Prefer header delimiters in auto detection

docs/src/examples.md

@@ -525,9 +525,9 @@ file = CSV.File(IOBuffer(data); dateformat="yyyy/mm/dd")
 ```julia
 using CSV
 
-# In many places in the world, floating point number decimals are separated with a comma instead of a period (`3,14` vs. `3.14`)
-# . We can correctly parse these numbers by passing in the `decimal=','` keyword argument. Note that we probably need to
-# explicitly pass `delim=';'` in this case, since the parser will probably think that it detected `','` as the delimiter.
+# In many places in the world, floating point number decimals are separated with a comma instead of a period (`3,14` vs. `3.14`).
+# We can correctly parse these numbers by passing in the `decimal=','` keyword argument. If the file has no header or
+# delimiter detection is otherwise ambiguous, pass `delim=';'` explicitly so commas are treated only as decimal markers.
 data = """
 col1;col2;col3
 1,01;2,02;3,03

src/detection.jl

@@ -119,12 +119,25 @@ function detectdelimandguessrows(buf, headerpos, datapos, len, oq, eq, cq, @nosp
     if delim == UInt8('\n')
         if nlines > 0
             d = UInt8('\n')
-            for attempted_delim in (UInt8(','), UInt8('\t'), UInt8(' '), UInt8('|'), UInt8(';'), UInt8(':'))
-                cnt = bvc.counts[Int(attempted_delim) + 1]
-                # @show Char(attempted_delim), cnt, nlines
-                if cnt > 0 && cnt % nlines == 0
-                    d = attempted_delim
-                    break
+            if headerpos > 0
+                # Prefer delimiters represented in the header over incidental data characters.
+                for attempted_delim in (UInt8(','), UInt8('\t'), UInt8(' '), UInt8('|'), UInt8(';'), UInt8(':'))
+                    hcnt = headerbvc.counts[Int(attempted_delim) + 1]
+                    cnt = bvc.counts[Int(attempted_delim) + 1]
+                    if hcnt > 0 && cnt > 0 && cnt % nlines == 0
+                        d = attempted_delim
+                        break
+                    end
+                end
+            end
+            if d == UInt8('\n')
+                for attempted_delim in (UInt8(','), UInt8('\t'), UInt8(' '), UInt8('|'), UInt8(';'), UInt8(':'))
+                    cnt = bvc.counts[Int(attempted_delim) + 1]
+                    # @show Char(attempted_delim), cnt, nlines
+                    if cnt > 0 && cnt % nlines == 0
+                        d = attempted_delim
+                        break
+                    end
                 end
             end
             if d == UInt8('\n')

test/basics.jl

@@ -132,6 +132,12 @@ f = CSV.File(IOBuffer("x\ty\n1\t\n2\t\n"))
 f = CSV.File(IOBuffer("x|y\n1|\n2|\n"))
 @test (length(f), length(f.names)) == (2, 2)
 
+f = CSV.File(IOBuffer("a;b;c\n1,1;2,2;3,3\n4,4;5,5;6,6\n"); decimal=',')
+@test (length(f), length(f.names)) == (2, 3)
+@test f.a == [1.1, 4.4]
+@test f.b == [2.2, 5.5]
+@test f.c == [3.3, 6.6]
+
 # type promotion
 # int => float
 f = CSV.File(IOBuffer("x\n1\n3.14"))