Sanity check: STAT names must match Designspace instance style names

[madig]

A single-axis VF must have entries that match instance style names. A multi-axis VF must have entries that when concatenated in the specified order must match instance style names.

[twardoch]

That should not be extensive ie. if I have a 6-axis font, I may want to define 6 named axis instances on each axis in STAT, but I may prefer not to define 46,656 (or whatever) fvar instances ;)

[twardoch]

In other words there may be STAT combinations that don't correspond to fvar instances

[twardoch]

I’d leave out the multiaxis check. There may be cases where I create fvar instance names concatenating style records in a particular order for legacy reasons but I may prefer to order the fvar axes differently

[twardoch]

But the single-axis check is reasonable

[twardoch]

Does your code reuse name table entries? Single-axis would bebefit from such a treatment that you check if the STAT name records match existing instance names, if they do, you reuse the name entry, if they don't, you warn

[twardoch]

Though a reverse check, i.e. if all fvar instance names can be expressed as a space concatenation of STAT axis instances in the axis order — that'd be useful

[twardoch]

with the special case "Regular", I guess

[madig]

Does your code reuse name table entries

Not yet: #17.

Though a reverse check, i.e. if all fvar instance names can be expressed as a space concatenation of STAT axis instances in the axis order — that'd be useful

That's what I was thinking. Though that then messes with "concatenating style records in a particular order for legacy reasons"? Hrm.

[twardoch]

Well, that's a special "I know what I'm doing" case, so a warning is still fine

[kontur]

Shouldn't statmake support writing STAT entries that are not defined as variable instances? E.g. the latest GF suggestion is to have only a subset of instances, but a full STAT table.

[madig]

Maybe. It currently explicitly prunes STAT entries that aren't present in the fvar. At least IIRC, haven't looked at the code in a while.