[Address Bar] Backspace re-triggers inline auto-complete sequentially instead of clearing it

[sojupie]

Summary

When typing a term in the address bar that matches multiple internal index sources (e.g., open tabs, bookmarks, or history entries), backspacing to clear the inline completion fails to persist. Instead of leaving the raw user text, deleting the auto-filled suffix causes the address bar to sequentially re-trigger auto-fill matches from remaining sources.

Steps to reproduce

1. Ensure a domain with repeating string tokens exists in your open tabs or bookmarks (e.g., pangrampangram.com).
2. Open a new tab and focus the address bar.
3. Type the initial prefix (e.g., pangram).
4. Observe the address bar auto-completing the rest of the URL (pangram.com).
5. Press the backspace key (🔙) once to clear the suggested inline completion.

Expected behavior

Pressing backspace once removes the inline auto-filled completion, leaving only the typed text (pangram), and suspends further auto-fill suggestions for the remainder of the active entry session.

Actual behavior

Backspace removes the current inline match, but the address bar immediately re-evaluates the typed input against secondary data sources (open tab -> bookmark -> history) and appends a new auto-fill suffix. The user must hit backspace 3–4 consecutive times to clear each layer before being able to submit a plain text search.

Device & build information

• Device: iPhone 13 Pro Max, iOS 27 public beta
• Build version: Firefox 153.2 (74961)
• First seen version: Not new to this version, has been an issue for a while but i can't say exactly how long, unfortunately.
• Keyboards Tested: Apple Native, Gboard

Notes

Attachments:
https://github.com/user-attachments/assets/73c4d612-fdc5-43f6-a325-10de187c9248

┆Issue is synchronized with this Jira Task

[EhsanAzish80]

I can reproduce and investigate this issue. I plan to trace the inline-completion deletion state in LocationTextField and SearchLoader, update the existing domain-autocomplete deletion coverage, and verify the result on an iOS simulator.

Would you be open to an external contribution for this fix? I’ll wait for confirmation before starting, in accordance with the repository’s contribution guidelines.