Next breaking release

[timholy]

If I'm not mistaken, #379 and #381 are breaking changes. Before the next release, I'd propose we:

• restrict to Julia 1.10 and higher (Remove cruft #384)
• eliminate SnoopCompileBot (rm SnoopCompileBot #383)
• eliminate SnoopPrecompile (Remove cruft #384)
• eliminate @snoopc and @snoopi (Remove cruft #384)
• possibly rename @snoopi_deep to just @snoopi or @snoop_infer (Rename the macros #386)
• update documentation (Rewrite the docs #389)

Other thoughts?

[aviatesk]

Thanks for putting together the list. Let's focus on finishing it for now. We might come across additional tasks as we go along.

[mkitti]

In #379 we eliminated the compat declaration for JET.jl. I'm not sure if that was necessary since you can also keep a compact section as long as you also add the package to extras:
https://pkgdocs.julialang.org/v1/creating-packages/#Weak-dependencies

Also, since we are restricting to Julia 1.10 and above here though, perhaps we can restrict the JET.jl compatibility to tighter bounds that make sense for Julia 1.10?

[mkitti]

Regarding release time, releasing this when Julia 1.11 comes out should be an objective. I'm not sure how long we have until then or if we have sufficient time to accomplish our goals above. While I see that we are trying to aggregate breaking changes and this is a good goal, I think that should be secondary to have a working package for a released and stable version of Julia.

[timholy]

If needed I think we could get this out by next week. Given the number of open milestones for Julia 1.11 I don't think it will be hard to have this ready in time for 1.11.

[aviatesk]

I support releasing version 3.0 as soon as we're ready, regardless of when 1.11 is released.

[timholy]

The only really big job now is rewriting the documentation. SnoopCompile 2.x was written when we didn't cache native code and didn't have PrecompileTools, and I suspect that this assumption is pervasive in the documentation.

For example, much of the documentation focuses on fixing inference problems, because before PrecompileTools there wasn't a good way to ensure that type-inferred got stored in the cache unless the chain of inference was unbroken. That's not nearly so relevant anymore. Conversely, what is relevant now is the edge-case of missing precompiled code because the package calls a method during module-definition outside a @compile_workload. Explaining what can go wrong there and how to detect and fix it is not covered in the current documentation, and should be.

[oscardssmith]

What's the status on this? I was just updating packages and realized that it is not currently possible to have latest versions of both JET and SnoopCompile installed.

[timholy]

The status is revealed in the checklist above. The only scary one is "update documentation" because there's a huge amount of documentation, and since precompilation the whole landscape has changed.

[timholy]

But with 1.11 nearing release, I'm motivated to fix this, so it won't be long now.

[timholy]

All done! 3.0.0 is on its way out: JuliaRegistries/General#111692