threads / discuss / 66794

[Outreachy] column: propose using early_scan_options() for --command

Subject: [Outreachy] column: propose using early_scan_options() for --command

## tl;dr

One message between Nov 6, 2026 and Nov 6, 2026.

replies: 0people: 1as markdown or json

Fien Dora· Nov 6, 2026, 17:26 UTC · lore
Hello Christian, Siddharth, and the Git community,

Following my earlier contribution, t1402: test --no-allow-onelevel option precedence, I have been exploring possible follow-up work related to Git's argument-scanning and parsing improvements.

I have been studying the new early_scan_options() API, particularly its implementation in parse-options.c and its use in fast-import.

I noticed that builtin/column.c still performs a manual early check for --command= before invoking parse_options():

if (argc > 1 && starts_with(argv[1], "--command="))

This is necessary because git column uses the option's value to load configuration before normal option parsing. However, the early check recognizes only the exact --command=<name> form, whereas parse_options() also recognizes separate-value and unambiguous abbreviated forms.

I tested the current implementation and observed that •⁠ ⁠--command=branch is accepted. •⁠ ⁠--command branch and --comm=branch are rejected. •⁠ ⁠--command=branch is rejected when preceded by another option. •⁠ ⁠Repeated identical --command values are accepted, while conflicting values are rejected. I would like to propose a focused follow-up patch that replaces the manual early check with early_scan_options(), mark --command with PARSE_OPT_EARLY, and add regression tests for the relevant argument-handling behavior.

I would appreciate guidance on whether the existing first-position and attached-value restrictions should be preserved or whether the conversion should accept additional forms recognized by parse_options().

I also noticed that Cynthia Uche mentioned git column as a possible conversion target in her internship proposal. I would like to acknowledge her proposal to prevent any potential overlap before starting any implementation.

Would this be a suitable independent follow-up contribution to undertake now, or would you recommend a different target?

Thank you for your time and guidance.

← back to recent threads