fix: prevent arbitrary font-family and font-weight from merging

[roneymoon]

What does this change do?

This PR fixes an issue where arbitrary font-family and font-weight utilities were incorrectly merged by tailwind-merge, causing one of them to be dropped when both were present.

The problem occurred because the font-weight class group was using an overly broad arbitrary validator, which also matched font-family values (e.g. family-name:). As a result, both utilities were classified into the same group and one would override the other.

This change makes the font-weight matching more precise so that each utility is correctly classified and both can safely coexist.

---

What was the issue?

Given the following input:

twMerge('font-(family-name:--custom-family) font-(--custom-weight)')

Before

'font-(--custom-weight)'

After

'font-(family-name:--custom-family) font-(--custom-weight)'

The same incorrect behavior happened regardless of class order.

---

How is this fixed?

• Introduced weight-specific arbitrary validators that only match:

• unlabeled variables intended for font weight (e.g. (--custom-weight))

• explicitly labeled weight values (e.g. (weight:...), (number:...))

• Replaced the generic isArbitraryVariable usage in the font-weight class group with these more specific validators

• Left font-family logic unchanged so it continues to correctly handle family-name: and other family-related arbitrary values

• This avoids the over-matching that caused both utilities to be treated as font-weight.

---

Why this approach?

• Keeps the fix minimal and localized

• Avoids changing validator execution order or introducing priorities

• Does not modify configuration or public APIs

• Preserves existing behavior while fixing the incorrect merge

---

Thanks for taking a look!

Closes #588

[codspeed-hq]

Merging this PR will not alter performance

✅ 7 untouched benchmarks

---

_{Comparing roneymoon:font-validator-conflict (b232b67) with main (95ce92f)}

[dcastil]

Hey @roneymoon! 👋

I expected a fix to be much more involved, awesome that you found a much simpler solution! Thanks for fixing the issue!

I fixed some additional minor nuances and added some tests and docs for the changes.

[github-actions]

This was addressed in release v3.4.1.