x/image/font/opentype: font metrics discrepancies wrt golang/freetype/truetype

[sbinet]

hi,

it seems parsing the same TTF files (Liberation fonts, for example) leads to different font metrics values.
I'd like to understand whether that's coming from a bug/oversight from x/image/font/sfnt or from freetype/truetype.

What version of Go are you using (go version)?

$ go version
go version devel +732ea4c2dc Thu Mar 18 08:25:24 2021 +0000 linux/amd64

Does this issue reproduce with the latest release?

yes.

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GO111MODULE=""
GOARCH="amd64"
GOBIN=""
GOCACHE="/home/binet/.cache/go-build"
GOENV="/home/binet/.config/go/env"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOINSECURE=""
GOMODCACHE="/home/binet/dev/go/gocode/pkg/mod"
GONOPROXY=""
GONOSUMDB=""
GOOS="linux"
GOPATH="/home/binet/dev/go/gocode"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/home/binet/sdk/go"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/home/binet/sdk/go/pkg/tool/linux_amd64"
GOVCS=""
GOVERSION="devel +732ea4c2dc Thu Mar 18 08:25:24 2021 +0000"
GCCGO="gccgo"
AR="ar"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD="/dev/null"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build1145417185=/tmp/go-build -gno-record-gcc-switches"

What did you do?

https://play.golang.org/p/BM_FTuuxrxF

What did you expect to see?

reasonnable agreement between font metrics extracted from a TTF file via x/image and via freetype/truetype.

• freetype/truetype doesn't extract XHeight, CapHeight nor CaretSlope
• Descent is defined negative in one case, positive in the other
  but apart from those known differences, (and the glyph bounds), nothing really matches.

What did you see instead?

$> go run main.go
fnter: truetype: units-per-em=2048
fnter: opentype: units-per-em=2048
fnter: truetype: ext=main.Extents{Ascent:9.7998046875, Descent:-3.0322265625, Height:12.83203125}
fnter: opentype: ext=main.Extents{Ascent:9.052734375, Descent:2.119140625, Height:11.4990234375}
fnter: truetype: bnds=fixed.Rectangle26_6{Min:fixed.Point26_6{X:-1114, Y:-621}, Max:fixed.Point26_6{X:2666, Y:2007}}
fnter: opentype: bnds=fixed.Rectangle26_6{Min:fixed.Point26_6{X:-1114, Y:-2007}, Max:fixed.Point26_6{X:2666, Y:621}}
fnter: truetype: met=font.Metrics{Height:640, Ascent:580, Descent:136, XHeight:0, CapHeight:0, CaretSlope:image.Point{X:0, Y:0}}
fnter: opentype: met=font.Metrics{Height:736, Ascent:579, Descent:136, XHeight:338, CapHeight:440, CaretSlope:image.Point{X:0, Y:1}}
fnter: delta(t-o):   font.Metrics{Height:-96, Ascent:1, Descent:0, XHeight:-338, CapHeight:-440, CaretSlope:image.Point{X:0, Y:-1}}
fnter: opentype: met=font.Metrics{Height:2355, Ascent:1854, Descent:434, XHeight:1082, CapHeight:1409, CaretSlope:image.Point{X:0, Y:1}}
fnter: truetype: line-gap=196
fnter: opentype: line-gap=21

[cagedmantis]

/cc @nigeltao

[sbinet]

gentle ping?

[nigeltao]

Sorry, I don't have a lot of free time right now to investigate.

Do you know what the C FreeType library reports? (Ideally with a small reproducible C program).

[tinne26]

I looked a bit into this, and I can share the following:

• Extents: the comparison is not valid. The code used to compute them for freetype is taking the font bounds, which are the union of the font glyph bounds, while the opentype version uses the font metrics, which are derived from the font headers.
• Bounds: these are actually matching, but since descents are negative in freetype, the Y axis is shifted between freetype and opentype, but the height is the same. Opentype here is matching font.Face.GlyphBounds() behavior (The glyph's ascent and descent are equal to -bounds.Min.Y and +bounds.Max.Y), which seems like the right thing for opentype to do, and it makes freetype behavior slightly strange, as it's also aware of the font interfaces but does not operate consistently with them.
• Metrics: the Height value in freetype is simply wrong, as far as I can tell. It uses the scale right away (see truetype/face.go#L258), while opentype derives Height from the proper HHead Ascent/Descent/LineGap values. This means that line gap calculations in freetype are also meaningless.

Despite all this, I still have some concerns:

• Extents: (edit: nevermind, seems like bounds height seems to exceed metrics height in almost all fonts I tried, or at most match it) freetype is reporting the bounding box that encloses all glyphs from the font, but that's bigger than the box reported by opentype, which is based on the sizes that the font defines in its headers. This means that some glyph (maybe some with carets or stuff above and below) exceeds the advertised font height itself, which seems ugly. Either that or the cbox is different from the bbox, but the difference seems too extreme for that to be the case. Someone who knows more about typography should let us know if any of these are common or there might still be an underlying issue here.
• Metrics: the ascent is still slightly different. I'd assume some conversion or floating point issue, but I couldn't really pinpoint it.