Use Base.require() to load Distributed internally #54571
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Base.require_stdlib() will exclusively load a stdlib from whatever was shipped with Julia. The problem is that when developing Distributed.jl this will cause the workers to always load the builtin Distributed module instead of the development version, which can break everything because now the master and worker are running different (potentially incompatible) versions of Distributed. This commit changes Base and LinearAlgebra to use `Base.require()` instead, which will respect `Base.LOAD_PATH`. I argue that this is safe because unlike the other stdlibs like REPL or Pkg, Distributed is only loaded if explicitly requested by the user with `-p` or through `addprocs()` or something, so it shouldn't be possible to get into quite the same tools-are-broken-because-I-broke-my-tools situation that motivated using `Base.require_stdlib()` in the first place. An alternative design would be to: 1. Move the if-block loading Distributed in `exec_options()` below the for-loop where it will execute `-e` options. 2. Require any implementation of `Distributed.launch(::ClusterManager)` to pass `-e 'using Distributed'` in their command to ensure that Distributed is loaded in a way respecting `Base.LOAD_PATH`. This would be more consistent with how the other stdlibs must be developed, but it requires implementers (i.e. Distributed and ClusterManagers.jl) to opt-in to allowing development versions of Distributed, which feels very annoying and easy to miss so I decided not to implement that.
vtjnash
approved these changes
May 24, 2024
59 tasks
KristofferC
pushed a commit
that referenced
this pull request
May 27, 2024
TL;DR: we changed to loading stdlibs exclusively from the default sysimg in #53326, and this breaks developing Distributed.jl because the workers will load the builtin version instead of the development version. I think this should be backported to 1.11? CC @jpsamaroo --- `Base.require_stdlib()` will exclusively load a stdlib from whatever was shipped with Julia. The problem is that when developing Distributed.jl this will cause the workers to always load the builtin Distributed module instead of the development version, which can break everything because now the master and worker are running different (potentially incompatible) versions of Distributed. This commit changes Base and LinearAlgebra to use `Base.require()` instead, which will respect `Base.LOAD_PATH`. I argue that this is safe because unlike the other stdlibs like REPL or Pkg, Distributed is only loaded if explicitly requested by the user with `-p` or through `addprocs()` or something, so it shouldn't be possible to get into quite the same tools-are-broken-because-I-broke-my-tools situation that motivated using `Base.require_stdlib()` in the first place. An alternative design would be to: 1. Move the if-block loading Distributed in `exec_options()` below the for-loop where it will execute `-e` options. 2. Require any implementation of `Distributed.launch(::ClusterManager)` to pass `-e 'using Distributed'` in their command to ensure that Distributed is loaded in a way respecting `Base.LOAD_PATH`. This would be more consistent with how the other stdlibs must be developed, but it requires implementers (i.e. Distributed and ClusterManagers.jl) to opt-in to allowing development versions of Distributed, which feels very annoying and easy to miss so I decided not to implement that. (cherry picked from commit 3b922b0)
KristofferC
removed
the
backport 1.11
DilumAluthge
pushed a commit
that referenced
this pull request
Jun 3, 2024
TL;DR: we changed to loading stdlibs exclusively from the default sysimg in #53326, and this breaks developing Distributed.jl because the workers will load the builtin version instead of the development version. I think this should be backported to 1.11? CC @jpsamaroo --- `Base.require_stdlib()` will exclusively load a stdlib from whatever was shipped with Julia. The problem is that when developing Distributed.jl this will cause the workers to always load the builtin Distributed module instead of the development version, which can break everything because now the master and worker are running different (potentially incompatible) versions of Distributed. This commit changes Base and LinearAlgebra to use `Base.require()` instead, which will respect `Base.LOAD_PATH`. I argue that this is safe because unlike the other stdlibs like REPL or Pkg, Distributed is only loaded if explicitly requested by the user with `-p` or through `addprocs()` or something, so it shouldn't be possible to get into quite the same tools-are-broken-because-I-broke-my-tools situation that motivated using `Base.require_stdlib()` in the first place. An alternative design would be to: 1. Move the if-block loading Distributed in `exec_options()` below the for-loop where it will execute `-e` options. 2. Require any implementation of `Distributed.launch(::ClusterManager)` to pass `-e 'using Distributed'` in their command to ensure that Distributed is loaded in a way respecting `Base.LOAD_PATH`. This would be more consistent with how the other stdlibs must be developed, but it requires implementers (i.e. Distributed and ClusterManagers.jl) to opt-in to allowing development versions of Distributed, which feels very annoying and easy to miss so I decided not to implement that.
lazarusA
pushed a commit
to lazarusA/julia
that referenced
this pull request
Jul 12, 2024
TL;DR: we changed to loading stdlibs exclusively from the default sysimg in JuliaLang#53326, and this breaks developing Distributed.jl because the workers will load the builtin version instead of the development version. I think this should be backported to 1.11? CC @jpsamaroo --- `Base.require_stdlib()` will exclusively load a stdlib from whatever was shipped with Julia. The problem is that when developing Distributed.jl this will cause the workers to always load the builtin Distributed module instead of the development version, which can break everything because now the master and worker are running different (potentially incompatible) versions of Distributed. This commit changes Base and LinearAlgebra to use `Base.require()` instead, which will respect `Base.LOAD_PATH`. I argue that this is safe because unlike the other stdlibs like REPL or Pkg, Distributed is only loaded if explicitly requested by the user with `-p` or through `addprocs()` or something, so it shouldn't be possible to get into quite the same tools-are-broken-because-I-broke-my-tools situation that motivated using `Base.require_stdlib()` in the first place. An alternative design would be to: 1. Move the if-block loading Distributed in `exec_options()` below the for-loop where it will execute `-e` options. 2. Require any implementation of `Distributed.launch(::ClusterManager)` to pass `-e 'using Distributed'` in their command to ensure that Distributed is loaded in a way respecting `Base.LOAD_PATH`. This would be more consistent with how the other stdlibs must be developed, but it requires implementers (i.e. Distributed and ClusterManagers.jl) to opt-in to allowing development versions of Distributed, which feels very annoying and easy to miss so I decided not to implement that.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
TL;DR: we changed to loading stdlibs exclusively from the default sysimg in #53326, and this breaks developing Distributed.jl because the workers will load the builtin version instead of the development version. I think this should be backported to 1.11?
CC @jpsamaroo
Base.require_stdlib()will exclusively load a stdlib from whatever was shipped with Julia. The problem is that when developing Distributed.jl this will cause the workers to always load the builtin Distributed module instead of the development version, which can break everything because now the master and worker are running different (potentially incompatible) versions of Distributed.This commit changes Base and LinearAlgebra to use
Base.require()instead, which will respectBase.LOAD_PATH. I argue that this is safe because unlike the other stdlibs like REPL or Pkg, Distributed is only loaded if explicitly requested by the user with-por throughaddprocs()or something, so it shouldn't be possible to get into quite the sametools-are-broken-because-I-broke-my-tools situation that motivated using
Base.require_stdlib()in the first place.An alternative design would be to:
exec_options()below the for-loop where it will execute-eoptions.Distributed.launch(::ClusterManager)to pass-e 'using Distributed'in their command to ensure that Distributed is loaded in a way respectingBase.LOAD_PATH.This would be more consistent with how the other stdlibs must be developed, but it requires implementers (i.e. Distributed and ClusterManagers.jl) to opt-in to allowing development versions of Distributed, which feels very annoying and easy to miss so I decided not to implement that.