[pve-devel] [PATCH v2 common] cli-formatter: avoid warning when trying to sort on undefined key

[Thomas Lamprecht]

On 11/28/19 10:18 AM, Fabian Grünbichler wrote:
> On November 28, 2019 10:12 am, Thomas Lamprecht wrote:
>> On 11/28/19 10:08 AM, Fabian Grünbichler wrote:
>>> On November 27, 2019 4:59 pm, Thomas Lamprecht wrote:
>>>> On 11/26/19 1:10 PM, Christian Ebner wrote:
>>>>> Example:
>>>>>     pvesh get /nodes/{node}/qemu/{vmid}/rrddata --timeframe day
>>>>>
>>>>> If the sorting key is not defined in the dataset, e.g. when a VM was not running
>>>>> for some time within the given timeframe, this resulted in several ugly warnings.
>>>>>
>>>>> Signed-off-by: Christian Ebner <c.ebner at proxmox.com>
>>>>> ---
>>>>>
>>>>> v2: Oops, v1 is nonsense and breaks sorting.
>>>>>
>>>>>  src/PVE/CLIFormatter.pm | 6 ++++--
>>>>>  1 file changed, 4 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/src/PVE/CLIFormatter.pm b/src/PVE/CLIFormatter.pm
>>>>> index 0e9cbe6..21fa2df 100644
>>>>> --- a/src/PVE/CLIFormatter.pm
>>>>> +++ b/src/PVE/CLIFormatter.pm
>>>>> @@ -175,9 +175,11 @@ sub print_text_table {
>>>>>      if (defined($sort_key) && $sort_key ne 0) {
>>>>>  	my $type = $returnprops->{$sort_key}->{type} // 'string';
>>>>>  	if ($type eq 'integer' || $type eq 'number') {
>>>>> -	    @$data = sort { $a->{$sort_key} <=> $b->{$sort_key} } @$data;
>>>>> +	    @$data = sort { $a->{$sort_key} <=> $b->{$sort_key}
>>>>> +		if defined $a->{$sort_key} && defined $b->{$sort_key} } @$data;
>>>>>  	} else {
>>>>> -	    @$data = sort { $a->{$sort_key} cmp $b->{$sort_key} } @$data;
>>>>> +	    @$data = sort { $a->{$sort_key} cmp $b->{$sort_key}
>>>>> +		if defined $a->{$sort_key} && defined $b->{$sort_key} } @$data;
>>>>
>>>>
>>>> a post-if in a sort condition feels a bit awkward and does not
>>>> clearly tells a reader what the behavior for undefined cases is, IMO
>>>>
>>>> Maybe use a ternary operation here:
>>>> sort {
>>>>     defined $a->{$sort_key} && defined $b->{$sort_key}
>>>>         ? $a->{$sort_key} cmp $b->{$sort_key}
>>>>         : 1
>>>> }
>>>>
>>>>
>>>> or as alternative a sorter method:
>>>>
>>>> my $hashsort = sub {
>>>>     my ($a, $b, $key) = @_;
>>>>     return 1 if !(defined $a->{$key} && defined $b->{$key});
>>>>     return $a->{$key} cmp $b->{$key};
>>>> }
>>>>
>>>> and use that?
>>>>
>>>> Or even:
>>>>
>>>> return 1 if defined $a->{$key} && !defined $b->{$key});
>>>> return -1 if !defined $a->{$key} && defined $b->{$key};
>>>> return 0 if defined $a->{$key} && defined $b->{$key};
>>>>
>>>> ?? that stuff confuses me after a longer day, sorry ^^
>>>
>>> complete would actually be, with the (hopefully) most likely  case up 
>>> front, and the assumption that undef is less than anything defined ;)
>>> could of course also be written as if / elsif / elsif / else, to make it 
>>> more obvious that all cases are handled.
>>>
>>> return $a->{$key} cmp $b->{key} if defined($a->{key}) && defined($b->{key});
>>> return 1 if defined($a->{$key}) && !defined($b->{$key});
>>> return -1 if !defined($a->{$key}) && defined($b->{$key});
>>> return 0 if !(defined($a->{$key}) && defined($b->{$key}));
>>>
>>
>> That's the exact same I proposed?? or do I miss something?
> 
> your first variants only cover two cases (both defined, or both undef), 
> your second variant only covers three cases and does no comparison when 
> both are defined ;) there are actually four cases (think of definedness 
> as bool).
> 
I mean isn't it _really_ obvious that the actual compare needs to be always
done, your proposed and mine are the same..