500
>>>>>>>>Subject: [PATCH] IDMAP: Add test to validate off by one
bug
>>>>>>>>
>>>>>>>>Resolves:
>>>>>>>>https://fedorahosted.org/sssd/ticket/2922
>>>>>>>>---
>>>>>>>>src/tests/cmocka/test_sss_idmap.c | 125
++++++++++++++++++++++++++++++++++++++
>>>>>>>>1 file changed, 125 insertions(+)
>>>>>>>>
>>>>>>>>diff --git a/src/tests/cmocka/test_sss_idmap.c
b/src/tests/cmocka/test_sss_idmap.c
>>>>>>>>index
00e03ffd9ab1532fb55795b9935b254c8a89ec16..2c99a09e6d11e8d731860eaad94eda0900fafabe 100644
>>>>>>>>--- a/src/tests/cmocka/test_sss_idmap.c
>>>>>>>>+++ b/src/tests/cmocka/test_sss_idmap.c
>>>>>>>>@@ -43,6 +43,17 @@
>>>>>>>>#define TEST_OFFSET 1000000
>>>>>>>>#define TEST_OFFSET_STR "1000000"
>>>>>>>>
>>>>>>>>+#define TEST_2922_MIN_ID 1842600000
>>>>>>>>+#define TEST_2922_MAX_ID 1842799999
>>>>>>>>+#define TEST_2922_MAX_ID_PLUS_ONE (TEST_2922_MAX_ID + 1)
>>>>>>>>+
>>>>>>>>+#define TEST_2922_DFL_SLIDE 9212
>>>>>>>>+#define TEST_2922_FIRST_SID TEST_DOM_SID"-0"
>>>>>>>>+/* Last SID = first SID + (default) rangesize -1 */
>>>>>>>>+#define TEST_2922_LAST_SID
TEST_DOM_SID"-199999"
>>>>>>>>+/* Last SID = first SID + rangesize */
>>>>>>>>+#define TEST_2922_LAST_SID_PLUS_ONE
TEST_DOM_SID"-200000"
>>>>>>>>+
>>>>>>>We should prefer to use constants instead of #defines.
>>>>>>>Constants are more type safe.
>>>>>>>
>>>>>>
>>>>>>Do you want Pavel to change it? If so, then I would prefer
>>>>>>at least the strings to remain as #defines. It is more
>>>>>>comfortable to concatenate string literals. But honestly
>>>>>>I am ok with the numbers to remain as #defines as well,
>>>>>>but I agree with your point in general.
>>>>>>
>>>>>Yes, please.
>>>>>
>>>>>Define can still be used for simplification of string initialisation
>>>>>e.g.
>>>>>#define DOM_SID "S-1-2-3-4"
>>>>>const char TEST_2922_LAST_SID[] = DOM_SID"-99";
>>>>>const char TEST_2922_LAST_SID_PLUS_ONE[] = DOM_SID"-100";
>>>>>
>>>>
>>>>Yes, they can. But do you require it? As I said I would
>>>>prefer the strings to remain #defines, but it is not strong
>>>>preference.
>>>>
>>>I'm sorry but your preference is not type safe.
>>>It's much simpler to identify errors caused by wrong
>>
>>Well, it's just your preference. It's not in out code style, right? Anyway
it's often used in other cmocka tests so it's consistent with out code base.
>>
>It's not just my preference constants are already used in test
>(module scope)
>
>sh$ cd src/tests
>sh$ git grep "^const" | wc -l
>64
grep '#define' `find src/tests -name '*.c'` | grep \" | wc -l
237
This is not technical reason for not using type safe constants.
In past we used debug macro with numbers(old-style) and hexadecimal
version (SSSDBG_*).
Even though the old-style version was used much more times we were
adding new debug macros with new-style.
And this is the similar case.
>
>Could you give me any reason why macros are better from technical
>point of view?
It doesn't matter. Both work and we are wasting time on fruitless discussion.
You can always improve code. It's import to know when it's good enough. I say
using defines in test is good enough and I don't see real benefit in changing
it that would justify time spend on fixing.
I'm sorry I cannot see a technical reason in this paragraph why
"#defines" are
better then constants. BTW the purpose of pre commit review is to improve code.
So I cannot see a reason why we should do it later.
>We needn't support sssd on platforms with old C compiler
>and we use c99(gnu99) anyway.
>
>If no then I will appreciate usage of new-style constants.
>
>>>usage of constants. If you like macros then you can be honored
>>>in future to try find bugs in macro generated code (nss-pam-ldapd).
>>I think this is totally unrelated, please try to focus on purpose of this thread -
test for id mapping.
>>
>It just an explanation of my reasoning.
>
>>>
>>>>>>>
>>>>>>>
>>>>>>>>struct test_ctx {
>>>>>>>> TALLOC_CTX *mem_idmap;
>>>>>>>> struct sss_idmap_ctx *idmap_ctx;
>>>>>>>>@@ -128,6 +139,53 @@ static int setup_ranges(struct
test_ctx *test_ctx, bool external_mapping,
>>>>>>>> return 0;
>>>>>>>>}
>>>>>>>>
>>>>>>>>+static int setup_ranges_2922(struct test_ctx *test_ctx)
>>>>>>>>+{
>>>>>>>>+ struct sss_idmap_range range;
>>>>>>>>+ enum idmap_error_code err;
>>>>>>>>+ const char *name;
>>>>>>>>+ const char *sid;
>>>>>>>>+ /* Pick a new slice. */
>>>>>>>>+ id_t slice_num = -1;
>>>>>>>>+
>>>>>>>>+ if (test_ctx == NULL) {
>>>>>>>>+ DEBUG(SSSDBG_FATAL_FAILURE,
>>>>>>>>+ "test context is NULL\n");
>>>>>>>>+ return 1;
>>>>>>>>+ }
>>>>>>>>+
>>>>>>>>+ name = TEST_DOM_NAME;
>>>>>>>>+ sid = TEST_DOM_SID;
>>>>>>>>+
>>>>>>>>+ err =
sss_idmap_calculate_range(test_ctx->idmap_ctx, sid, &slice_num,
>>>>>>>>+ &range);
>>>>>>>>+ if (err != IDMAP_SUCCESS) {
>>>>>>>>+ DEBUG(SSSDBG_FATAL_FAILURE,
>>>>>>>>+ "sss_idmap_calculate_range failed:
exp: [%d] but got [%d]\n",
>>>>>>>>+ IDMAP_SUCCESS, err);
>>>>>>>>+ return 1;
>>>>>>>Please use different numbers for different failures.
>>>>>>
>>>>>>I think it is OK to use the same number in tests, as long as
>>>>>>the debug message is present. We can identify the error
>>>>>>based on stderr messages.
>>>>>>
>>>>>No it's not OK to have same return code.
>>>>>The return codes have to be unique.
>>>>
>>>>They do not need to be unique here. Compare:
>>>>
>>>>[ RUN ] test_map_id_2922
>>>>(Fri Feb 5 19:00:17:723118 2016) [sssd]
>>>>[test_sss_idmap_setup_with_domains_2922] (0x0010): test contex is NULL
>>>>Could not run the test - check test fixtures
>>>>[ ERROR ] test_map_id_2922
>>>>
>>>>
>>>>With this:
>>>>
>>>>[ RUN ] test_map_id_2922
>>>>(Fri Feb 5 19:01:17:376602 2016) [sssd]
>>>>[test_sss_idmap_setup_with_domains_2922] (0x0010): test contex is NULL
>>>>Could not run the test - check test fixtures
>>>>[ ERROR ] test_map_id_2922
>>>>
>>>>I returned different error code in both cases but the result
>>>>is the same.
>>>>
>>>That's because of insufficient desing of libcmocka.
>>>Unit test should not contain any if conditions
>>>which are not tested. If you run code coverage of unit test
>>>it should cover 100% of code path of unit test.
>>>
>>>However, the "official way" of handling errors in setup
>>>and teardown function does not allow to do this.
>>>In other words, returning error code is useless.
>>>As you pointed out there's no difference and we need to
>>>use debug messages to find out where it failed.
>>>that's the worst situation. The error message
>>>and line where it happened should be logged automaticaly
>>>or thearduwn function should return unique identifier
>>>of failure.
>>>
>>>The usage of debug message (printf error) is kind of workaround
>>>for current state. Workarounds should be removed after fixing the staff.
>>>So we will need to retun a unique error code anyway in the end
>>>and there's no reason to do it twice.
>>>
>>
>>We are not working on cmocka code here and we cannot predict its evolution.
>>
>On the other hand, this patch introduced the new style of error
>handling in fixtures (setup and teardown functions).
>This code will be seen as a precedence in future
>because we do not have a coding guidelines for this.
>So it would be better to do it properly.
Well, that's the problem - you think your way is proper and I and Michal don't
share it.
I'm sorry I did a mistake in previous paragraph.
s/precedence/precedent/
We do not have a coding guidelines for handling errors in fixtures.
This introduced new way of handling errors will be in future seen
as a refferential implementation. And there should be common agreement
how to do it otherwise different develepers will do it in different way
and in the end it will be mess.
This is a main *BLOCKER* for me.
I will repeat here a compromise from previous mail which I will accept.
If you think that usage of new-style error handling in fixtures should
not block this patch then please use old-style. In another words,
Please use old-style of everything or new-style of everything.
It's up to you which way you will choose.
LS