test: unify exec framework

[ThomasWaldmann]

Repurposed as requested to fix #243.

Every shell now gets the same black-box helper: give it a generated completion and a
command line, get back the candidates that shell actually offers.
candidates(shell, completion, cmdline) dispatches, so the per-shell if/elif
branches collapse into a single assertion and most skip("WiP")s are gone.

	before	after
bash	Bash.compgen, poking the generated script's internal arrays	bash_candidates: sets up COMP_WORDS/COMP_CWORD the way readline does (COMP_WORDBREAKS splitting included) and calls the registered completion function
zsh	zsh_specs (the _arguments spec array)	zsh_candidates: zsh has no complete -C, so drive an interactive zsh through a pty with a widget that only ever lists, so even a unique match shows up
tcsh	pty + autolist, script written into cwd	shares the pty driver; lists via ^D (list-choices) so unique matches show up, and the script is written outside cwd so it is not a candidate itself
fish	complete -C	same mechanism, shared (completion, cmdlines, cwd) signature

zsh_specs is kept rather than renamed: it answers a different question (which spec was
generated), which is what the #224 quoting tests need.

The first commit is the original content of this PR (Bash.test captured neither stdout
nor stderr, so assert not stdout was vacuous) — capturing the output is what makes
reading back bash's candidates possible. Bash itself has no users left and is removed.

Test counts: 71 passed / 33 skipped → 105 passed / 9 skipped / 5 xfailed (runtime 2.6s → 24s,
the pty-driven shells).

Running the same expectations on every shell surfaces five real differences

Marked xfail (so they show up under the existing -rxs) rather than skipped — happy to
split any of these off into their own issues:

• tcsh completes files, not subcommands, after a global option's value (fish's fish: no subcommand completion after a global option that takes a value #228)
• tcsh offers the options of sibling subcommands (fish's fish: completions vanish when an argument value equals a subcommand name #229)
• zsh re-offers an earlier positional's choices at a later slot (fish's fish: positional argument order ignored (earlier positionals' candidates re-offered at later slots) #230)
• zsh's _describe reads the : in a sub:cmd subcommand as its name/description separator
• tcsh treats the / in a word list as a pathname separator, so sub/cmd lists as sub

The last two are exactly what test_subparser_colons/test_subparser_slashes were written
to guard against; they were invisible while only bash was exercised.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 89.18%. Comparing base (fe5f177) to head (201bf5c).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #209   +/-   ##
=======================================
  Coverage   89.18%   89.18%           
=======================================
  Files           3        3           
  Lines         370      370           
=======================================
  Hits          330      330           
  Misses         40       40           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[ThomasWaldmann]

Brought to you by google antigravity. :-)

[casperdcl]

Would prefer if you could repurpose this PR to fix #243 :)

[ThomasWaldmann]

Would prefer if you could repurpose this PR to fix #243 :)

I asked Claude to do it, this is the result. Review carefully! :)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.20%. Comparing base (f16f4b4) to head (32480fe).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #209   +/-   ##
=======================================
  Coverage   85.20%   85.20%           
=======================================
  Files           3        3           
  Lines         473      473           
  Branches       93       93           
=======================================
  Hits          403      403           
  Misses         39       39           
  Partials       31       31           

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[yangfan-yf-yf]

I found three paths where the new execution harness can report expected test outcomes without having exercised a successful completion. Details inline. The hosted checks are green, but they do not cover these failure paths.

[yangfan-yf-yf]

This checks only stderr and parses stdout even when Bash exits with a non-zero status. For example, bad() { exit 7; }; complete -F bad myprog is reported as an empty candidate list, so a broken completion is indistinguishable from a valid no-match result. The trailing printf also masks a completion function that returns non-zero without exiting. Please propagate the function status, check proc.returncode, and add regression cases for both exit and return.

[yangfan-yf-yf]

When the deadline expires, read() returns partial output without verifying that the requested prompt was observed. The caller then parses that output as candidates, so an empty result can satisfy negative assertions even though zsh or tcsh never completed setup or completion. Please raise on a missing prompt and apply a bounded deadline to the drain phase as well.

[yangfan-yf-yf]

This calls pytest.xfail() before the affected candidate query and assertion. The branch therefore never checks the documented tcsh behavior and can never XPASS after it is fixed. The same pattern occurs in the other four shell-specific xfails on this head. Please mark the affected parameter with pytest.mark.xfail(..., strict=True) so that the query and assertion still run.