Check for conflicting @ccallable name before JIT registration#55813
Check for conflicting @ccallable name before JIT registration#55813topolarity wants to merge 1 commit intoJuliaLang:masterfrom
@ccallable name before JIT registration#55813Conversation
This turns the existing segfault into a warning, as on 1.9: ```julia julia> using Foo, Bar WARNING: @ccallable was already defined for this method name ``` Resolves JuliaLang#54878
43c389a to
450fc9d
Compare
|
Should we instead turn it into a proper error? Especially since we are deleting an assert? |
I think we should do that, but also make These functions aren't actually available to
The assert was just wrong - it assumed there is only one (pkg/sys)image loaded system-wide and therefore naming conflicts cannot occur. |
I think so that Lines 154 to 177 in c3af4fc |
Backported PRs: - [x] #55773 <!-- Add compat entry for `Base.donotdelete` --> - [x] #41244 <!-- Fix shell `cd` error when working dir has been deleted --> - [x] #55795 <!-- fix #52986, regression in `@doc` of macro without REPL loaded --> - [x] #55829 <!-- [Dates] Make test more robust against non-UTC timezones --> - [x] #55641 <!-- fall back to slower stat filesize if optimized filesize fails --> - [x] #55744 <!-- fix #45494, error in ssa conversion with complex type decl --> - [x] #55783 <!-- use `inferencebarrier` instead of `invokelatest` for 1-arg `@assert` --> - [x] #55739 <!-- Add `invokelatest` barrier to `string(...)` in `@assert` --> Need manual backport: - [ ] #55798 <!-- Broadcast binary ops involving strided triangular --> Contains multiple commits, manual intervention needed: - [ ] #55509 <!-- Fix cong implementation to be properly random and not just cycling. --> - [ ] #55569 <!-- Add a docs section about loading/precomp/ttfx time tuning --> - [ ] #55824 <!-- Replace regex package module checks with actual code checks --> Non-merged PRs with backport label: - [ ] #55845 <!-- privatize annotated string API, take two --> - [ ] #55828 <!-- Fix some corner cases of `isapprox` with unsigned integers --> - [ ] #55813 <!-- Check for conflicting `@ccallable` name before JIT registration --> - [ ] #55743 <!-- doc: heap snapshot viewing --> - [ ] #55741 <!-- Change annotations to use a NamedTuple --> - [ ] #55534 <!-- Set stdlib sources as read-only during installation --> - [ ] #55499 <!-- propagate the terminal's `displaysize` to the `IOContext` used by the REPL --> - [ ] #55458 <!-- Allow for generically extracting unannotated string --> - [ ] #55457 <!-- Make AnnotateChar equality consider annotations --> - [ ] #55453 <!-- Privatise the annotations API, for StyledStrings --> - [ ] #55355 <!-- relocation: account for trailing path separator in depot paths --> - [ ] #55220 <!-- `isfile_casesensitive` fixes on Windows --> - [ ] #55169 <!-- `propertynames` for SVD respects private argument --> - [ ] #54457 <!-- Make `String(::Memory)` copy --> - [ ] #53957 <!-- tweak how filtering is done for what packages should be precompiled --> - [ ] #51479 <!-- prevent code loading from lookin in the versioned environment when building Julia --> - [ ] #50813 <!-- More doctests for Sockets and capitalization fix --> - [ ] #50157 <!-- improve docs for `@inbounds` and `Base.@propagate_inbounds` -->
|
Hmm yeah - feels like that behavior ought to be opt-in though |
Another variant how this is used (since I did encounter this myself again) Really, that's the same error we want to give here, and that the error you would get in a systemimage. |
Throwing the error isn't really the annoying part imo - it's that it's not always avoiding a real problem:
Which is why I think it might be better to separate these two use cases |
|
For my use case, I wanted to test side-by-side two versions of a package for use with PackageCompiler. I ended up writing a terrible wrapper to save all the ccallable registrations, and then defer their registration until an explicit point in time that I can activate in the PackageCompiler build Otherwise it was impossible to import both modules without warnings / errors |
Yeah, but that is the universe we live in. Even if I want those packages for use in PackageCompiler I need to use prefixes to ensure that my names are unique. The question is "could we come up with a better namespace strategy than the user"? Maybe, we could automatically add the package (module?, FQN?) as a prefix. Even for your use-case you likely had to do some shenaigans like changing the UUID and renaming a package to |
Yeah, for PackageCompiler you're specifically exporting for the linker, so we're forced to accept the realities of C namespacing But only when you're actually building the sysimage - so for my use case, I can load vA and vB just fine in an interactive session
I think it's pretty much equivalent to the gensym problem - but I have to admit I don't know how much we try to make gensym unique across pkgimages |
Backported PRs: - [x] #55849 <!-- Mmap: fix grow! for non file IOs --> - [x] #55863 <!-- Update TaskLocalRNG docstring according to #49110 --> - [x] #54433 <!-- Root globals in toplevel exprs --> - [x] #55828 <!-- Fix some corner cases of `isapprox` with unsigned integers --> - [x] #55890 <!-- Profile: fix order of fields in heapsnapshot & improve formatting --> - [x] #55884 <!-- inference: add missing `TypeVar` handling for `instanceof_tfunc` --> - [x] #55881 <!-- Install terminfo data under /usr/share/julia --> - [x] #55909 <!-- do not intentionally suppress errors in precompile script from being reported or failing the result --> - [x] #55355 <!-- relocation: account for trailing path separator in depot paths --> - [x] #55917 <!-- fix rawbigints OOB issues --> - [x] #55892 <!-- TOML: Avoid type-pirating `Base.TOML.Parser` --> - [x] #55798 <!-- Broadcast binary ops involving strided triangular --> - [x] #55919 <!-- Limit `@inbounds` to indexing in the dual-iterator branch in `copyto_unaliased!` --> Contains multiple commits, manual intervention needed: - [ ] #54009 <!-- allow extensions to trigger from packages in [deps] --> - [ ] #55509 <!-- Fix cong implementation to be properly random and not just cycling. --> - [ ] #55569 <!-- Add a docs section about loading/precomp/ttfx time tuning --> - [ ] #55824 <!-- Replace regex package module checks with actual code checks --> Non-merged PRs with backport label: - [ ] #55932 <!-- REPL: make UndefVarError aware of imported modules --> - [ ] #55910 <!-- Prevent extensions from blocking parallel pre-compilation --> - [ ] #55908 <!-- add logic to prefer loading modules that are already loaded --> - [ ] #55886 <!-- irrationals: restrict assume effects annotations to known types --> - [ ] #55871 <!-- lowering: don't reverse handler order in `(pop-handler-list ...)` --> - [ ] #55870 <!-- fix infinite recursion in `promote_type` for `Irrational` --> - [ ] #55867 <!-- update `hash` doc string: `widen` not required any more --> - [ ] #55851 <!-- [REPL] Fix #55850 by using `safe_realpath` instead of `abspath` in `projname` --> - [ ] #55813 <!-- Check for conflicting `@ccallable` name before JIT registration --> - [ ] #55743 <!-- doc: heap snapshot viewing --> - [ ] #55741 <!-- Change annotations to use a NamedTuple --> - [ ] #55534 <!-- Set stdlib sources as read-only during installation --> - [ ] #55499 <!-- propagate the terminal's `displaysize` to the `IOContext` used by the REPL --> - [ ] #55458 <!-- Allow for generically extracting unannotated string --> - [ ] #55457 <!-- Make AnnotateChar equality consider annotations --> - [ ] #55220 <!-- `isfile_casesensitive` fixes on Windows --> - [ ] #55169 <!-- `propertynames` for SVD respects private argument --> - [ ] #54457 <!-- Make `String(::Memory)` copy --> - [ ] #53957 <!-- tweak how filtering is done for what packages should be precompiled --> - [ ] #51479 <!-- prevent code loading from lookin in the versioned environment when building Julia --> - [ ] #50813 <!-- More doctests for Sockets and capitalization fix --> - [ ] #50157 <!-- improve docs for `@inbounds` and `Base.@propagate_inbounds` -->
|
This is no longer relevant now that #56987 has merged |
Fixes #54878. For pkgimages with conflicting
@ccallabledefinitions:This turns the existing segfault into a warning, as on 1.9: