fix(config): strip whitespace around comma-separated config values

[ntdatt812]

list=cpu, mem, load does not work, and glances.conf documents it that way

load_limits splits a config list and keeps the spaces:

self._limits[limit] = config.get_value(self.plugin_name, level).split(",")

So list=cpu, mem, load becomes ['cpu', ' mem', ' load']. That is how anyone writes a comma-separated value, and it is how conf/glances.conf writes one in its own comment:

# Available stats are: cpu,mem,load,swap, gpu_mem, gpu_proc

A user copying that line gets a list where four of the six items carry a leading space.

The loudest symptom is show/hide

Every item in those lists is used as a re.fullmatch pattern, so a leading space makes the pattern match nothing at all. With hide=sda2, loop.*:

=== on develop ===
hide patterns: ['sda2', ' loop.*']
  sda2   hidden=True   displayed=False
  loop0  hidden=False  displayed=True     <-- not hidden
  sda1   hidden=False  displayed=True

=== with this patch ===
hide patterns: ['sda2', 'loop.*']
  sda2   hidden=True   displayed=False
  loop0  hidden=True   displayed=False
  sda1   hidden=False  displayed=True

The rule is accepted, logged, and silently does nothing. Nothing tells the user their filter is dead.

Quicklook: a config mistake showed more than the default

Separately, quicklook answered an unrecognised list by falling back to everything available:

if not set(self.stats_list).issubset(self.AVAILABLE_STATS_LIST):
    logger.warning(f'Quicklook plugin: Invalid stats list: {self.stats_list}')
    self.stats_list = self.AVAILABLE_STATS_LIST

AVAILABLE_STATS_LIST contains gpu_mem and gpu_proc, and the next three lines use exactly those to flip gpu_stats.get_gpu_mem / get_gpu_proc. So a typo — or, before the first fix, an ordinary space — started polling the GPU on a machine whose owner never asked for it. That is not a neutral default; on the AMD runtime-PM hardware in #3590 it is reads against a suspended card.

Measured on develop:

list='cpu,mem,load'      -> ['cpu', 'mem', 'load']
list='cpu, mem, load'    -> ['cpu', 'mem', 'swap', 'load', 'gpu_mem', 'gpu_proc']
list='cpu,mem,typo'      -> ['cpu', 'mem', 'swap', 'load', 'gpu_mem', 'gpu_proc']

All three now resolve to what the user asked for, or to DEFAULT_STATS_LIST — the same list get_conf_value already uses as its default=. The warning also names which entries were unrecognised, instead of printing the whole list back.

Scope

Two changes, kept together because fixing either alone leaves the other reachable: with only the strip, a genuine typo still turns on the GPU; with only the fallback, list=cpu, mem, load still silently degrades.

The strip lives in load_limits, so it is one place rather than one per plugin. It touches only the edges of each item — internal spaces survive, which matters for values like alias=sda1:System Disk, and there is a test pinning that.

I looked for anything that could depend on the old behaviour. ConfigParser already strips the value as a whole, so a leading or trailing space on a single-item value was never reachable; only interior items were affected, and for those the space was never wanted. bar_char reads [0] of its list and is a single item, so it is unaffected either way.

Verification

11 new tests in tests/test_plugin_quicklook.py. Seven of them are red against develop:

FAILED test_whitespace_around_items_is_ignored[cpu, mem, load]
FAILED test_whitespace_around_items_is_ignored[cpu ,mem ,load]
FAILED test_whitespace_around_items_is_ignored[ cpu , mem , load ]
FAILED test_whitespace_around_items_is_ignored[cpu,\tmem,\tload]
FAILED test_a_typo_falls_back_to_the_default_not_to_everything
FAILED test_limits_carry_stripped_items
FAILED test_internal_spaces_are_kept
7 failed, 3 passed

(The hide/show test was added after that run and is red too; it is the loop0 case shown above.)

Full suite, same invocation both times, only glances/ reverted for the baseline:

	failed	passed
develop (new test file excluded)	40	641
this branch	40	652

652 = 641 + the 11 new tests, and the two failure sets are identical — I diffed the FAILED names, and both comm directions are empty. The 40 are pre-existing and environmental on this box: test_restful.py and test_xmlrpc.py cannot start their servers here, plus test_perf and test_memoryleak. test_mcp.py and the browser/webui suites were excluded from both runs for the same reason.

ruff check and ruff format --check clean on all three files.

[nicolargo]

Thanks for the PR @ntdatt812

Just merged into develop.