[varLib] Clarification regarding avar segment maps

The avar spec says:

Axis value maps can be provided for any axis, but are required only if the normalization mapping for an axis is being modified. If the segment map for a given axis has any value maps, then it must include at least three value maps: -1 to -1, 0 to 0, and 1 to 1. These value mappings are essential to the design of the variation mechanisms and are required even if no additional maps are specified for a given axis.

1. Currently, varLib._add_fvar_avar doesn’t enforce this. Should it?

2. When an axis doesn’t cover the full range from -1 to 1, it makes no sense to include the entries as described.

For example, take a wdth axis from Condensed to Normal. The mappings may be:

-1.0 to -1.0
-0.5 to -0.3
 0.0 to  0.0

What would an additional 1 to 1 mapping, as required by the spec, do? There is no master at 1.0 for this axis. If something has to be added to the segment map, it should be 1.0 to 0.0, making the range from 0 to 1 effectively flat. If that is the intent of the spec, the wording should be changed.

[PeterConstable]

Yes, varLib should enforce this to ensure fonts are valid. This is an old requirement, originating in GX:

“A segment must always have at least three correspondence pairs: minus one mapping to minus one, zero mapping to zero, and one mapping to one.”

Your example considers a width axis from Condensed to Normal, but those values are defined only in “user” space, not normalized space. The ‘avar’ table is a transformation entirely within normalized space which is always defined as having axes going from -1 to +1. An ‘fvar’ table may happen to define an axis in which default and min or default and max are the same, in which case nothing changes between -1 and 0, or between 0 and +1. But that is a consideration in the mapping from “user” space to normalized space that is independent of the ‘avar’ transformation that happens entirely within normalized space.

[jenskutilek]

Thanks all. #1014 fixes the issue about the default mappings. The rest was a misinterpretation on my part.