I’m not opposed to it, but I think our team would suggest that it’s a good use of a multi-repo solution where each repo can have its own enforced style.
What would your thoughts be on that?
- Bret
Bret,
Details aside this ECC issue got me thinking it might be useful to some how tag code that follows different, subsetted, or alternate rules, or uses different build environments. I was kind of thinking of doing something like how we tag
the licenses in the header comments [1]. I would say nothing means edk2 rules. I can see vendors maybe having different internal rules, or us wanting to distinguish test code from production code. The general idea if we start this tools can be smarter and
that seems like a good thing.
This is just a wild idea, so I’d like to see what other people think?
[1] SPDX-License-Identifier: BSD-2-Clause-Patent
That’s actually what got me here. EccCheck runs as part of our CI now (but it didn’t when I wrote these tests a year ago). I need to either figure out how to get this code to pass EccCheck in a reasonable way, or just not contribute the
tests and say “go to Mu for the tests, if you want them”.
Skipping the contribution isn’t a desirable outcome at all for a number of reasons, not the least of which is that we (MS and others) are trying encourage all contributions to come with tests, so the process of writing them needs to be
simple and painless.
I’ve had this issue with EFI code too. It will compile with for DEBUG and RELEASE as the optimizer removes the memcpy/memset. So you only see a build failure when you compiler NOOPT (and there are no intrinsic libs). I mostly see this in
platform code when I try to compile a single driver/lib NOOPT and it fails to link due to the missing intrinsic.
The easy to enforce this is to compile with optimizations enabled and don’t enable intrinsic libs. Not sure if that is really practical from the test point of view.
Seems the tool caught the coding style violation so I guess we could try to “make running that tool easier”. Maybe hooking into patchcheck.py, making some kind of githook, or adding a git command?
So for context, this is a new host-based test that should only run within a platform OS, so intrinsics aren’t the big deal that they would be in FW code.
But we do need to figure out how to simultaneously adhere to the coding convention while enabling test authoring.
Or we chose to not enforce quite as many things for tests.
If the structure is a non-static local variable, most compilers will silently inject an intrinsic call to memcpy in function initialization. This leads to an intermittent linker error.
If the compiler you use automatically supports an intrinsic memcpy in the given architecture or optimizes out the memcpy, it will build for you and you won’t know you need to link to an intrinsic support library
in order to build cross platform. This leads to code that builds for you, but not for me.
ERROR - EFI coding style error
ERROR - *Error code: 5007
ERROR - *There should be no initialization of a variable as part of its declaration
ERROR - *file: //home/corthon/_uefi/edk2_qemu_ci/edk2/MdeModulePkg/Library/VariablePolicyLib/VariablePolicyUnitTest/VariablePolicyUnitTest.c
ERROR - *Line number: 333
ERROR - *Variable Name: MatchCheckPolicy
SIMPLE_VARIABLE_POLICY_ENTRY ValidationPolicy = {
VARIABLE_POLICY_ENTRY_REVISION,
sizeof(VARIABLE_POLICY_ENTRY) + sizeof(TEST_VAR_1_NAME),
sizeof(VARIABLE_POLICY_ENTRY),
TEST_POLICY_MIN_SIZE_NULL,
TEST_POLICY_MAX_SIZE_NULL,
TEST_POLICY_ATTRIBUTES_NULL,
TEST_POLICY_ATTRIBUTES_NULL,
VARIABLE_POLICY_TYPE_NO_LOCK
But you can’t init this structure separately without addressing each field.
Can a brother get an override?
<95B370A7BEED49AAB797022E8F260C8F.png>