FYI @wg-incr-comp: I have a series of PRs changing the way the dep-graph is persisted on disk: #82780, #83214 and #83322. @mw has been reviewing the implementation. We would feel more comfortable merging early in the release cycle, and with a team sign-off. Please take a look and tell me if a MCP is required.
Yup, #82780 is probably closest to being finished. It does not fundamentally change things - it just makes the compiler write the dep-graph to disk as it goes instead of all at once at the end. So I don't think an MCP is required (as long as the working group is aware of the changes).
#83214 is also something that makes a lot of sense. I haven't had time to review it in detail. I don't think this one requires an MCP.
#83322 I have not really looked at much. I think the general direction definitely makes sense but we'll need to gather some data on how it affects file sizes. In general, getting rid of the LEB128 encoding (at least for incr. comp. caches) is a good idea and opens things up for further optimizations.
Lots of great stuff from @cjgillot lately! :smile:
Heads up! I plan to r+ #82780 (i.e. streaming the depgraph) once it is rebased. @Wesley Wiser, you are currently assigned, so please let me know if you have any objections or want to take a closer look yourself.
I haven't had time to look closely at it but the description and perf looks good to me. Thanks for reviewing @mw!
Thanks, @Wesley Wiser! I just r+ed it.
Last updated: Oct 21 2021 at 20:03 UTC