whos() considerably slow

[hyiltiz]

julia> @time whos()
Base 24573 KB Module : Base
Core 3043 KB Module : Core
Main 808579 KB Module : Main
ans 0 bytes Void : nothing
x 781250 KB 10000x10000 Array{Float64,2} : [0.2345839889343943…
α̂₂ 8 bytes Int64 : 2
203.371041 seconds (1.58 G allocations: 34.022 GB, 4.22% gc time)

[andreasnoack]

That is really bad. The problem is mainly this line which prints the object. Unfortunately, it prints all the values of the array, which takes quite long time because your array is large. It shouldn't print all the values, but instead do something like we are doing when printing an array in the REPL, but our array printing code is really convoluted, so I'm not sure there is an easy fix.

Hopefully #14052 will be a long term fix for easier array printing, but I'm not sure it is worth the trouble to print the objects in whos. It is mainly their size and type that are interesting.

[Keno]

Wouldn't replacing that line with showcompact be sufficient?

[hyiltiz]

I am fairly new to Julia, and am really fascinated by it. All I did was x = rand(1000,1000) and a call to whos(). But I really do not expect to wait about a minute to find out about my workspace. I am not sure what needs to be done, but definitely not trying to print implicitly or explicitly the variables in the workspace.

[andreasnoack]

Unfortunately, showcompact won't work. It also prints all the values of the array and I think it ends up calling the same method. The more efficient printing in the REPL completely bypasses show and showcompact using writemime and showarray. The reasons for the two different paths for array printing are a bit of a mystery because I don't think it's ever a good idea to fetch all the values of big arrays. For DArray this would probably make the system hang forever.

[hyiltiz]

@andreasnoack So is that a feature to come in a future release?

[andreasnoack]

@hyiltiz The fix in #14177 should be possible to backport. The more general refactoring of the printing functions is for future releases.

[hyiltiz]

I really liked the current behavior (printing head and tail of a big matrix) of the printing function. I hope #14177 does not change that. Instead, whos() only needs to represent properties of the variables in the namespace. So, instead of showing everything, a (maybe hardcoded) head -n 5 could be enough as far as whos() functionality is concerned.

[JeffBezanson]

We should not be trying to show whole values during whos. We should use something like summary.

[andreasnoack]

@JeffBezanson That is also what I've ended up doing in #14177.