[dmd-internals] [OT] Re: DMD build still borked

[Brad Roberts]

On Thu, 27 Sep 2012, Jason House wrote:

> It's been a while since I looked at the auto tester web page. I really 
> like the new layout!
> 
> The clean front page made me poke around a bit more. I'm s bit confused 
> by the pull tester page. When I looked just now, there was a "has 
> passes" section and a "has failures" section. Many pull requests seem 
> like they could go into either section.

The intent is that everything has just passes (ie no failures yet) 
(unknowns are included here as well) first, then everything that has known 
failures, then everything that's completely unknown.  Unknown means never 
has has a build, or the previous builds are out of date (due to either a 
new master branch checkin or due to the sha of the pull request being 
updated).

There's a section that should probably be first above the 'has passes' 
block which is 'completely passes' or some label like that, where all of 
the platforms have completed a current run successfully.  It'd be easy to 
add, but I'm not clear on the value.
 
> How do people use that page? If it's to find something to review and 
> commit, I'd expect all green pulls to be emphasized more. If it should 
> help authors to review/fix their pull requests, I would have expected 
> things to be sorted by author (or better yet, a distinct page per 
> author)

When I have the time and desire to look for something to pull, I 
definitely look at the 'has passes' block, and focus on those that have 
completed a full set of runs.

I'm not overly fond of the way this page works.  It lacks a good view of 
history.  I worry a little about the case of a pull request that 
introduces a periodic / random failure.  Easy to spot if you look across 
10 runs, but potentially hard to spot looking only at the most recent run.  
It's not terribly likely, but it's not impossible either.  A good bit of 
the problem is the magnitude of the number of platforms and even more so, 
the number of open pull requests.  It's just an enormous amount of data to 
present.  I'm very much open to suggestions both on layout and run 
ordering. There's some issues I need to address in how work is assigned to 
the build hosts to make things as flexible as I want, but even the current 
implementation has a good deal of flexibility.

Thanks for your feedback, I really appreciate it. :)

Later,
Brad