Enzyme.jl icon indicating copy to clipboard operation
Enzyme.jl copied to clipboard

Update CIs to use `julia-actions/install-juliaup`

Open prbzrg opened this issue 1 year ago • 6 comments

Also replaced the actions/cache with julia-actions/cache.

prbzrg avatar Oct 19 '24 10:10 prbzrg

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 70.38%. Comparing base (037dfed) to head (1ef67b4). Report is 351 commits behind head on main.

Additional details and impacted files
@@            Coverage Diff             @@
##             main    #1987      +/-   ##
==========================================
+ Coverage   67.50%   70.38%   +2.88%     
==========================================
  Files          31       33       +2     
  Lines       12668    14593    +1925     
==========================================
+ Hits         8552    10272    +1720     
- Misses       4116     4321     +205     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

codecov-commenter avatar Oct 19 '24 10:10 codecov-commenter

what's the advantage of juliaup here? Also we should keep the names 1.10, 1.11, etc explicitly

wsmoses avatar Oct 19 '24 19:10 wsmoses

juliaup is the recommended way to install and use Julia. The benefits are few but IMHO, we should choose it over setup-julia.

prbzrg avatar Oct 19 '24 20:10 prbzrg

…but for CI where versions are always fixed?

imo we should use whatever reduces setup latency since these jobs are run constantly.

cc @vchuravy if you have thoughts here

wsmoses avatar Oct 19 '24 20:10 wsmoses

~I looked into some CI runs, and it seems that install-juliaup is the faster setup. (1~2 seconds less)~ Both have similar setup time.

prbzrg avatar Oct 19 '24 20:10 prbzrg

If that's the case I really don't see the value add here?

My understanding of juliaup is that essentially it's a script which download's the right julia version binary as required, then dispatches to that. Which, is all good for making users lives' easier.

But in this CI context, we'll never need that functionality, and it feels like it would add extra points of failure/latency, no?

wsmoses avatar Oct 21 '24 00:10 wsmoses

I don't think it would be any more failures. but feel free to close this one. If in the future, there were more points, we can open this again.

prbzrg avatar Oct 23 '24 09:10 prbzrg