-
-
Notifications
You must be signed in to change notification settings - Fork 5.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Intermittently broken precompile #13200
Comments
I've seen this kind of thing several times (on OS X). Usually it's fine the next time I start julia though. |
Ref #13182 (comment) This might not be OSX specific (but it only happens on Travis OSX and not Travis Linux AFAICT) |
I've just seen this locally as well. It happend when I run |
It can be reproduced ~50% of the time running the World of Julia script. |
This happens 100% of the time if you run
edit: this is actually probably a JSON.jl issue, not a base issue: JuliaIO/JSON.jl#123 (comment) |
I just saw this, not sure if it's related, but seems bad as well:
|
That error means that you've got some conflict between |
I got this to happen fairly reliably when doing |
My issue which was closed as a dupe includes a reproducible (simple) example which depends on no external packages. The problem occurs when a dependent module 'uses' a package which does not have precompile set and then you use this module. This only occurs first time round with an empty lib directory. |
that is simply a usage error. |
@vtjnash as a user error I'd expect the top level package to error on precompile, it does not and causes the crash outlined here. I'd happily see the the precompile error out if all child modules are not so annotated, but I suspect it will break current usage. Perhaps a warning ? The reproduction supplied boils the issue down to something which does not require a lot of moving parts, but is based on the real problem - The manifestation of this issue in iJulia etc is due to BinDeps.jl not having precompile set and Conda.jl does (which calls import BinDeps). In isolation this causes no issue (e.g. when you call these on their own), but when you install a package through REQUIRE the 'build step' breaks (this is due to the reuse of the build child process in the package manager.) Specifically you can reproduce the problem by having LibCURL and PyCall in a parent REQUIRE file. The reason that people see this as intermittent is that it depends on the order of the build and the set of packages specified in the build. Additionally once you have successfully precompiled you are good to go from then on. It is only when you try and build a package with many dependencies in a clean area that this rears its head. I hope that helps. |
If BinDeps is safe to precompile, we could fix that case by adding the precompile directive to it. Better checking, if possible, would be worth trying to do. |
It certainly seems safe since any users of Conda are doing this today. |
I think we should reopen this. If we don't support precompiling a module whose dependencies can't be precompiled, we should throw an error as soon as possible to prevent unpredictable breakage later. |
that's not really something the system can test for |
We can't check whether dependencies have marked |
FWIW, I'm running into this today with Actually, I'm getting the Edit: This was fixed by nuking |
I've just installed VS Code and Julia v0.7 and i'm having this same problem: C:\Users\joe.vscode-insiders\extensions\julialang.language-julia-0.9.1\scripts\languageserver\packages\Nullables\src\nullable.jl:392 and so on |
Update: Just to be clear, this issue doesn't seem to be OSX specific althought it does happen more often on OSX travis.
Original post:
This started to happen on the OSX Travis CI since a few days ago for both 0.4 and master. (Also possible that the OSX travis was just not enabled before that). This is a typical failed build (output attached in case someone restarted that build).
The corresponding Linux build looks fine.
The text was updated successfully, but these errors were encountered: