Bret,
Can you provide the compiler log?
You can still use STATIC CONST and copy it to a 2nd local that allows mods.
Mike
From: Bret Barkelew <Bret.Barkelew@microsoft.com>
Sent: Wednesday, October 7, 2020 9:45 AM
To: Laszlo Ersek <lersek@redhat.com>; Andrew Fish <afish@apple.com>; devel@edk2.groups.io
Cc: Kinney, Michael D <michael.d.kinney@intel.com>
Subject: RE: [EXTERNAL] Re: [edk2-devel] VariablePolicy: Final Changes Thread 2 - ECC & UnitTest
I can test it again, but I hit a compiler complaint around the STATIC CONST solution.
Also, it doesn’t apply in all cases because I have at lease half a dozen cases that say “test X, prove negative, twiddle value, test Y, prove positive”.
I’ll try a few more things, but I may reissue the patch series and withhold the tests until they can be rewritten to match. I don’t want them to hold up VarPol any longer.
- Bret
From: Laszlo Ersek
Sent: Wednesday, October 7, 2020 8:51 AM
To: Andrew Fish;
devel@edk2.groups.io
Cc: Kinney, Michael D;
Bret Barkelew
Subject: [EXTERNAL] Re: [edk2-devel] VariablePolicy: Final Changes Thread 2 - ECC & UnitTest
On 10/07/20 16:27, Andrew Fish wrote:
> For case 1 I thought the size had to be > 8 bytes, not just a struct? Maybe that is compiler specific?
Honestly, I've got no clue. I just remember we must avoid initializers
for objects that do not have static storage duration.
Laszlo
>
> Sent from my iPhone
>
>> On Oct 7, 2020, at 6:43 AM, Laszlo Ersek <lersek@redhat.com> wrote:
>>
>> On 10/07/20 03:46, Michael D Kinney wrote:
>>>
>>> Bret,
>>>
>>> Initializing variable in declaration for structures and arrays
>>> introduces use of intrinsics. Since it is possible for unit test
>>> sources to be used for both host and target tests, I recommend we
>>> continue to follow the EDK II coding style for unit tests to support
>>> maximum compatibility and code reuse.
>>>
>>> Using a module global variable with initializers instead of
>>> initializing a local declaration is the same amount of work, so I do
>>> not believe that will result in fewer tests.
>>>
>>> I agree it is useful to have the test data next to the test code. This
>>> can be accomplished by breaking up into more files so the test data is
>>> immediately above the test function the test data is used. Does ECC
>>> raise an error if a module global is placed between 2 functions? A
>>> 2nd approach to put the module global immediately above the test
>>> function the test data is used.
>>
>> Consider the following example structure type, for the sake of
>> discussion:
>>
>> typedef struct {
>> UINT32 Value;
>> } TEST_DATA;
>>
>>
>> * Case#1: block scope, automatic storage duration
>>
>> EFI_STATUS
>> FoobarTest (
>> VOID
>> )
>> {
>> TEST_DATA TestData = { 42 };
>> // ...
>> }
>>
>> Problem: uses intrinsics.
>>
>>
>> * Case#2: file scope, static storage duration.
>>
>> STATIC CONST TEST_DATA mTestData = { 42 };
>>
>> EFI_STATUS
>> FoobarTest (
>> VOID
>> )
>> {
>> // ...
>> }
>>
>> Problem: either "mTestData" is textually far from FoobarTest(), or -- if
>> we keep them close to each other -- we mix variable definitions with
>> function definitions, at file scope.
>>
>>
>> * Case #3: block scope, static storage duration.
>>
>> EFI_STATUS
>> FoobarTest (
>> VOID
>> )
>> {
>> STATIC CONST TEST_DATA TestData = { 42 };
>> // ...
>> }
>>
>> Problem: there should be none. Does not involve intrinsics, and the
>> object definition is part of the function's scope.
>>
>>
>> If ECC does not recognize case#3 as valid, then that is an *ECC bug*.
>>
>> ECC has no reason to prevent case#3, as case#3 does not involve
>> intrinsics, and is a generally valid and useful C language construct (it
>> combines the life cycle of case#2 with the visibility of case#1).
>>
>> Again, if ECC rejects case#3, that's *definitely* a bug in ECC, and we
>> should fix it first. Given that ECC includes a full-blown C language
>> parser, the fix should not be too difficult -- check if the declaration
>> has the "static" storage-class specifier.
>>
>> ... In fact, I think that purely CONST-qualifying TestData might suffice
>> for shutting up ECC. See the following in
>> "BaseTools/Source/Python/Ecc/c.py", method
>> "CheckFuncLayoutLocalVariable":
>>
>>> for Result in ResultSet:
>>> if len(Result[1]) > 0 and 'CONST' not in Result[3]:
>>> PrintErrorMsg(ERROR_C_FUNCTION_LAYOUT_CHECK_NO_INIT_OF_VARIABLE, 'Variable Name: %s' % Result[0], FileTable, Result[2])
>>
>> So case#3 should work through that avenue already, because case#3 has
>> CONST *too*.
>>
>> Now, in case#3, if "TestData" needs to undergo modifications, and so
>> CONST is not immediately desirable, that's solvable:
>>
>> EFI_STATUS
>> FoobarTest (
>> VOID
>> )
>> {
>> STATIC CONST TEST_DATA TestDataTemplate = { 42 };
>> TEST_DATA TestData;
>>
>> CopyMem (&TestData, TestDataTemplate, sizeof (TEST_DATA));
>> // ...
>> }
>>
>> Thanks
>> Laszlo
>>
>>>
>>> Best regards,
>>>
>>> Mike
>>>
>>> From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of Bret Barkelew via groups.io
>>> Sent: Tuesday, October 6, 2020 5:28 PM
>>> To: devel@edk2.groups.io
>>> Subject: [edk2-devel] VariablePolicy: Final Changes Thread 2 - ECC & UnitTest
>>>
>>> Ive worked through all the ECC issues with Variable Policy (AND the UnitTests) on this branch:
>>> Commits · corthon/edk2 (github.com)<https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fcorthon%2Fedk2%2Fcommits%2Fvar_policy_dev_submission_v8&data=04%7C01%7Cbret.barkelew%40microsoft.com%7C8286accbe81740bde8cc08d86ad8c8ae%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637376826604910138%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=XSNAxTGdbKGlrPLV%2BqWVNEAQjyyeSoKQ8zcfG1%2B4W1s%3D&reserved=0>
>>>
>>> I even wrote the Main() entry point lib that Laszlo suggested (it works rather nicely):
>>> TEMP: Staging for HostTest entry point · corthon/edk2@4ce5210 (github.com)<https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fcorthon%2Fedk2%2Fcommit%2F4ce52108b3e1bcb2ba78995be94c3949fe647eda&data=04%7C01%7Cbret.barkelew%40microsoft.com%7C8286accbe81740bde8cc08d86ad8c8ae%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C637376826604910138%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=46MZeja1WTnq44vSZOwJQ%2BAs61TbdFQtFfyj7wgm5mY%3D&reserved=0>
>>>
>>> However, theres one that I just cant get past and I would like to take it up with the community. I dont think that UnitTests should have to deal with the cant initialize variables in declaration check. Almost none of the solutions that I tested worked,
and the ones that did were too cumbersome. They failed on two key points that are important for test writing:
>>>
>>> * They were annoying to write ===> fewer tests.
>>> * They moved even more of the test case data away from the test ===> harder to read tests.
>>>
>>> I would like to move for an exception for unit tests (or at least host-based unit tests), but I dont know how to accomplish that from a technical standpoint.
>>>
>>> Thoughts?
>>>
>>> - Bret
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>>
>>
>>
>