I was thinking to submit a round table focusing on resolving FIXMEs in the codebase of clang and llvm. The idea is FIXME hunters to bring a laptop and start hunting down the 13K FIXMEs in both codebases.
Ideally we would invite code owners to provide key insights and to possibly review patches quickly.
I am wondering if that sounds fun and are there people willing to engage?
I like the idea, and would be happy to help in an advisory capacity - answering folks questions about whether a given FIXME is easy/hard (as far as I know) and where to go to maybe fix it. But doubt I have the motivation to be driving/working on fixes myself, unfortunately.
This topic is very relevant indeed. I made a quick check in the clang sources and this is the FIXME count for different releases:
Release #FIXME Increase
14 6101 12
13 6089 57
12 6032 242
11 5790 134
10 5656 320
9 5336 286
8 5050 53
7 4997 200
6 4797 157
5 4640 214
In summary, the number of FIXMEs keeps growing… While I think we should definitely address this I feel it might not be easy within 1h slot. This could be an excellent task for a Hacker Lab for example. I am not sure though whether it is planned for this year?
Other formats we can consider are:
Brainstorming our FIXME policy and what can be done to avoid/reduce the growth. Should we start tagging FIXMEs by a domain or severity level and etc to help find the most relevant ones?
Build some tooling e.g. finding the oldest FIXMEs or checking which domains have more FIXMEs. I would particularly be interested in a way to classify them… It’s been a while I am trying to find a way to filter OpenCL-specific FIXMEs. I think finding relevant FIXME could be a good first step forward.
Overall I would be happy to join this session in whichever format as soon as it is in reasonable enough timing for me (AM hours would work best!).
The proposal looks interesting. But I have a basic question to ask.
What does a round table mean? Does it require us to attend a meeting at real time.
Or does it mean a project that people could interact with each other more quickly?
I imagined that people interested will show up, at real time, and work on some pre-picked items/areas. I think the benefit from the real time participation is we could ask code owners for reviews/help.
I agree that a Hackers Lab might have been a better format but I do not think there is such a thing this year. I am open to any suggestions how to a better format or make this more efficient.
I got it. I agree that it is more effective to interact for a concret small topic. But there might be people couldn’t attend due to many reasons.
So I guess it may be better to record the
TODO left in the meeting a online document so that other people could work asynchronously.