CSV.read leaves file open on Windows 7?

[jaakkor2]

using CSV
CSV.read("data.txt")

looks like to leave the file open in Windows 7. If I try to delete the file data.txt in Windows Explorer, I get the dialog "File In Use; The action can't be completed because the file is open in Julia Programming Language; Close the file and try again". If I close Julia, I can delete the file. However, I can delete the file in Git Bash without closing Julia.

On Mac it seems to work fine.

Versions

• Julia Version 0.5.0
• CSV v"0.1.2"

[quinnj]

Hmmm.....I'm not sure we can do anything here; we're calling Mmap.mmap from julia Base and I think that's just part of the behavior on windows with mmapping (i.e. you're not allowed to delete a file that is currently mmapped). As long as the CSV.Source is in scope, I think we have to expect this behavior. But if you're just calling CSV.read, you can probably call gc(); gc(); afterwards and then be able to delete the file (this forces a garbage collection on the mmap and should release the file to be deleted).

[jaakkor2]

Thanks. That workaround works.

CSV.read("data.txt")
if is_windows()
    gc()
    gc()
end

Does not look nice though. If there is no handle of the file given to the user, I think the memory should be unmapped after reading.

[jaakkor2]

This works as well

a=CSV.read("data.txt"; use_mmap=false)

[bkamins]

readdlm in Julia Base by default sets use_mmap to false on Windows by testing is_windows() ? false : true probably because of the problem mentioned in this issue.

Maybe the same approach could be used in definition of Source like:

Source(fullpath::Union{AbstractString,IO};

              delim=COMMA,
              quotechar=QUOTE,
              escapechar=ESCAPE,
              null::AbstractString="",

              header::Union{Integer,UnitRange{Int},Vector}=1, # header can be a row number, range of rows, or actual string vector
              datarow::Int=-1, # by default, data starts immediately after header or start of file
              types::Union{Dict{Int,DataType},Dict{String,DataType},Vector{DataType}}=DataType[],
              nullable::Bool=true,
              weakrefstrings::Bool=true,
              dateformat::Union{AbstractString,Dates.DateFormat}=Dates.ISODateFormat,

              footerskip::Int=0,
              rows_for_type_detect::Int=100,
              rows::Int=0,
              use_mmap::Bool=is_windows() ? false : true)

[quinnj]

Here's the tradeoff: mmap is much faster, and this issue only comes up when you read a file and then quickly try to delete the underlying file. I could add some documentation around what needs to happen to safely delete, but it seems worth it to keep the speedups by using mmap by default on windows.

[bkamins]

I would find this additional documentation very useful (and maybe the conclusions could be also used in readdlm in Base). Is the problem only related to deletion of the file or also to its modification?