Reducing startup time by reusing the pre-compiled directory. about evcxr HOT 6 OPEN

I'm guessing your config is loading some crates? Setting up sccache might save some time, but it won't make 1 as fast as 2. Using EVCXR_TMPDIR, as you were before should make 1 as fast as 2, but as you found in the other thread, has the downside that you can only have one copy of evcxr open at a time.

I don't think there's currently any ideal solution. The issue stems from there not really being a good way to share a target directory between multiple cargo workspaces.

from evcxr.

I tested the processes with both using EVCXR_TMPDIR, but the 2 is 10 times faster than 1. I'm not sure whether I described the processes clearly, I can make a gif to show the 1 and 2 if it is needed.

from evcxr.

What's your ~/.config/evcxr/init.evcxr like? Also, your ~/.config/evcxr/prelude.evcxr if you've got one? Mostly what I'm interested in is how many :dep commands you have and in what file. If you've got more than 1 :dep command in init.evcxr, have you tried with evcxr version 0.15.0, which was released earlier today? It included a change that would have improved startup performance for the case of having more than one :dep command in init.evcxr.

from evcxr.

I tested above with evcxr version 0.14.2(it is unpublished in that time, 3 days ago) in windows. The init.evcxr is empty, the prelude.rs is following.

I just know a new version has published, I'm testing the same processes with new version.

from evcxr.

Wow, that's quite a lot of crates. Thinking about it some more, my guess is that the difference in startup time could be rust-analyzer processing all those crates. Unfortunately there's currently no way to cache the analysis that rust-analyzer does. I'll have a look at adding some extra instrumentation to check what's taking time so that we can confirm if that's what's taking time.

from evcxr.

I test the following processes with using version 0.15.0:
(there is a test.ipynb to test, which has the first cell :clear :load_config )

1. I start my computer(windows 11), then open vscode and open the test.ipynb file. Then I run the first cell, while recording the time of finishing running.
2. I click the restart kernel in notebook, and waiting the vscode info (show up in the left bottom corner after I clicked the restart kernel button)Restarting kernel: Rust: Waiting for Jupyter Session to be idle to finish(it takes a few seconds). Then I run the first cell, while recording the time of finishing running.
3. I run the first cell, while recording the time of finishing running.
   The consuming time of the above four processes.
   1 . 60 secs,
   2 . 6 secs,
   3 . 6 secs.

from evcxr.