Allow the user to set julia-args for precompile execution
#931
Conversation
It seems reasonable to me that PackageCompiler should just hardcode |
|
I guess some precompilation workloads are not particularly light computationally and could take longer without optimizations... But yeah maybe |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #931 +/- ##
=======================================
Coverage 84.62% 84.62%
=======================================
Files 3 3
Lines 826 826
=======================================
Hits 699 699
Misses 127 127 ☔ View full report in Codecov by Sentry. |
ah true, good point. 👍 |
|
Are the tests on Nightly unrelated? |
|
I checked a couple of nighly failures and there seems to be a bounds error: |
|
This is a good idea. My hope is that propagating and adding all sorts of these configurations will be easier with, #929 Can we split the difference and do |
| @@ -834,7 +843,8 @@ function create_app(package_dir::String, | |||
| sysimage_build_args::Cmd=``, | |||
| include_transitive_dependencies::Bool=true, | |||
| include_preferences::Bool=true, | |||
| script::Union{Nothing, String}=nothing) | |||
| script::Union{Nothing, String}=nothing, | |||
| precompile_execution_args::Cmd=`-O0`) | |||
There was a problem hiding this comment.
Can we split the difference and do -O1?
There was a problem hiding this comment.
Seems plausible to me, but what is that based on, @sjkelly? My gut is that -O0 would be better for almost all snoop workloads, since the developer is already pretty strongly incentivized to keep data sizes small, and thus to have compilation-dominated workloads.
What leads you to prefer -O1?
So they can run their potentially expensive precompile execution scripts with
-O0.Not sure if hardcoding
-O0would be preferable... I'm open to suggestions!