WG-Polonius meeting minutes: 2019/05/14

Relevant links:

A recap of last week’s work:

Niko and Albin went through a lot of details of liveness, in the dedicated Zulip thread (around this point).
With the previous week’s steps, and these conversations, Albin has continued the work on liveness, and is now basically done :tada:. With the 2 PRs, Polonius is able to compute the same region_live_at facts as rustc. The remaining work being mostly clean-up, various minor fixes, more testing, and finally, reviews. The remaining tasks are listed in this comment.
Rémy has continued looking at the rustc test suite under Polonius. As expected, over #60171, there are less than 20 failures. A first pass at analyzing the results is available here with links to the necessary information (test source, NLL/polonius outputs, diff) to categorize the failures but the TL;DR is: some of them are trivially fixable (being artifacts of test construction, and identical under NLLs and Polonius), some are duplicates or similar to of other failures, a lot are simply different diagnostics, and a handful need more in-depth investigation to understand the difference in behaviour, if it’s a bug, and so on. More information will generally be available in the dedicated thread.
@lokalmatador is working through profiling tasks and may soon be able to gather benchmark data.
On the more administrative side of things, Rémy wrote up the meeting notes from the previous weeks, updated the roadmap to add testing and validation, as we mentioned during last week’s meeting.

Useful information:

various members will be absent at different points in the upcoming weeks. There may not necessarily be a meeting sometimes (TBD), but async updates will be available on Zulip in any case.

Useful future tasks we mentioned:

with the liveness work about to land, we mentioned some important optimizations rustc has about liveness (in particular, trying to avoid computing it unless it is necessary), and that Polonius (or its inputs) will surely need as well. This topic was discussed at this point in the meeting thread.

For the following week:

Albin will tackle the remaining final liveness.
Rémy will continue on the rustc test suite results (fix the trivial differences, etc) and time permitting, help with the final liveness tasks.
@lokalmatador will continue their benchmarking endeavour.