EkType · revurpk · May 24, 2026
diff --git a/BalooTammudu2-Telugu/vedic-build/.gitignore b/BalooTammudu2-Telugu/vedic-build/.gitignore
@@ -0,0 +1,10 @@
+# Donor font tree — unpacked Noto Sans Devanagari release (~45MB)
+donor/
+
+# Intermediate TTFs compiled from ../TTX as graft input
+intermediate/
+
+# Ephemeral artifacts from the GitHub issue-draft step
+make_issue_url.py
+issue-body.md
+issue-prefilled-url.txt
diff --git a/BalooTammudu2-Telugu/vedic-build/README.md b/BalooTammudu2-Telugu/vedic-build/README.md
@@ -0,0 +1,177 @@
+# Vedic Extensions support for Baloo Tammudu 2 (Telugu)
+
+Prototype build that adds rendering support for the Unicode **Vedic
+Extensions** block (U+1CD0–U+1CFF) plus the four **Devanagari-block Vedic
+accents** (U+0951–U+0954) to all five weights of Baloo Tammudu 2.
+
+This is a **non-authoritative prototype** built on top of the released TTX
+artefacts in [`../../TTX/`](../../TTX/). The font's `.vfb` sources in
+[`../Regular/VFB/`](../Regular/VFB/) etc. are untouched — for permanent
+inclusion the outlines need to land there. A feature request tracking the
+broader work has been opened against the upstream
+[EkType/Baloo2](https://github.com/EkType/Baloo2/issues) repo.
+
+## What this produces
+
+Five patched TTFs (one per weight) with **45 codepoints** of Vedic mark
+support:
+
+| Block | Range | Count |
+|---|---|---|
+| Devanagari Vedic accents | U+0951–U+0954 | 4 |
+| Vedic Extensions | U+1CD0–U+1CFF | 41 |
+
+Each grafted codepoint is:
+
+- Mapped in `cmap` (Windows BMP subtable).
+- Classified as a combining mark (GDEF class 3) with zero advance width.
+- Positioned over Telugu consonants via a new `abvm` GPOS lookup with
+  mark-to-base anchors on 74 Telugu base glyphs.
+
+The seven codepoints in U+1CD0–U+1CFF that aren't covered are either too
+recent (U+1CF7 from Unicode 12, U+1CFA from Unicode 13) or unassigned
+(U+1CFB–U+1CFF). The donor font doesn't have them either.
+
+## Pipeline
+
+```
+TTX/BalooTammudu2-{Weight}.ttx
+  │
+  │  fontTools.ttx
+  ▼
+intermediate/BalooTammudu2-{Weight}.ttf
+  │
+  │  graft_vedic.py
+  ▼  (donor: NotoSansDevanagari-{Weight}.ttf, OFL)
+BalooTammudu2-{Weight}.ttf  (cmap + glyf + GDEF wired)
+  │
+  │  add_vedic_gpos.py
+  ▼
+BalooTammudu2-{Weight}.ttf  (+ abvm MarkBasePos lookup on tel2/telu)
+```
+
+## Scripts in this directory
+
+| File | Purpose |
+|---|---|
+| `graft_vedic.py` | Copies Vedic glyph outlines from a donor TTF into a recipient TTF. Flattens composites, weight-matches, adapts glyph geometry: above-marks have their bbox-bottom aligned to Y=600 (Baloo's existing mark band); all marks have their bbox-center shifted to X=-300 to land over a typical Telugu base. Adds cmap entries, GDEF mark classification, sets zero advance. |
+| `add_vedic_gpos.py` | Adds a `MarkBasePos` (GPOS lookup type 4) connecting Vedic above-marks to Telugu consonants. Base anchors at consonant `(xmid, ymax + 220)`, mark anchors at `(xmid, ymin)`. Above-base vowel signs (matras U+0C3E–U+0C56) and halant are excluded from the base coverage so the mark attaches to the consonant directly — the matra is allowed to overlap visually but the mark sits over the consonant column. Registers a new `abvm` feature on the `tel2` and `telu` scripts. |
+
+## Donor font
+
+[Noto Sans Devanagari v2.006](https://github.com/notofonts/devanagari) (SIL
+Open Font License). Chosen because it has 41/48 of the assigned Vedic
+Extensions codepoints and weight-matches Baloo Tammudu 2 (Regular through
+ExtraBold). Outlines are imported as-is and then transformed; no other
+changes to the donor are carried over.
+
+The attribution is recorded in the top-level
+[`OFL.txt`](../../OFL.txt) as required by OFL §4.
+
+## How to rebuild
+
+From the repo root:
+
+```powershell
+# 1. Compile Baloo TTX -> TTF (5 weights, ~12s total)
+$weights = 'Regular','Medium','SemiBold','Bold','ExtraBold'
+foreach ($w in $weights) {
+  python -m fontTools.ttx -q `
+    -o "BalooTammudu2-Telugu/vedic-build/intermediate/BalooTammudu2-$w.ttf" `
+    "TTX/BalooTammudu2-$w.ttx"
+}
+
+# 2. Download donor zip (one-time)
+[Net.ServicePointManager]::SecurityProtocol = [Net.SecurityProtocolType]::Tls12
+$assets = (Invoke-WebRequest 'https://api.github.com/repos/notofonts/devanagari/releases/latest' `
+  -UseBasicParsing -Headers @{'User-Agent'='curl'}).Content | ConvertFrom-Json
+Invoke-WebRequest $assets.assets[0].browser_download_url `
+  -OutFile 'BalooTammudu2-Telugu/vedic-build/donor/NotoSansDevanagari.zip'
+Expand-Archive -Force `
+  'BalooTammudu2-Telugu/vedic-build/donor/NotoSansDevanagari.zip' `
+  'BalooTammudu2-Telugu/vedic-build/donor/'
+
+# 3. Graft outlines + add GPOS anchors for each weight
+foreach ($w in $weights) {
+  python BalooTammudu2-Telugu/vedic-build/graft_vedic.py `
+    --recipient "BalooTammudu2-Telugu/vedic-build/intermediate/BalooTammudu2-$w.ttf" `
+    --donor     "BalooTammudu2-Telugu/vedic-build/donor/NotoSansDevanagari/full/ttf/NotoSansDevanagari-$w.ttf" `
+    --output    "BalooTammudu2-Telugu/vedic-build/BalooTammudu2-$w.ttf"
+  python BalooTammudu2-Telugu/vedic-build/add_vedic_gpos.py `
+    --input  "BalooTammudu2-Telugu/vedic-build/BalooTammudu2-$w.ttf" `
+    --output "BalooTammudu2-Telugu/vedic-build/BalooTammudu2-$w.ttf"
+}
+```
+
+Requires `fontTools` (`pip install fontTools`).
+
+## Known limitations
+
+These are inherent to a prototype that grafts outlines without involving
+the FontLab `.vfb` sources or the original designers.
+
+- **Style mismatch.** Noto's stroke contrast and weight don't match Baloo's
+  heavy, spurless display character. Vedic marks will look thinner and
+  out-of-family next to native Telugu glyphs in the same line.
+- **Approximate vertical clearance.** The static Y-gap of 220 units above
+  each consonant works for most clusters, but tall above-base matras like
+  `matraIi` (Y up to 1051) come within ~80 units of the marks. Tweak
+  `BASE_ANCHOR_Y_GAP` in `add_vedic_gpos.py` if more clearance is wanted.
+- **No anchors for below-marks.** Only the `abvm` (above-base) lookup was
+  added. U+0952 anudatta and the few below-marks in Vedic Extensions still
+  rely on the static negative-X-baked positioning from the graft step.
+- **Coverage gaps.** Seven Vedic Extensions codepoints (U+1CF7, U+1CFA,
+  U+1CFB–U+1CFF) aren't in the donor and remain unrendered.
+- **Sources still untouched.** Patches live in this directory only. The
+  `.vfb` files need updates from a designer for permanent inclusion.
+
+## TODO — what a proper fix requires
+
+This prototype is a working approximation, not the real fix. For
+upstreamable Vedic support in Baloo Tammudu 2 the following needs to
+happen in FontLab on the `.vfb` sources, then propagate through the
+existing build chain (VFB → AFDKO → TTF/TTX):
+
+- [ ] **Draw native Vedic glyphs** for U+0951–U+0954 and U+1CD0–U+1CFF
+      in all five weights, matched to Baloo's heavy spurless display
+      style (stroke contrast, weight, x-height). This is the blocker;
+      everything else depends on it.
+- [ ] **Cover the seven gaps** the donor doesn't have: U+1CF7 (Unicode
+      12), U+1CFA (Unicode 13). U+1CFB–U+1CFF are unassigned in Unicode
+      and should remain unmapped.
+- [ ] **Add entries to** `BalooTammudu2-Telugu/GlyphOrderAndAliasDB`
+      mapping each new glyph name (`uni0951` … `uni1CFA`) to its
+      codepoint, following the existing naming convention.
+- [ ] **Classify in** `BalooTammudu2-Telugu/GDEF` as combining marks
+      (class 3) with zero advance width.
+- [ ] **Real GPOS anchor design.** Replace the `add_vedic_gpos.py`
+      heuristic (consonants at `ymax + 220`, marks at `ymin`) with
+      anchors drawn intentionally in FontLab for each base and mark.
+      Add anchors for **below-marks too** (U+0952 anudatta etc.) — the
+      prototype only handles above-marks.
+- [ ] **Handle the above-base matras properly.** The prototype excludes
+      them from base coverage so the mark attaches to the consonant
+      instead. The right solution is either (a) `mkmk` anchors so the
+      Vedic mark attaches to the matra's top, or (b) contextual GPOS
+      that raises the mark when a matra is present in the cluster.
+- [ ] **Register `abvm` and `blwm` features** on `tel2`/`telu` script
+      records pointing at the new lookups, following Baloo's existing
+      feature-table structure rather than the runtime patch this
+      prototype does.
+- [ ] **Visual QA** across a Vedic test corpus (e.g. svaras over
+      common Telugu consonants, conjuncts, and aksharas with above-base
+      matras). The static `BASE_ANCHOR_Y_GAP = 220` prototype value is
+      a compromise and needs designer review.
+- [ ] **Remove this `vedic-build/` directory** once the proper sources
+      ship — it's a temporary workspace, not a permanent build product.
+
+Tracking issue: <https://github.com/EkType/Baloo2/issues> (filed as
+"Add rendering support for Unicode Vedic Extensions (U+1CD0–U+1CFF)").
+
+## Files ignored by `.gitignore`
+
+`donor/` (~45MB Noto release tree), `intermediate/` (compiled TTX), the
+ephemeral issue-draft files (`issue-body.md`, `issue-prefilled-url.txt`,
+`make_issue_url.py`), and `*.ttf` (handled by the repo-wide
+`.gitignore`). The 5 patched output TTFs are therefore not tracked —
+regenerate them with the pipeline above when needed.