fix race condition in mv where the file might get removed after we checked its existance#42253
fix race condition in mv where the file might get removed after we checked its existance#42253KristofferC merged 1 commit intomasterfrom
Conversation
…ecked its existance
|
(We shared the following via other channels as well, but reporting here for good measure and other folks' benefit.) With this patch the failure mode changes to: |
|
We just managed to get our CI pipeline testing the 1.6.3 candidate, and hit an analog under 1.6.3 likely due to the Pkg bump in df3d351. Relevant part of the stacktrace: |
|
For those following along, the commit in Pkg.jl that introduced the issue has been reverted on Pkg.jl's release-1.6 and release-1.7 branches (via the two pull requests linked just above). #42282 brings the relevant Pkg.jl rev onto julialang/julia release-1.6 branch, and #42255 brings it onto the julialang/Julia release-1.7 branch. I'm not sure where that this leaves this pull request. Thoughts Kristoffer? Close this and address the issue more precisely on Pkg.jl master I imagine? :) |
This still looks like an improvement to me though? It turns a scenario from a fatal error to doing what is expected. |
👍 I defer to your better judgement, not being familiar with the subtleties (if any) of |
…ecked its existance (JuliaLang#42253)
…ecked its existance (JuliaLang#42253)
We've observed an error with the following stacktrace:
In short, the
rmin the line changed in this PR errors because the file does not exist, even though we checked its existence a while earlier. This could be because some other process deleted it. Therefore, passforce=truetormso that it is not an error if the file has already been deleted.cc @Sacha0, @NHDaly