Fix UnboundLocalError in _border and document rgba() color format

[veeceey]

Summary

• Fix bug in ImageOps._border(): When given a tuple with a length other than 2 or 4 (e.g. a 1-tuple or 3-tuple), the function raised an unhelpful UnboundLocalError because the left, top, right, bottom variables were never assigned. This now raises a clear ValueError with a descriptive message instead.

• Document rgba() color format in ImageColor: The rgba(red, green, blue, alpha) color string format has been supported in code and tested for a long time, but was missing from the ImageColor documentation. This adds it to the list of supported color string formats.

Test plan

• Added parametrized test test_expand_invalid_border that verifies ValueError is raised for 1-tuple, 3-tuple, and 5-tuple border values
• Verified existing tests still pass (ruff and black checks pass)
• Manually verified the fix resolves the UnboundLocalError

[radarhere]

Hi. These seem like two unrelated changes. In the future, it's worth considering whether these should be two separate PRs, or at least two separate commits.

I'm not sure how much of this you used AI for, but if you wouldn't mind explaining your human motivation a bit - did you actually find yourself triggering the UnboundLocalError unintentionally?

[veeceey]

Fair point on splitting, noted for next time.

On the motivation -- I was looking at the ImageOps.expand code and noticed the border validation raises ValueError for negative values, but if you pass something like a 3-tuple (which is not a valid border spec), the unpacking just silently fails and border stays partially unpacked, leaving bottom unbound. Then the subsequent code hits an UnboundLocalError, which is confusing because the real issue is the invalid input. So yes, I did trace through it manually -- the UnboundLocalError masks the actual problem.

Will apply the test simplification you suggested.

[veeceey]

Hi @radarhere, just checking in — I've applied your suggested test simplification and responded to your earlier questions. All CI checks are green. Would you be able to take another look when you get a chance? Thanks!

[radarhere]

#9433 (comment) might like us to consider our policy towards AI before merging

[radarhere]

Suggested change

	msg = "border must be an integer or a 2- or 4-tuple"
	msg = "border must be an integer, or a tuple of two or four elements"

Feel free to ignore this idea if you want, it just seems a bit more natural to me.