Open-CMSIS-PackOpen-CMSIS-Pack Technical Meeting 2022-05-31

Open-CMSIS-Pack Technical Meeting 2022-05-31


+Kyle Dando




Meeting Notes

Specification Change Control

BF: Held the review committee meeting. 6 issues were reviewed. Followed the process previously described. Feedback all welcome.

FR: What are the next steps? For #27 (#5 in the list approvals list) I can work on a PR

JK: Will assign you.

FR: How should the assignment work generally?

JK: Can have multiple assignees

JK: At the review stage the issues will be assigned to Bill->set up a review meeting


RK: Tensorflow has no concept of sematic versioning at the component level - not prepared to guarantee backward compatibility. AWS similarly has an umbrella pack which has months/years versioning. Do we allow packs that are not semantically versioned? Individual AWS packs do have semantic versioning.

DB: Semantic versioning is primarily to sort the packs.

RK: Believe there’s no specific requirement to enforce semantic versioning.

DJ: FatFs is using also different format: FatFs R0.14b

FR: SemVer gives more than this : we know API breaks, features introduction

RK: Think we need to write it down and potentially educate the industry.

ED: Worse because we can’t specify a range

RK: Have to assume they’re not compatible. Can’t mix and match between the versions. For the ML we’d have to replace the complete set of packs.

FR: The drawback is for the provider himself. Can’t force them.

DJ: I think it is not possible to force all authors to use semver.

See following opinion:

JK: Suggest we come back with some suggestions. Not comfortable with just having a string.

MD: Or we do not provide the features if the provider does not use semantic versioning.


HS: For single project can just provide name. If one csolution contains 2 projects which depend on each other. They may have different targets and compiler options. How to present this to the pdsc? Each project has one element in the pdsc. DonI can sp’t know how to present them. We have “hello world” for both M4 and M0 simultaneously.

KD: Both projects run at the same time. The two cores interact. It’s not either/or.

FR: Think you have only 1 csolution - have 2 x C projects. Have concept of phase to say if projects run concurrently. Your question is about the pdsc?

JK: Assume IDE loads csolution and displays 2 projects.

HS: Propose environment loaded as csolution file?


Same situation for TF-M where TF-M is board specific but the upper layer is board agnostic.


JK: Would like more opinions on this


HS: My concern is in the comment.

JK: So you would fix the build type for these projects?

HS: Yes


HS: We are supporting different toolchains/compilers.

RK: Is the problem compiler specific?

ED: By component?

JK: Needs further thinking. Do we aim to have a complete decoupling between csolution and cproject. I don’t think we have complete independence. What’s our expectation here?




TB: Have libraries sometimes compiled in soft FPU and sometimes in hard FPU mode. ‘Mixed’ not supported today. Cannot specify which ABI to use to get the linker to work correctly.

RK: Extensions to processor type? Also working on compiler agnostic extension feature?

TB: Yes

RK: Mapping in CMake only or deeper into csolution?

TB: Floating ABI is everywhere. Should be in the yaml. It goes through the full flow.

JK: Could be many options (MISRA …). We need to agree the most common set of options we want to set.

TB: Agree more in the enumeration values and propagate to the CMake.

Meeting Recording