New crater reports: 1.1 stable vs beta-2015-07-10 and nightly-2015-07-10


#1

Here’s stable vs. beta: https://gist.github.com/brson/e4d8184be1d9af0f901a

And here’s stable vs. nightly: https://gist.github.com/brson/9362098f0ca2ef38d1c2

Beta has 4 root regressions vs. stable and nightly has 24 root regressions vs. stable.

Here are the beta regressions:

And nightly:

Analysis - stable vs. beta

3 of the beta regressions are because #[packed] is no longer accepted. Looks like it was probably cause by this PR. It needs to be either fixed or mentioned in the relnotes as a breaking change.

The 4th beta regression though is a type inference failure. Somebody please investigate.

Analysis stable vs. nightly

  • ‘missing lifetime specifier’ - protobuf
  • ‘type annotations required’ - ncollide_queries
  • ‘missing documentation’ - nickel, startuppong, hueclient, coinbaser, dcpu16-gui, piston32-gfx_voxel, piston_window_game, dev_menu, poglgame, recaptcha, rust-cleverbot, rustfmt, rustful, shiplift, strava, tba, webdriver
  • ‘type of value must be known’ - timely, simple_stats
  • #[packed] - libnetfilter_queue, netfilter_queue
  • ‘illegal cast: cast through raw pointer first’ - rust_events

I don’t know what caused any of these.


Turning `#[packed]` into a permanent warning and backporting to 1.2
#2

The immediate question we need to answer is whether to do anything about the #[packed] regression on beta (this PR). We might restore the previous behavior of #[packed] and leave it deprecated.

It would also be delightful if somebody could figure out the other beta regression and start figuring out the nightly regressions.


#3

This at least is the improvements to the missing doc lint, and seems to suggest the importance of:

https://github.com/rust-lang/rfcs/pull/1193


#4

[ Since timely does not break in beta, it can’t be just Extend breaking it – timely’s issue is #26952 ]

These both look like they come from the addition of impl Extend<&T> for Vec<T> (and other collections. With both Extend<T> and Extend<&T> now, type inference is now sometimes ambiguous. Issue #26279


#5

This crate is already updated and I assume it is fixed. The author says “Type inference regression?”

Edit: Maybe that projections can’t drive inference anymore, could it be that? Though the time is wrong, that landed in nightly weeks ago.


Regression analysis for 1.2 beta and 1.3 nightly
#6

I found the cause of the problem. I’ll be opening a PR soon.