Files
fish-config/tests/test-guards.fish
T
rootiest 9d7cddf758 docs: strip gitignored AGENTS/ and JOB-BRIEF-FINDINGS references from code comments
Comments in tracked files pointed readers at AGENTS/specs/*.md,
AGENTS.md, and JOB-BRIEF-FINDINGS.md -- all gitignored, transient
dev notes not part of the base repo. Rewrote each to be self-contained
or point at tracked docs (docs/fish-config.md) instead.
2026-09-08 12:54:53 -04:00

332 lines
12 KiB
Fish

#!/usr/bin/env fish
# Copyright (C) 2026 Rootiest
# SPDX-License-Identifier: AGPL-3.0-or-later
#
# Table-driven coverage of the opinionated-component guard system:
# __fish_variable_check, __fish_config_op_cascade,
# __fish_config_op_registry_lookup and __fish_config_op_enabled.
#
# Runs isolated (no `# MODE:` marker, which means isolated): the driver gives
# this process temp XDG dirs and --no-config, so there is no loaded config and
# no universal variables. That is load-bearing in both directions --
#
# * these cases `set -e` real guard variable names, including
# __fish_config_op_logging and __fish_config_opinionated. Under a plain
# `fish` those erase from the innermost scope holding the variable, which
# for a user with the variable set universally means destroying a real
# universal variable. The temp XDG_CONFIG_HOME is what makes that
# impossible.
# * the two `set -p` / `source` lines below are what make the FORK's code the
# code under test. Their failure modes differ and the difference matters:
# forgetting the fish_function_path prepend fails LOUDLY (the guard
# functions are unresolvable under --no-config and a call exits 127, which
# can never equal an expected 0/1/2/3, so every case turns red), while
# forgetting the registry source fails SILENTLY (__fish_config_op_enabled
# fail-opens on a missing entry, so a third of the table would pass for
# entirely the wrong reason). Hence the explicit precondition below.
source (realpath (dirname (status filename)))/lib.fish
set -p fish_function_path $repo_root/functions
source $repo_root/conf.d/__fish_config_op_registry.fish
section "guards: preconditions"
check "the fork's registry is loaded" 65 (count $__fish_config_op_registry_keys)
section "__fish_variable_check: truthy"
for v in 1 true yes on y Y TRUE ON
set -g __probe_v $v
__fish_variable_check __probe_v
check "truthy '$v' -> 0" 0 $status
end
section "__fish_variable_check: falsy"
for v in 0 false no off n OFF FALSE
set -g __probe_v $v
__fish_variable_check __probe_v
check "falsy '$v' -> 1" 1 $status
end
section "__fish_variable_check: neither"
set -e __probe_v
__fish_variable_check __probe_v
check "unset -> 2" 2 $status
set -g __probe_v ""
__fish_variable_check __probe_v
check "empty string -> 2" 2 $status
set -g __probe_v banana
__fish_variable_check __probe_v
check "unrecognized -> 3" 3 $status
__fish_variable_check
check "no argument -> 2" 2 $status
set -g __probe_v a b
__fish_variable_check __probe_v
check "multi-element list -> 3" 3 $status
set -e __probe_v
section "cascade: category only"
# Invented variable names the real config can never set, so nothing ambient can
# perturb these -- only the C5 and master cases below need real names.
set -e __fish_config_opinionated __probe_cat __probe_cat_sub
__fish_config_op_cascade __probe_cat
check "all unset -> enabled" 0 $status
set -g __probe_cat 0
__fish_config_op_cascade __probe_cat
check "category falsy -> disabled" 1 $status
set -g __probe_cat 1
__fish_config_op_cascade __probe_cat
check "category truthy -> enabled" 0 $status
set -g __probe_cat garbage
__fish_config_op_cascade __probe_cat
check "category unrecognized defers to master -> enabled" 0 $status
set -e __probe_cat
section "cascade: subcategory beats category"
set -g __probe_cat 0
set -g __probe_cat_sub 1
__fish_config_op_cascade __probe_cat __probe_cat_sub
check "sub on, category off -> enabled" 0 $status
set -g __probe_cat 1
set -g __probe_cat_sub 0
__fish_config_op_cascade __probe_cat __probe_cat_sub
check "sub off, category on -> disabled" 1 $status
set -e __probe_cat_sub
set -g __probe_cat 0
__fish_config_op_cascade __probe_cat __probe_cat_sub
check "sub unset, category off -> disabled" 1 $status
set -g __probe_cat_sub garbage
__fish_config_op_cascade __probe_cat __probe_cat_sub
check "sub unrecognized defers, category off -> disabled" 1 $status
set -e __probe_cat __probe_cat_sub
__fish_config_op_cascade __probe_cat ""
check "empty subcategory argument -> category-only chain" 0 $status
section "cascade: the master is an off switch only"
set -g __fish_config_opinionated 0
__fish_config_op_cascade __probe_cat
check "master off, category unset -> disabled" 1 $status
set -g __probe_cat 1
__fish_config_op_cascade __probe_cat
check "master off, category on -> enabled" 0 $status
set -e __probe_cat
set -g __fish_config_opinionated 1
__fish_config_op_cascade __probe_cat
check "master on, category unset -> enabled" 0 $status
set -g __fish_config_opinionated garbage
__fish_config_op_cascade __probe_cat
check "master unrecognized, category unset -> enabled" 0 $status
set -e __fish_config_opinionated
section "cascade: C5 logging is opt-in"
# Uses the REAL __fish_config_op_logging name because the opt-in list lives
# inside the cascade keyed on it. C5 logging is opt-in by design (see
# docs/fish-config.md's "C5 -- Logging and Capture" section): unset or
# unrecognized means off, and the master switch cannot enable it. If any of
# these three fail, that is a regression in the guard itself, not the test.
set -e __fish_config_op_logging __fish_config_opinionated
__fish_config_op_cascade __fish_config_op_logging
check "C5 unset -> disabled" 1 $status
set -g __fish_config_opinionated 1
__fish_config_op_cascade __fish_config_op_logging
check "C5 unset + master truthy -> still disabled" 1 $status
set -e __fish_config_opinionated
set -g __fish_config_op_logging garbage
__fish_config_op_cascade __fish_config_op_logging
check "C5 unrecognized is not consent -> disabled" 1 $status
set -g __fish_config_op_logging on
__fish_config_op_cascade __fish_config_op_logging
check "C5 explicit truthy -> enabled" 0 $status
set -g __fish_config_op_logging off
__fish_config_op_cascade __fish_config_op_logging
check "C5 explicit falsy -> disabled" 1 $status
set -e __fish_config_op_logging
section "cascade: C5 subcategories inherit opt-in"
# The opt-in check reads chain[-1], which is always the CATEGORY variable, so
# nesting inherits "off unless explicit" with no per-subcategory special case.
set -e __fish_config_op_logging_terminal_capture
__fish_config_op_cascade __fish_config_op_logging __fish_config_op_logging_terminal_capture
check "C5 sub unset, C5 unset -> disabled" 1 $status
set -g __fish_config_op_logging_terminal_capture 1
__fish_config_op_cascade __fish_config_op_logging __fish_config_op_logging_terminal_capture
check "C5 sub explicit truthy -> enabled" 0 $status
set -g __fish_config_op_logging_terminal_capture 0
set -g __fish_config_op_logging on
__fish_config_op_cascade __fish_config_op_logging __fish_config_op_logging_terminal_capture
check "C5 sub falsy, C5 truthy -> disabled" 1 $status
set -e __fish_config_op_logging_terminal_capture __fish_config_op_logging
section "registry lookup"
__fish_config_op_registry_lookup cat "" >/dev/null
check "known unsited key found" 0 $status
set -l t (__fish_config_op_registry_lookup cat "")
check "cat's tags" aliases/filesystem "$t"
__fish_config_op_registry_lookup nosuchthing "" >/dev/null
check "unknown identity -> not found" 1 $status
set -l t2 (__fish_config_op_registry_lookup config cdpath)
check "sited key config:cdpath" overrides/environment "$t2"
# The key is the identity:site PAIR, not the identity alone.
__fish_config_op_registry_lookup cat wrongsite >/dev/null
check "known identity, wrong site -> not found" 1 $status
section "op_enabled: against the real registry"
set -e __fish_config_op_aliases __fish_config_op_aliases_filesystem
set -e __fish_config_opinionated
__fish_config_op_enabled cat
check "cat, nothing set -> enabled" 0 $status
__fish_config_op_enabled cat.fish
check "a .fish suffix is stripped" 0 $status
set -g __fish_config_op_aliases 0
__fish_config_op_enabled cat
check "cat, aliases off -> disabled" 1 $status
set -g __fish_config_op_aliases_filesystem 1
__fish_config_op_enabled cat
check "cat, aliases off but its subcategory on -> enabled" 0 $status
set -e __fish_config_op_aliases __fish_config_op_aliases_filesystem
# Fail-open: an unclassified identity, or one whose doc header has no
# # COMPONENT section, resolves to enabled. This is what keeps user-authored
# and third-party functions unaffected.
__fish_config_op_enabled __totally_unregistered somesite
check "no registry entry -> fail open" 0 $status
section "op_enabled: always/* and AND, via a synthetic registry"
# Why this fixture exists, so nobody deletes it as redundant:
#
# The generated registry has 65 entries, EVERY ONE carrying exactly one tag,
# and contains no always/on or always/off anywhere (measured 2026-09-07
# against conf.d/__fish_config_op_registry.fish). So three documented
# semantics -- always/off, always/on, and AND-across-tags -- have no reachable
# case in production data and would otherwise go completely untested.
#
# The registry is just two global lists, and in isolated mode nothing else in
# this process reads them, so overriding them is free.
set -g __fish_config_op_registry_keys "syn_on:" "syn_off:" "syn_and:" "syn_bare:" "syn_multi:"
set -g __fish_config_op_registry_values \
"always/on" \
"always/off" \
"aliases/filesystem integrations/notifications" \
"aliases" \
"always/off always/on"
set -e __fish_config_op_aliases __fish_config_op_integrations __fish_config_opinionated
__fish_config_op_enabled syn_off
check "always/off -> disabled" 1 $status
set -g __fish_config_op_aliases 1
__fish_config_op_enabled syn_off
check "always/off ignores an enabled category" 1 $status
set -e __fish_config_op_aliases
__fish_config_op_enabled syn_on
check "always/on -> enabled" 0 $status
set -g __fish_config_op_aliases 0
__fish_config_op_enabled syn_on
check "always/on short-circuits a disabled category" 0 $status
set -e __fish_config_op_aliases
__fish_config_op_enabled syn_multi
check "always/off beats always/on" 1 $status
section "op_enabled: AND across tagged sub-categories"
__fish_config_op_enabled syn_and
check "both categories default -> enabled" 0 $status
set -g __fish_config_op_aliases 0
__fish_config_op_enabled syn_and
check "first tag's category off -> disabled" 1 $status
set -e __fish_config_op_aliases
set -g __fish_config_op_integrations 0
__fish_config_op_enabled syn_and
check "second tag's category off -> disabled" 1 $status
set -e __fish_config_op_integrations
set -g __fish_config_op_aliases 1
set -g __fish_config_op_integrations 1
__fish_config_op_enabled syn_and
check "both explicitly on -> enabled" 0 $status
set -e __fish_config_op_aliases __fish_config_op_integrations
section "op_enabled: a tag with no slash"
# Degenerate but harmless: with no '/', $parts[2] is empty and the derived
# subcategory name gets a trailing underscore. That name is simply always
# unset, so the chain falls through to the category. Pinned so a future reader
# does not mistake it for a bug.
__fish_config_op_enabled syn_bare
check "bare tag 'aliases' -> enabled by default" 0 $status
set -g __fish_config_op_aliases 0
__fish_config_op_enabled syn_bare
check "bare tag honors its category" 1 $status
set -e __fish_config_op_aliases
section "op_enabled: C5 through the guard"
# The path production code actually takes, as opposed to calling the cascade
# directly. Same rule: unset means off and the master cannot enable it.
set -g __fish_config_op_registry_keys "syn_log:"
set -g __fish_config_op_registry_values "logging/terminal-capture"
set -e __fish_config_op_logging __fish_config_op_logging_terminal_capture
__fish_config_op_enabled syn_log
check "C5-tagged component, nothing set -> disabled" 1 $status
set -g __fish_config_opinionated 1
__fish_config_op_enabled syn_log
check "C5-tagged component + master truthy -> still disabled" 1 $status
set -e __fish_config_opinionated
set -g __fish_config_op_logging 1
__fish_config_op_enabled syn_log
check "C5-tagged component + explicit C5 on -> enabled" 0 $status
set -e __fish_config_op_logging
report