Fix incorrect length of UTF-8 string for pango layout

[tomyun]

I believe this is also related to GiovineItalia/Gadfly.jl#1410.

julia> using Compose
julia> import Cairo, Fontconfig

before

julia> Compose.text_extents("", 8pt, "α")

(process:35143): Pango-WARNING **: 21:27:47.868: pango_layout_set_markup_with_accel: Error on line 1 char 10: Invalid UTF-8 encoded text in name — not valid “\xce”
1-element Array{Tuple{Measures.Length{:mm,Float64},Measures.Length{:mm,Float64}},1}:
 (1.5075737847222221mm, 1.522732204861111mm)

julia> Compose.text_extents("", 8pt, "αα")
1-element Array{Tuple{Measures.Length{:mm,Float64},Measures.Length{:mm,Float64}},1}:
 (1.5075737847222221mm, 1.522732204861111mm)

Note that the first case complained about a clipped string array, but still got a correct width. The length of UTF-8 string array for "α" supposed to be 2 ("\xce\xb1"), but pango_text_extents internally used a length of Julia String which was 1 for this.

The second case seemed working fine with no complaints, but due to spurious string length, only the first letter was accounted for layout. That's why we had same widths for them which are not we'd expect.

after

julia> Compose.text_extents("", 8pt, "α")
1-element Array{Tuple{Measures.Length{:mm,Float64},Measures.Length{:mm,Float64}},1}:
 (1.5075737847222221mm, 1.522732204861111mm)

julia> Compose.text_extents("", 8pt, "αα")
1-element Array{Tuple{Measures.Length{:mm,Float64},Measures.Length{:mm,Float64}},1}:
 (3.231499565972222mm, 1.522732204861111mm)

Now we no longer have pango warning and get correct widths for the both cases.

[bjarthur]

thanks for this! sorry for the delay in merging. travis CI hung on julia 1.4 windows initially.

[ScottPJones]

@tomyun The length is 1 (it is one character). If you want the number of bytes, you need to use the sizeof function instead.

[ScottPJones]

This change is incorrect. The convert(String, text) should be left as-is.
Further down, the length needs to be changed to sizeof.

[tomyun]

I don't think this change is necessarily incorrect. pango_layout_set_markup() requires an Ptr{UInt8} array and now we explicitly provide one with codeunits(). I believe this is a canonical way as Julia documentation explains. And textarray is now an AbstractVector{UInt8} wrapper of String text, so we can still use length() on it to get a correct length of the byte array without complication.

[tomyun]

Maybe now I understand where your concern came from. If text was not a regular UTF-8 backed String, but something different like UTF16String/UTF32String from LegacyStrings.jl, code units would be no longer in UInt8 and their length shouldn't be used for counting byte array size. In such case, convert(String, text) is still needed to make sure we always have a UTF-8 String.

[bjarthur]

should probably either open a new issue or PR so we don't forget about this