Fix a few allocation/inference issues (#885)

This was from an effort to play around with precompilation some more.
There's a tricky balance where if we precompile too much of `CSV.File`,
then we run into the Base bug where there are Base methods missing and
cause painful allocations via invoke
(JuliaLang/julia#35972). The state of
precompilation for CSV.jl is thus pretty disappointing: precompiling
currently takes ~7.5s (which is fine, this is the number we're fine to
have keep growing), loading CSV.jl precompiled takes ~2.5s (seems a
little high, but not terrible), but the first call to `CSV.File` takes
~6.5s :sad_face:. At this point, I've spent too much time and don't
understand the ins and outs of precompilation interactions with later
runtime code to debug further, but hopefully we can recruit help from
some experts to take CSV.jl precompilation to the next level. As-is,
it's workable, so let's go with this for now and get this darn release
out.

src/CSV.jl

@@ -70,7 +70,8 @@ include("precompile.jl")
 _precompile_()
 
 function __init__()
- # CSV.File(IOBuffer(PRECOMPILE_DATA))
+ CSV.Context(IOBuffer(CSV.PRECOMPILE_DATA))
+ # CSV.File(IOBuffer(CSV.PRECOMPILE_DATA))
  # foreach(row -> row, CSV.Rows(IOBuffer(PRECOMPILE_DATA)))
  # CSV.File(joinpath(dirname(pathof(CSV)), "..", "test", "testfiles", "promotions.csv"))
 end

src/context.jl

@@ -118,7 +118,7 @@ struct Context
  names::Vector{Symbol}
  rowsguess::Int64
  cols::Int
- buf::AbstractVector{UInt8}
+ buf::Vector{UInt8}
  datapos::Int64
  len::Int
  datarow::Int
@@ -141,6 +141,61 @@ struct Context
  streaming::Bool
 end
 
+# user-facing function if just the context is desired
+function Context(source::ValidSources;
+ # file options
+ # header can be a row number, range of rows, or actual string vector
+ header::Union{Integer, Vector{Symbol}, Vector{String}, AbstractVector{<:Integer}}=1,
+ normalizenames::Bool=false,
+ # by default, data starts immediately after header or start of file
+ datarow::Integer=-1,
+ skipto::Integer=-1,
+ footerskip::Integer=0,
+ transpose::Bool=false,
+ comment::Union{String, Nothing}=nothing,
+ ignoreemptyrows::Bool=true,
+ ignoreemptylines=nothing,
+ select=nothing,
+ drop=nothing,
+ limit::Union{Integer, Nothing}=nothing,
+ buffer_in_memory::Bool=false,
+ threaded::Union{Bool, Nothing}=nothing,
+ ntasks::Union{Nothing, Integer}=nothing,
+ tasks::Union{Nothing, Integer}=nothing,
+ rows_to_check::Integer=DEFAULT_ROWS_TO_CHECK,
+ lines_to_check=nothing,
+ # parsing options
+ missingstrings=String[],
+ missingstring="",
+ delim::Union{Nothing, Char, String}=nothing,
+ ignorerepeated::Bool=false,
+ quoted::Bool=true,
+ quotechar::Union{UInt8, Char}='"',
+ openquotechar::Union{UInt8, Char, Nothing}=nothing,
+ closequotechar::Union{UInt8, Char, Nothing}=nothing,
+ escapechar::Union{UInt8, Char}='"',
+ dateformat::Union{String, Dates.DateFormat, Nothing, AbstractDict}=nothing,
+ dateformats=nothing,
+ decimal::Union{UInt8, Char}=UInt8('.'),
+ truestrings::Union{Vector{String}, Nothing}=TRUE_STRINGS,
+ falsestrings::Union{Vector{String}, Nothing}=FALSE_STRINGS,
+ # type options
+ type=nothing,
+ types=nothing,
+ typemap::Dict=Dict{Type, Type}(),
+ pool::Union{Bool, Real, AbstractVector, AbstractDict}=DEFAULT_POOL,
+ downcast::Bool=false,
+ lazystrings::Bool=false,
+ stringtype::StringTypes=DEFAULT_STRINGTYPE,
+ strict::Bool=false,
+ silencewarnings::Bool=false,
+ maxwarnings::Int=DEFAULT_MAX_WARNINGS,
+ debug::Bool=false,
+ parsingdebug::Bool=false
+ )
+ return @refargs Context(source, header, normalizenames, datarow, skipto, footerskip, transpose, comment, ignoreemptyrows, ignoreemptylines, select, drop, limit, buffer_in_memory, threaded, ntasks, tasks, rows_to_check, lines_to_check, missingstrings, missingstring, delim, ignorerepeated, quoted, quotechar, openquotechar, closequotechar, escapechar, dateformat, dateformats, decimal, truestrings, falsestrings, type, types, typemap, pool, downcast, lazystrings, stringtype, strict, silencewarnings, maxwarnings, debug, parsingdebug, false)
+end
+
 @refargs function Context(source::ValidSources,
  # file options
  # header can be a row number, range of rows, or actual string vector

src/file.jl

@@ -213,13 +213,13 @@ function File(source::ValidSources;
  # select=nothing;drop=nothing;limit=nothing;threaded=nothing;ntasks=Threads.nthreads();tasks=nothing;rows_to_check=30;lines_to_check=nothing;missingstrings=String[];missingstring="";
  # delim=nothing;ignorerepeated=false;quoted=true;quotechar='"';openquotechar=nothing;closequotechar=nothing;escapechar='"';dateformat=nothing;
  # dateformats=nothing;decimal=UInt8('.');truestrings=nothing;falsestrings=nothing;type=nothing;types=nothing;typemap=Dict{Type,Type}();
- # pool=CSV.DEFAULT_POOL;downcast=false;lazystrings=false;stringtype=String;strict=false;silencewarnings=false;maxwarnings=100;debug=true;parsingdebug=false;buffer_in_memory=false
+ # pool=CSV.DEFAULT_POOL;downcast=false;lazystrings=false;stringtype=String;strict=false;silencewarnings=false;maxwarnings=100;debug=false;parsingdebug=false;buffer_in_memory=false
  # @descend CSV.Context(CSV.Arg(source), CSV.Arg(header), CSV.Arg(normalizenames), CSV.Arg(datarow), CSV.Arg(skipto), CSV.Arg(footerskip), CSV.Arg(transpose), CSV.Arg(comment), CSV.Arg(ignoreemptyrows), CSV.Arg(ignoreemptylines), CSV.Arg(select), CSV.Arg(drop), CSV.Arg(limit), CSV.Arg(buffer_in_memory), CSV.Arg(threaded), CSV.Arg(ntasks), CSV.Arg(tasks), CSV.Arg(rows_to_check), CSV.Arg(lines_to_check), CSV.Arg(missingstrings), CSV.Arg(missingstring), CSV.Arg(delim), CSV.Arg(ignorerepeated), CSV.Arg(quoted), CSV.Arg(quotechar), CSV.Arg(openquotechar), CSV.Arg(closequotechar), CSV.Arg(escapechar), CSV.Arg(dateformat), CSV.Arg(dateformats), CSV.Arg(decimal), CSV.Arg(truestrings), CSV.Arg(falsestrings), CSV.Arg(type), CSV.Arg(types), CSV.Arg(typemap), CSV.Arg(pool), CSV.Arg(downcast), CSV.Arg(lazystrings), CSV.Arg(stringtype), CSV.Arg(strict), CSV.Arg(silencewarnings), CSV.Arg(maxwarnings), CSV.Arg(debug), CSV.Arg(parsingdebug), CSV.Arg(false))
  ctx = @refargs Context(source, header, normalizenames, datarow, skipto, footerskip, transpose, comment, ignoreemptyrows, ignoreemptylines, select, drop, limit, buffer_in_memory, threaded, ntasks, tasks, rows_to_check, lines_to_check, missingstrings, missingstring, delim, ignorerepeated, quoted, quotechar, openquotechar, closequotechar, escapechar, dateformat, dateformats, decimal, truestrings, falsestrings, type, types, typemap, pool, downcast, lazystrings, stringtype, strict, silencewarnings, maxwarnings, debug, parsingdebug, false)
  return File(ctx)
 end
 
-function File(ctx::Context, chunking::Bool=false)
+function File(ctx::Context, @nospecialize(chunking::Bool=false))
  @inbounds begin
  # we now do our parsing pass over the file, starting at datapos
  if ctx.threaded
@@ -235,6 +235,7 @@ function File(ctx::Context, chunking::Bool=false)
  rows = zeros(Int64, ntasks) # how many rows each parsing task ended up actually parsing
  @sync for i = 1:ntasks
  Threads.@spawn multithreadparse(ctx, pertaskcolumns, rowchunkguess, i, rows, wholecolumnslock)
+ # CSV.multithreadparse(ctx, pertaskcolumns, rowchunkguess, i, rows, wholecolumnslock)
  end
  finalrows = sum(rows)
  if ctx.limit < finalrows
@@ -740,9 +741,7 @@ end
 function detectcell(buf, pos, len, row, rowoffset, i, col, ctx, rowsguess)::Tuple{Int64, Int16}
  # debug && println("detecting on task $(Threads.threadid())")
  opts = col.options
- # stats = Base.gc_num()
  code, tlen, x, xT = detect(pass, buf, pos, len, opts, false, ctx.downcast, rowoffset + row, i)
- # diff = GC_Diff(gc_num(), stats)
  if x === missing
  col.anymissing = true
  @goto finaldone

src/precompile.jl

@@ -1,7 +1,8 @@
 const PRECOMPILE_DATA = "int,float,date,datetime,bool,null,str,catg,int_float\n1,3.14,2019-01-01,2019-01-01T01:02:03,true,,hey,abc,2\n2,NaN,2019-01-02,2019-01-03T01:02:03,false,,there,abc,3.14\n"
 function _precompile_()
- # ccall(:jl_generating_output, Cint, ()) == 1 || return nothing
- # CSV.File(IOBuffer(PRECOMPILE_DATA))
+ ccall(:jl_generating_output, Cint, ()) == 1 || return nothing
+ while false; end
+ # CSV.Context(IOBuffer(CSV.PRECOMPILE_DATA))
  # foreach(row -> row, CSV.Rows(IOBuffer(PRECOMPILE_DATA)))
- # CSV.File(joinpath(dirname(pathof(CSV)), "..", "test", "testfiles", "promotions.csv"))
+ CSV.Context(joinpath(dirname(pathof(CSV)), "..", "test", "testfiles", "promotions.csv"))
 end