Topics for Brno F2F
In yesterday's BAC monthly meeting, each community was required to have a topic list before the Brno F2F meeting in May. For the committers community, what do you guys like to put on the list?
For me personally, I find involving the Rust language into the project interests me a lot.
So, collected topics:
-
AI related topics
AI policy finalized - we experience a lot of PRs (and issues) generated by AI, what is the policy? (Frederik Wedel-Heinen) - together with a related topic raised by Dmitry: AI as a part of contribution and review process
the Mythos, how will these kind of models affect the security of open source software, say OpenSSL as an example (Paul Yang)
-
Product functionality & features
Rust. Providers implemented in Rust? Or we can refactor some core code of OpenSSL by using Rust? Richard was also playing with Zig for provider implementation (Paul Yang)
I'd like to see a discussion around the function of addrev, and especially where it takes its data from. (Richard)
Increasing OpenSSL plugability (Dmitry)
Privacy computing related features (protocol and algorithms), such as FHE, SMPC protocol, ZKP, Distance-preserving Encryption and so forth... (Paul Yang)
-
Clean code (Norbert & Shane)
What areas of the code would be beneficial to clean up/refactor?
What are the things that make the code unreadable and hard to understand which should go and be replaced? Hard to Read(/Debug) nested Macros vs expanded form e.g. https://github.com/openssl/openssl/pull/30926/changes#diff-7a39d0fcd854bb417feefebe7756553703ff6faf0f8e811055c1030db999ff88R460
-
Project management, development and review process
Pull request backlog is growing out of control and needs stronger management (Frederik Wedel-Heinen)
Need for an OpenSSL roadmap to align development across core team and contributor boundaries (Frederik Wedel-Heinen)
OpenSSL development process: PR review/approval process (Dmitry)
Dealing with the stream of CVEs (Dmitry)
-
Committers management
Do we need some committers tech chat/meetings/etc? (Dmitry)
How should the committers list be modified? (Dmitry)
Pull request backlog is growing out of control and needs stronger management
Need for an OpenSSL roadmap to align development across core team and contributor boundaries
AI policy finalized - we experience a lot of PRs (and issues) generated by AI, what is the policy?
Here are some more points. Please let me know if it needs elaboration.
@Frederik Wedel-Heinen I am interested in 2 and 3
@Frederik Wedel-Heinen In my committer's hat I'm interested in all these points
Rust is definitely interesting. I am concerned if we have enough knowledge to process PR’s. I think a strategy needs to be in place and possible risks should be addressed.
I would like to raise a new one: the Mythos, how will these kind of models affect the security of open source software, say OpenSSL as an example
Note on writing providers in Rust: if someone external already does, I'd like to see it!
I've played with writing a provider in Zig, and found a major difficulty... ERR upcall core_vset_error use a C va_list parameter , and I got into a type conflict trying to use the Zig construct for those. It would be interesting to see how that's dealt with in Rust.
@Richard Levitte (individual) a lot of unsafe {} is needed to deal with C interfaces, especially OpenSSL which is found of function pointers and pointer overloading, here you can see an example here: https://github.com/QUBIP/openssl-provider-forge-rs
@ssorce, thanks for the link! That gave me cause to add a section on support libraries in https://github.com/provider-corner?view_as=public
I'd like to see a discussion around the function of addrev, and especially where it takes its data from.
Inspired by git's gpg.ssh.allowedSignersFile, I think it would be possible to have a minimum bit of data necessary to perform what addrev does on file in the repository, and that file would be subject to review and approval like every other file.
This would make things much more transparent, and verifiable on per commit basis if anyone finds that important.
@Richard Levitte (individual) If you are collecting suggestions for that page then:
1) PKCS#11 Provider is now: https://github.com/openssl-projects/pkcs11-provider so both link and org needs to be updated
2) You may want to consider whether to add Kryoptic: https://github.com/latchset/kryoptic
This one is tricky, but when built with --feature=fips it creates a statically linked module that is both a PKCS#11 softoken and an OpenSSL FIPS provider (it currently embeds the OpenSSL 4.0 provider when built that way).
@ssorce, thanks for the update. Provider-corner being a provider corner, I think I'll stick to that, even though Kryoptic was interesting to look at.
But, for the future, it's preferable to raise an issue directly with that repo, so as not to saturate the thread here, which isn't quite related.
My suggestion for topic would be clean code.
What areas of the code would be beneficial to clean up/refactor?
What are the things that make the code unreadable and hard to understand which should go and be replaced?
Related to 2)
Hard to Read(/Debug) nested Macros vs expanded form e.g. https://github.com/openssl/openssl/pull/30926/changes#diff-7a39d0fcd854bb417feefebe7756553703ff6faf0f8e811055c1030db999ff88R460
Adding API Helper functions to hide OSSL_PARAM for common parameters (for KDF's MAC's etc) e.g. https://github.com/openssl/openssl/pull/30989
Nested Macros are the biggest problem preventing developers to understand the code...
Not necessarily nested, but code-generating (and especially function/structure-defining) macros make code significantly more difficult to navigate.
My particular pet peeve is overabundance of "methods" without use of designated initialisers for method table definitions, which make it nigh impossible to quickly find where all concrete implementations for a specific method are defined (as opposed to Linux, which also relies heavily on this concept, but doesn't have the similar problem due to consistent use of designated initialisers for almost everything).
ANSI C doesn't support designated initialisers and that's the standard OpenSSL was coding to until quite recently. Now that the project has embraced C99, they are usable. Adding them in is not a small task. Try raising a PR to see how large it is.
IMO: adding them opportunistically is an easier approach and is what should be done.
I agree 100% with all these remarks about code readability, do I remember correctly that in the past openssl was relying on a C standard so old that designated initializer could not be used? Would be good to have some historic/present perspective on what can be change and want can't going forward.
@ssorce The base C language syntax level was set to C89/C90 until recently, when it was elevated to C99.
We've veered from that base level in places, especially surrounding std functions and some included headers, with heavy guards around such sections.
My personal 2c
1. OpenSSL development process: PR review/approval process
2. AI as a part of contribution and review process
3. Increasing OpenSSL plugability
Dealing with the stream of CVEs
- AI as a part of contribution and review process
Incidently, I've started playing with https://pi.dev on a personal project...
Not on OpenSSL code yet, but I gotta say, it's tempting to start experimenting.
@Richard Levitte (individual)you can evaluate my PRs on symmetric keys in PKCS#12 with heavy Claude cooperation
@beldmit, I suppose I can do the same with any agent and a sufficiently good model. There are some relatively good prompts available for this sort of thing.
@Richard Levitte (individual) Sure
2 more topics
Do we need some committers tech chat/meetings/etc?
How should the committers list be modified?
Do we need to publish these topics to the general discussion? IIRC, this committers community is not accessible for non-committers
@Paul Yang I think it's still visible
@Dmitry Belyavsky , OK, then I will add a link in your thread in general discussion
Sounds like a good topic