Check for nested components

[RosaWagner]

Observed behaviour

Component referring to another component (nested) are not exported correctly for ttf variable format. They take back the coordinates of the original glyph.

Expected behaviour

There should be a check to find any nested components.

Resources and exact process needed to replicate

I used Public Sans at commit uswds/public-sans@0a09945

But you can reproduce it with any font which have nested components (typically ohorn/uhorn with another accent)

[arrowtype]

This may be a more severe issue than just something messing up Samsa.

At arrowtype/recursive#412, I have described issues that occur when printing static TTFs with many printers. In summary: PostScript-based printers really mess up nested components in TTF fonts. This includes even high-end setups used for offset printing. One example of where things go wrong in current Recursive TTFs:

(There are more examples in the issue.)

There are two possible culprits:

• A problem in ttfautohint, or in how PostScript handles the hints from it
• A fundamental problem with PostScript’s handling of TTF nested components

Worse, printers that have PostScript installed apparently never update that software.

So, this check plus a script that un-nests components are probably relatively high-priority, if Google Fonts wants to give people any confidence in using the library for more than just web design. Not all of the alignment problems with TTFs & PostScript would be solved by this – it also messes up the positioning of glyph substituted via stylistic sets – but it seems that it would solve the majority of them.

[madig]

There's a ufo2ft filter to unnest composites automatically. Insert the following into your UFO's lib.plist:

    <key>com.github.googlei18n.ufo2ft.filters</key>
    <array>
      <dict>
        <key>name</key>
        <string>flattenComponents</string>
        <key>pre</key>
        <integer>1</integer>
      </dict>
    </array>

[arrowtype]

@madig THANK YOU, this is amazing news!

[madig]

It's actually been in there for decades, but we don't have time to write docs so all the filters are secret.

[anthrotype]

quite depressing that in 2020 we can't still have nested glyf components

[graphicore]

It's actually been in there for decades, but we don't have time to write docs so all the filters are secret.

I could be a good start to just drop some general lines about the existence of this mechanism and where to find the available filters and how to generally access and configure them.

[madig]

I started something at googlefonts/ufo2ft#286 last century but died before I could finish the job.

The stems of all .py files in https://github.com/googlefonts/ufo2ft/tree/master/Lib/ufo2ft/filters can be used as filters.

[simoncozens]

@madig: Would it be worth making FlattenComponents the default instead of DecomposeComponents?

[madig]

Where?

[simoncozens]

ufo2ft

[madig]

Where in ufo2ft? No idea what you're talking about.

[anthrotype]

He means as default filter in the ufo2ft.preProcessor. I would rather not do that. We can make it easier for users to specify that, if a custom filter in the lib.plist is too obscure. E.g. we can add a new option to ufo2ft.compileTTF function and even add a CLI flag to fontmake. But changing the default is not a good idea. There may be legitimate use-cases that may prefer to keep the current beaviour that allows for nested components.

[moyogo]

Should we add a --filters argument to fontmake instead or additionally?

[simoncozens]

Can’t we just have a single option which does this?

If we aren’t prepared to make creating non-broken fonts the default, we should at least make it as easy as possible.

[madig]

Technically, _the fonts_ aren't broken 😬

[moyogo]

Can’t we just have a single option which does this?

What I’m suggesting as an alternative or as an additional option is to have a --filter argument to fontmake that works like --feature-writer does.
So one could do:

fontmake -m DS.designspace --filter flattenComponents

This would also work for propagateAnchors, decomposeTransformedComponents, or any custom filter.

[arrowtype]

Note to others here (and maybe this should be included in the check rationale, etc):

At least for Recursive (and another project, Name Sans) it is essential to not only flatten components but to first also decompose transformed components. Otherwise, accents like the caron (which is a 180° rotated circumflex) get really misaligned.

In a lib.plist, that looks like this:

    <key>com.github.googlei18n.ufo2ft.filters</key>
    <array>
      <dict>
        <key>name</key>
        <string>decomposeTransformedComponents</string>
        <key>pre</key>
        <integer>1</integer>
      </dict>
      <dict>
        <key>name</key>
        <string>flattenComponents</string>
        <key>pre</key>
        <integer>1</integer>
      </dict>
    </array>

See arrowtype/recursive#412 for details.