Measuring width string

Hi Alex,

This is indeed a very useful information. You can also put it on StackOverflow, then it will be Googlable.

Cheers.

Uko

On 27 Oct 2015, at 12:02, Aliaksei Syrel <[hidden email]> wrote:

Hi

I'm sending it here so that it doesn't get lost.

There are multiple ways to measure string width. In the following examples performance will be tested measuring 10`000 times the width of the following string:

string := 'abcdefghijklmnopqrstuvwxyz 0123456789!@#$%^&*()_+'.

1) The most straightforward way is to send #widthOfString: to the font. Time to run gives around 250ms.

http://ws.stfx.eu/2Q5YA9DFTRDR

Resulting value is rounded to integer and seems to be not absolutely correct and precise.

2) More complex way is to go deeper on the level of glyphs and manually summarise the width of each character in the string sending #getGlyphWidth: character to the font. Time to run gives around 750ms.

http://ws.stfx.eu/ETBEW1EHAAZ8

Resulting value is float and looks like correct and precise value.

3) Even more complex is to use CairoFontMetricsProvider instead of font's methods. The same as in 2nd case we measure each character. Time to run around 350ms.

http://ws.stfx.eu/7I89DMD0ZLM3

Resulting value is exactly the same as in the 2nd case. With almost equal performance to 1st it is nice alternative.

4) One more way is to let native cairo to calculate everything for us. Calls happen through nativeboost. Time to run around 120ms.

http://ws.stfx.eu/HYD76OMIOM7L 

As result it returns CairoTextExtents which allows to calculate width and height with one call.

Cheers,

Alex

Note that you shouldn't use 2) or 3) for non-monospace fonts because of kerning.

font := StandardFonts defaultFont.

font widthOfString: 'eT'. "16"

font widthOfString: 'Te'. "15"

'eT' inject: 0 into: [ :width :c | width + (font getGlyphWidth: c) ]. "13.416015625"

'Te' inject: 0 into: [ :width :c | width + (font getGlyphWidth: c) ]. "13.416015625"

4) seems fine

font := StandardFonts defaultFont.

provider := CairoScaledFont fromFreetypeFont: font realFont.

width := [ :string |

converter := CairoUTF8Converter new.

converter convert: string from: 1 to: string size.

stringExtents := CairoTextExtents new.

provider getExtentsOf: converter buffer into: stringExtents.

stringExtents width ].

width value: 'eT'. "12.463638305664063"

width value: 'Te'. "12.561752319335938"

On Tue, Oct 27, 2015 at 12:27 PM, Yuriy Tymchuk <[hidden email]> wrote:

Hi Alex,

This is indeed a very useful information. You can also put it on StackOverflow, then it will be Googlable.

Cheers.

Uko

On 27 Oct 2015, at 12:02, Aliaksei Syrel <[hidden email]> wrote:

Hi

I'm sending it here so that it doesn't get lost.

There are multiple ways to measure string width. In the following examples performance will be tested measuring 10`000 times the width of the following string:

string := 'abcdefghijklmnopqrstuvwxyz 0123456789!@#$%^&*()_+'.

1) The most straightforward way is to send #widthOfString: to the font. Time to run gives around 250ms.

http://ws.stfx.eu/2Q5YA9DFTRDR

Resulting value is rounded to integer and seems to be not absolutely correct and precise.

2) More complex way is to go deeper on the level of glyphs and manually summarise the width of each character in the string sending #getGlyphWidth: character to the font. Time to run gives around 750ms.

http://ws.stfx.eu/ETBEW1EHAAZ8

Resulting value is float and looks like correct and precise value.

3) Even more complex is to use CairoFontMetricsProvider instead of font's methods. The same as in 2nd case we measure each character. Time to run around 350ms.

http://ws.stfx.eu/7I89DMD0ZLM3

Resulting value is exactly the same as in the 2nd case. With almost equal performance to 1st it is nice alternative.

4) One more way is to let native cairo to calculate everything for us. Calls happen through nativeboost. Time to run around 120ms.

http://ws.stfx.eu/HYD76OMIOM7L 

As result it returns CairoTextExtents which allows to calculate width and height with one call.

Cheers,

Alex

Hi Alex,

thanks, this is very usefull. Can you tell which one of these measures is the closest to the effective length of a string morph? Number 4?

Thierry

2015-10-27 12:02 GMT+01:00 Aliaksei Syrel <[hidden email]>:

Hi

I'm sending it here so that it doesn't get lost.

There are multiple ways to measure string width. In the following examples performance will be tested measuring 10`000 times the width of the following string:

string := 'abcdefghijklmnopqrstuvwxyz 0123456789!@#$%^&*()_+'.

1) The most straightforward way is to send #widthOfString: to the font. Time to run gives around 250ms.

http://ws.stfx.eu/2Q5YA9DFTRDR

Resulting value is rounded to integer and seems to be not absolutely correct and precise.

2) More complex way is to go deeper on the level of glyphs and manually summarise the width of each character in the string sending #getGlyphWidth: character to the font. Time to run gives around 750ms.

http://ws.stfx.eu/ETBEW1EHAAZ8

Resulting value is float and looks like correct and precise value.

3) Even more complex is to use CairoFontMetricsProvider instead of font's methods. The same as in 2nd case we measure each character. Time to run around 350ms.

http://ws.stfx.eu/7I89DMD0ZLM3

Resulting value is exactly the same as in the 2nd case. With almost equal performance to 1st it is nice alternative.

4) One more way is to let native cairo to calculate everything for us. Calls happen through nativeboost. Time to run around 120ms.

http://ws.stfx.eu/HYD76OMIOM7L 

As result it returns CairoTextExtents which allows to calculate width and height with one call.

Cheers,

Alex

StandardFonts defaultFont

 

"a LogicalFont

 familyName: Source Sans Pro

 emphasis: nil

 pointSize: 10

 realFont: FreeTypeFont('Source Sans Pro' 'Regular' 10) weight: 400 stretch: 5 slant: 0" 

StandardFonts defaultFont realFont

 

"FreeTypeFont('Source Sans Pro' 'Regular' 10)" 

StandardFonts defaultFont realFont face hasKerning

"false"

Free type source sans pro in Pharo just does not support kerning.

Ok, answer is quite simple.

Current string rendering on athens does not support kerning. Because it is not cairo's job. Kerning should be took into account after creating glyphs with cairo but before extent measurement and obviously rendering. It can be done with FreeType.

I made small refactoring of text measurement and added kerning support. 

So, in the latest version of Athens-Cairo you can get extents easily:

extents := CairoFontMetricsProvider new
font: StandardFonts defaultFont;
extentsOf: 'Ta ffl'.

extent is an instance of CairoTextExtents.

Rendering on athens with Times New Roman size 200 gives: (Mac OSX Yosemite)

And here is what I get in Keynote with the same font

Close enough :D

Cheers,

Alex

On Wed, Nov 4, 2015 at 2:04 PM, Aliaksei Syrel <[hidden email]> wrote:

However, Helvetica Neue on mac has kerning.

Will check

Cheers,

Alex

On Wed, Nov 4, 2015 at 1:59 PM, Aliaksei Syrel <[hidden email]> wrote:

StandardFonts defaultFont

 

"a LogicalFont

 familyName: Source Sans Pro

 emphasis: nil

 pointSize: 10

 realFont: FreeTypeFont('Source Sans Pro' 'Regular' 10) weight: 400 stretch: 5 slant: 0" 

StandardFonts defaultFont realFont

 

"FreeTypeFont('Source Sans Pro' 'Regular' 10)" 

StandardFonts defaultFont realFont face hasKerning

"false"

Free type source sans pro in Pharo just does not support kerning.