Regression in Precompile time with 1.10.0-rc2 compared to the rc1 version

I recently upgraded to the 1.10.0-rc2 version and I see a regression in the precompilation time of my package. There is no option in juliaup to revert back to rc1. Do you also see this with the rc2 version?

juliaup add 1.10.0-rc1

ah, thanks. It wasn’t showing up in the
juliaup list. But is the regression with compile time something that you also see or is it just me?

As I understood, in the rc1 version once a package is precompiled, the next time you load it, its instant. This behavior is not present with the rc2 version. Each time I load the package it precompiles dependencies.

On top of that I see an increase in precompilation time around ~20%.

There is something wrong then. Are you using local environments? What is the output of

using Pkg; Pkg.status()

Interesting:

% juliaup list | grep '1\.10\.0-rc1'
 1.10.0-rc1           1.10.0-rc1+0.x64.linux.gnu      
 1.10.0-rc1~x64       1.10.0-rc1+0.x64.linux.gnu      
 1.10.0-rc1~x86       1.10.0-rc1+0.x86.linux.gnu 

I don’t know, I haven’t tried it. But if you can quantify the regression (0.1%? 1%? 10%? 100%?) and have a reproducer it may be useful to open an issue about it.

1 Like

As I understood, in the rc1 version once a package is precompiled, the next time you load it, its instant. This behavior is not present with the rc2 version. Each time I load the package it precompiles dependencies.

@svretina can you retry with ENV["JULIA_DEBUG"]="loading" and show the output you see when you hit repeat precompilation

Hi,
I am trying to time the precompilation time for each version. In order to force precompilation, do I need to remove the contents of the cached libraries?.

Both of the versions load the cached libraries, according to the ENV["JULIA_DEBUG"]="loading" suggestion from @ianshmean .