low level API: how to skip field, test for EOL, EOF about csv.jl HOT 3 CLOSED

To test if a CSV.Source is done, I would use

Data.isdone(s, row, col)

but otherwise, eof(s.io) should work fine too.

1. There's an internal CSV.STATE variable of type Ref{CSV.ParsingState}. CSV.ParsingState is an enum with values

@enum ParsingState None Delimiter EOF Newline

Basically, each call to CSV.parsefield implicitly sets this CSV.STATE variable with the state of parsing when the parsing was finished:

• None: the default state
• Delimiter: parsing encountered a delimiter character which ended the CSV.parsefield function
• EOF: parsing encountered the EOF
• Newline: parsing encountered a newline which caused parsing to stop

1. Currently there is no way to "skip" a field, though I have some notes jotted down somewhere with some considerations around doing something like this. I'll try to dig them up and post them here and maybe see how much work it would take to get it working. For now, though, what you're doing is best practice (just parse a string and continue on).

from csv.jl.

Thanks, this works fine. Could you consider adding API functions eof and eol, ideally with a CSV.Source as their argument? (I am assuming that eventually the state will be migrated into CSV.Source to avoid globals, but I could be mistaken as I did not study the source deeply).

from csv.jl.

On current master, you can now do CSV.File(file; kwargs...) which returns a lazy-materializing row-iterator. You could accomplish the original goal by doing something like:

struct CustomType
    a::Int
    b::Float64
    c::String
end

customtypes = CustomType[]

for row in CSV.File(file; kwargs...)
    if should_keep(row.a)
        push!(customtypes, CustomType(row.a, row.b, row.c)
    end
end

In this case, only the a, b, and c column values will actually be parsed from the csv file.

from csv.jl.