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 Sent: Wednesday, October 7, 2020 9:45 AM To: Laszlo Ersek ; Andrew Fish ; devel@edk2.groups.io Cc: Kinney, Michael D 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 > 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 > 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) >>> >>> 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) >>> >>> 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 >>> >>> >>> >>> >>> >>> >>> >> >> >> >> >> >> >