Regression analysis for 1.2 beta and 1.3 nightly


#1

Today @nikomatsakis and I categorized all the reported recent regressions for the current beta (1.2) and nightly (1.3).

I’ve put the results in a spreadsheet that tries to categorize the reports in various ways.

Of the recent regressions most of them have been satisfactorily dealt with. I only have two remaining concerns:

  1. Both the simple_stats and timely regressions are attributed to the Extend<&T> impl, but the regressions happen in different branches. This shouldn’t be possible - either the regression happened in 1.2 or 1.3. There must be more to this.
  2. The ncollide_queries regression is optimistically classified as ‘mystery bugfix’. The breakage looks arguably right to @nikomatsakis but he’s not sure why it happened.

I’ll try to keep this spreadsheet updated for future releases.


#2

Bluss indicated that the Extend<&T> change landed before the regression for timely seems to have happened (in the gap between nightlies at 2015-07-01 and 2015-07-07). So, perhaps it is a different issue than Extend<&T> after all…


#3

I believe I have found the problem and it is a genuine bug. I am testing a fix now. I am going to check whether it accounts for the @frankmcsherry regression as well.


#4

PR posted: https://github.com/rust-lang/rust/pull/27258


#5

Thanks @nikomatsakis. I’ve updated the spreadsheet. Does your patch by chance fix the protobuf regression too?