fix(glab): tokenize glab args instead of walking them by hand

[KuSh]

The bug

glab_cmd.rs's extract_identifier_and_extra_args picked the MR/issue identifier with a hand-rolled walker: a hardcoded eight-entry list of value-taking flags (-R, --repo, -g, --group, -F, --output, -m, --message), a skip_next bool, and arg.starts_with('-'). Any value-taking flag missing from that list had its value hoisted out and re-emitted as the identifier. The list also mixed flags from unrelated subcommands — -m/--message belongs to mr merge/mr note, not to mr view, while mr view's own -p, -P and --jq were absent.

run() additionally had no -- awareness, and rtk's own -R/-g were appended after the user's arguments.

Reproducers

Verified against a glab stub that records its argv (tests/glab_argv_test.rs), flags transcribed from glab 1.36.0 and 1.117.0 --help, and every expectation checked against real glab 1.117.0 (registry.gitlab.com/gitlab-org/cli:latest, LC_ALL=C):

typed	glab received (develop)	glab receives (this PR)
rtk glab mr view --page 2	mr view 2 -F json --page	mr view -F json --page 2
rtk glab issue view -P 50	issue view 50 -F json -P	issue view -F json -P 50
rtk glab mr view --jq .iid	mr view .iid -F json --jq	mr view --jq .iid
rtk glab mr view --page 2 42	mr view 2 -F json --page 42	mr view 42 -F json --page 2
rtk glab -R o/r mr diff -- 42	mr diff -- 42 -R o/r	mr diff -R o/r -- 42
rtk glab mr view -- --web 42	mr view 42 -- --web	mr view -- --web 42
rtk glab mr view -- --web	mr view -- --web	mr view -- --web

What changed

• The walker and its flags_with_value array are gone. split_identifier runs arg_tokenizer::tokenize_grammar and takes the first free positional (Token::is_free_positional).
• One takes_value table per glab subcommand, each transcribed from that subcommand's own --help and taken as the union over 1.36.0 and 1.117.0. mr merge, mr approve, mr note and mr update do not share one: -s is the boolean --squash on merge but the value-taking --sha on approve, -d is --remove-source-branch on merge but --description on update, and -m is --message on merge but --milestone on update. mr view and issue view do share a table — their value-taking flag sets are identical on both versions.
• Value-taking flags use .claiming_dash_dash(): measured on 1.117.0, glab mr view --page -- 5 reports Invalid argument "--" for "-p, --page" flag, so pflag reads -- as the flag's value, not as the boundary.
• run() restores the -- clap's trailing_var_arg consumes, over the reassembled [subcommand] + args region rather than args alone; over args alone it lands one token off and duplicates the subcommand. trailing_var_arg only ever lets clap strip one --, the one at the head of that region — before the subcommand, or right after it, since -R/-g are parsed before the subcommand positional either way. That one is rtk's own option terminator and glab never saw it: glab -- mr view 42 prints root help on 1.117.0, glab mr -- view 42 prints the mr help instead of dispatching to view, and glab api -- projects/1 --paginate is Accepts 1 arg(s), received 2. So that one -- is dropped and every later one reaches glab verbatim.
• glab keeps reading positionals past -- (glab mr view -- 42 views MR 42), so the identifier search reaches past the boundary — but only where unescaping is a no-op: one token, still read as a positional without the --. Re-emitting the identifier in front of the boundary unescapes it along with whatever trailed it, and rtk glab mr view -- --web or -- --web 42 would then run glab with --web live, opening a browser the user had asked glab to treat as a positional. Anything else glab rejects on arity whatever rtk sends (Accepts at most 1 arg(s), received 2), so it goes to glab untouched.
• rtk's -R/-g now go in at arg_tokenizer::injection_point instead of being appended. Position is unchanged when there is no --.
• has_output_flag now counts --jq. With the value correctly attached, glab returns the user's own projection, and reformatting it as an MR printed a summary of fields the projection does not have (? MR !0: ???).

Tests

tests/glab_argv_test.rs asserts on the child argv, and on rtk's stdout for the --jq case; each of the 13 tests was confirmed failing before the corresponding change and passing after. The run is isolated with its own HOME and RTK_DB_PATH, which it otherwise read and wrote for real. Sixteen unit tests in glab_cmd.rs cover the per-subcommand tables, including the -s merge/approve split.

Not in scope

has_output_flag and should_passthrough_view still scan with a == "--output", so attached spellings (--jq=.iid) and short clusters miss. That is the separate args.iter().any(...) bug class, not the identifier walker; left for a follow-up.

[rtk-wshm-sync-bot]

wshm · Automated triage by AI

📊 Automated PR Analysis

🐛 Type	bug-fix
🟡 Risk	medium

Summary

Replaces the hand-rolled arg walker in glab_cmd.rs's identifier extraction with a proper tokenizer-based approach, using per-subcommand value-taking flag tables and correct -- boundary handling. This fixes bugs where value-taking flags missing from the old hardcoded list would have their values mistakenly hoisted out as the MR/issue identifier.

Review Checklist

• Tests present
• Breaking change
• Docs updated

---

Analyzed automatically by wshm · This is an automated analysis, not a human review.

[aeppling]

lgtm