Re: [boost] PSA: Travis OS X bottleneck, <cxxstd> new Boost.Build feature

Subject: Re: [boost] PSA: Travis OS X bottleneck, <cxxstd> new Boost.Build feature
From: Robert Ramey (ramey_at_[hidden])
Date: 2017-12-02 16:59:29

On 11/29/17 1:10 AM, Boris Kolpackov wrote:

 > In this light, I think it makes more sense to ask Boost users
 > who are using Boost on Mac OS to contribute hardware resources
 > for CI testing.
This is the right idea.  Let's go the full monte and take it to it's 
logical conclusion.

a) Encourage users to run tests on the libraries they use by making 
testing more user friendly.

b) Create a CMake or CMake-like dashboard where users can more or less 
automagically post their results to a common database.  Presumable boost 
regression.py does something similar already.

c) Create a program to permit display of this giant "data cube" 
according to the needs of the user/developer.

My personal contribution to moving in this direction can be found at

https://github.com/robertramey/library_status

This permits me to:

a) run the serialization test suite with boost build.  This is a single 
button which builds all the libraries that the serialization depends 
upon, and runs all the tests.

b) The results are displayed on a giant cumulative matrix which permits 
me to see which tests are failing on which platforms.  Here is a recent 
copy of that matrix

http://htmlpreview.github.com/?https://github.com/robertramey/library_status/blob/master/library_status.html

more information is available here:

http://htmlpreview.github.com/?https://github.com/robertramey/library_status/blob/master/index.html

If this idea were implemented we would have:

a) recent tests on all platforms actually being used regardless of how 
obscure.
b) an easily browsable data base of test results.
c) less dependency on centralized CI testing resources
d) less pressure on boost testers
e) more people participating and being a member of the boost community - 
helping to promote the "Boost Brand (c)"
f) Since Appveyor, Travis, regression.py, and individual users would be 
posting all results to a common database, we'd be decoupling test 
results presentation from any particular testing platform.
g) Should CMake solution every get accepted, it could easily be adjusted 
to post the the database as well.

Boost needs this going back at least a decade.

Robert Ramey