Don't specialize groupreduce! on result array element type #2335
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
Works around crash seen in #2326. The inferred return type of `groupreduce_init` is `Union{Vector{Any}, SentinelVector{Float64}}` and it seems the compiler then crashes when trying to correctly identify `U` from that union of types. Part of my conclusion here is based on the fact that if you remove all other argument type constraints and just make `groupreduce!` return `res` directly, it still crashes; thus, by deduction, the crash has something to do with the compiler having trouble with the `::AbstractVector{U}` argument type constraint/specialization. The work-around is pretty uncontroversial; we were already calling `eltype(res)` in several other places, and I've checked that it infers the same. I didn't add a test since this seems like such an obscure compiler bug that it doesn't really seem necessary for DataFrames to be testing core compiler behavior. Also note that this bug exists in Julia <= 1.5, so current Julia master (pending 1.6), which includes a number of compiler refactorings/changes, seems to have resolved whatever the issue was. cc: @Keno, @vtjnash, @JeffBezanson
bkamins
approved these changes
Jul 27, 2020
bkamins
reviewed
Jul 27, 2020
| @@ -962,8 +962,8 @@ function copyto_widen!(res::AbstractVector{T}, x::AbstractVector) where T | |||
| return res | |||
| end | |||
|
|
|||
| function groupreduce!(res::AbstractVector{U}, f, op, condf, adjust, checkempty::Bool, | |||
| incol::AbstractVector{T}, gd::GroupedDataFrame) where {U,T} | |||
| function groupreduce!(res, f, op, condf, adjust, checkempty::Bool, | |||
There was a problem hiding this comment.
Just maybe leve res::AbstractVector for documentation reasons. I understand it should be OK, or this also crashes?
|
I understand it would be good to backport it to 0.21 branch - right? If yes I will do it. |
|
Yeah, I think that's a good idea. |
|
(a small note: as a rule we typically squash-merge the PRs in DataFrames.jl) |
bkamins
pushed a commit
that referenced
this pull request
Jul 27, 2020
Don't specialize groupreduce! on result array element type
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.
Works around crash seen in #2326. The inferred return type of
groupreduce_initisUnion{Vector{Any}, SentinelVector{Float64}}andit seems the compiler then crashes when trying to correctly identify
Ufrom that union of types. Part of my conclusion here is based on the
fact that if you remove all other argument type constraints and just
make
groupreduce!returnresdirectly, it still crashes; thus, bydeduction, the crash has something to do with the compiler having
trouble with the
::AbstractVector{U}argument typeconstraint/specialization.
The work-around is pretty uncontroversial; we were already calling
eltype(res)in several other places, and I've checked that it infersthe same. I didn't add a test since this seems like such an obscure
compiler bug that it doesn't really seem necessary for DataFrames to be
testing core compiler behavior.
Also note that this bug exists in Julia <= 1.5, so current Julia master
(pending 1.6), which includes a number of compiler refactorings/changes,
seems to have resolved whatever the issue was.
cc: @Keno, @vtjnash, @JeffBezanson