The Garden & the Fire:
Engineering OpenType Nasta’liq

The problem

Alif joins only on its right side, so Nasta’liq text is full of unconnected alifs: at the start of words and in the middle of them. A Nasta’liq letter group is a cascade. The first letter starts high, and each following letter steps down until the last one rests on the baseline. When an alif comes before a group, a calligrapher does not drop it to the baseline. The alif hangs in the air, and its foot meets the point where the next stroke begins.

Gulzar’s engineering notes call this hard to express in OpenType Layout, and the released font does not attempt it. The obstacle is in the design of cursive attachment. A cursive lookup gives each glyph one entry anchor and one exit anchor, and every entry connects to every adjacent exit inside the same lookup. Gulzar’s main cursive lookup uses this to build the cascade: medial and final glyphs carry entry, initial and medial glyphs carry exit. If you now add an entry anchor to the initial letters, you do not get an alif connection. You get every medial and initial exit in the font connecting into initial letters, which is wrong everywhere except after an alif. The notes propose a contextual cursive lookup as a way out, marked untried.

The fix: scope by anchor name, not by context

The contextual lookup is not necessary. OpenType allows more than one cursive lookup, and each lookup only connects its own anchor pair. So Golshan adds a second cursive lookup with a pair that nothing else uses:

• cursive_exit exists only on the isolated alif glyphs, placed at the foot of the stroke.
• cursive_entry exists only on initial and isolated forms, placed at the pen-entry point of the glyph.

Inside this lookup, the only possible connection is alif to a following letter. The anchor names do the scoping. No context, no duplicated lookups, no interaction with the main cascade rules. In the FEZ source the whole feature is one line next to the main cursive routine:

Feature curs {
  Routine CursiveAttachment { Attach &entry &exit cursive; } IgnoreMarks RightToLeft;
  Routine AlifRaise { Attach &cursive_entry &cursive_exit cursive; } IgnoreMarks RightToLeft;
};

The RightToLeft flag matters. It tells the shaper to resolve the attachment so that the later glyph holds still and the earlier glyph moves. The word stays planted on the baseline, and the alif is the glyph that rises.

The alif rides the cascade

The best property of this fix comes free. HarfBuzz resolves cursive attachments from all lookups into one attachment tree per word. The alif attaches to the first letter of the group, and that letter already rides the group’s cascade from the main lookup. So the alif does not rise by a fixed amount. It rises to the entry height of the group, wherever the cascade puts it.

The shaper output shows the chain. In «امید» the initial mim rises 229 units and the alif lands at 266: the rise of the mim plus the height of its entry point. In «الرحمن» the lam rises 433 units with its cascade and the alif rises with it. One lookup, and the alif tracks any group the font can build.

Anchors from a script, not by hand

Gulzar has 341 initial and isolated glyph forms. Placing an anchor on each one by hand is a week of work, and every outline revision breaks it. Golshan computes the anchors at build time in scripts/add-alif-anchors.py, before the FEZ rules run:

• For each target glyph, the script gathers the outline points, with components resolved through their transforms. The pen-entry point is the highest point in the right part of the glyph (the rightmost 30% of its width).
• On the alif glyphs, cursive_exit goes at the foot of the stroke, with a small tuck constant so the alif overlaps the entry, as in the manuscripts.

The first version placed cursive_entry at the pen-entry height for every glyph, and it failed in an instructive way. For body-height letters (mim starts at 229, sin near 260, be near 300) the rule is right: the alif foot meets the start of the connecting stroke. But for tall-stemmed letters the pen enters at the top of the ascender, 450 to 670 units up. The alif climbed to the stem top, plus the cascade, and landed as high as 810 units in «الرحمن». On screen it read as a detached tick floating over the word. The word images looked broken.

The manuscripts show the real rule. Before a lam, the alif does not climb to where the pen enters. It stands beside the stem the way the lam-alif ligature stands: feet level at the stem base, tops level, and the cascade lifts the pair together. So the anchor script now has two regimes:

• Body letters (pen entry below 400 units): cursive_entry at the pen-entry height, capped at 300.
• Tall stems (pen entry at 400 or above): cursive_entry at the stem base, height zero. The alif rises by the group’s cascade alone, and «الرحمن» lands at 292 units instead of 810.

Isolated forms get entry anchors too, so an alif before a non-joining letter («اب», «او») also rises to that letter’s start height. Alif is excluded as a target, so an alif before another alif stays on the baseline.

What is not solved yet

The anchor heights come from a geometric heuristic tuned against one reference bismillah, not from a calligrapher’s review of many models. The two thresholds (400 for a tall stem, 300 for the entry cap) are first guesses. Gulzar’s automatic kerning also still assumes a baseline alif, so a raised alif can now stand further from its neighbor than it needs to. The next pass will review the heights against manuscript models, tune the tuck, and re-run the kerning and collision passes. The shaping test suite (make test-shaping) also needs a run to catch side effects in sequences we did not think about.

For the record, the full before and after test sheets: