refactor: remove emoji-datasource from dependcies

[i1skn]

Description

emoji-datasource was only being used as a static source for emoji lookups and avatar picker data, but it adds unnecessary dependency, which results into additional 0.96Mb for bundle size.

This PR replaces that dependency with a compact in-repo emoji catalog (emojiByName + avatarEmojis) so emoji resolution and avatar selection keep the same behavior while cutting the large JSON dataset and removing the external package.

As a result of these changes the bundle size is decreased by 0.96Mb!

[i1skn]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• refactor: remove emoji-datasource from dependcies #7174 👈 (View in Graphite)
• refactor: add regression tests for emoji resolution paths #7173
• develop

This stack of pull requests is managed by Graphite. Learn more about stacking.

[github-actions]

🧪 Flashlight Performance Report (AWS Device Farm)

🔀 Commit: ade1831

📎 View Artifacts

Metric	Current	Δ vs Baseline
Time to Interactive (TTI)	5321 ms	🟢 -104.5 ms (-1.9%)
Average FPS	55.79	🔴 -1.1 (-2.0%)
Average RAM	388.7 MB	🟢 -6.3 MB (-1.6%)

[github-actions]

Launch in simulator or device for ade1831

[olerass]

Nice cleanup! Didn't fully review yet, but one thought I had was if we should not keep commiting the generated file to git? If we commit it:

• Zero friction for new setups onboarding (just yarn install), one less step to break in yarn setup
• CI and local builds become more predictable
• Code review is easier, can see exactly what gets shipped and what changes on package or script updates. With the script as source it's hard to review for correctness as we don't see the output
• It's not a large file anyway (anymore)

I know we do compile time generation of e.g networks.json, but this is a non-deterministic generation fetching data from a live api vs the emoji process which is a pure deterministic transform.

[olerass]

I think we should keep generated files out of features. Mixing generated files with actual src is confusing and makes it hard to judge what is actual source and what is not. Also gitignoring things deep inside src seems wrong (that goes away if we commit the file though which I rec, see my overall comment on the PR).

src/references is the canonical place for generated data today, keeping it there is more predictable IMO. Optimally, it should prob be named assets or similar, and the generated graphql should move there too, but that's for another time.

[olerass]

yarn generate:emojis also fail on my machine because the script uses cjs but constants.js uses ESM.

[olerass]

Sorry for the many comments, I didn't expect this to be a thorough review in my evening, but I started and coulnd't stop myself 🤣

I had Claude generate a comparison script between the current emoji data on develop and what the generation script produces. There's a pretty significant gap:

• 937 emoji names that exist today are missing in the new catalog
• Every avatar picker category has fewer emojis (e.g. smileys goes from 408 to 241, activities from 60 to 55)
• 18 currency flag lookups break

I think the root cause is that on develop, the avatar picker and the name lookup (resolveEmoji) use two completely separate data sources. The avatar picker imports emoji-datasource directly and uses the full array (just filters obsoleted_by, groups by category, sorts by sort_order). The name lookup uses the curated src/references/emojis.json which is a different dataset with different naming.

289 entries in emoji-datasource have name: null. From what I can tell these are "variation-selector" versions of emojis. I don't fully understand the Unicode side of this, but afaiu some emojis exist as two entries in the dataset; a base version and a version with an invisible character (FE0F) appended that hints to the OS "render this as an emoji glyph." The FE0F variants seem to be the the ones with null name.

This didn't matter before because the old avatar picker code used the raw array and never looked at the name field. This PR funeling both through a the name-keyed catalog causes the drops: no name = no key = not in the catalog = not in the picker. The name lookup side also loses coverage because emoji-datasource uses different names than the old curated file.

For example, the 5 missing Activities emojis are all name: null in the package: 🎖️🎗️ 🎟️ 🕹️ ⛸️ as seen on below screenshots.

Develop	PR