Is it possible to have multiple CASE in a GROUP BY that can make it so it doesn't group at all? - sql

Question

I'm trying to query come datas and on a spcific case I might have to group datas by multiple values. But most of the time it needs not to be grouped at all. So I'm using multiple CASE WHEN {...} inside the GROUP BY, and all the WHEN basically have the same condition. The problem is that if the condition is met, everything works fine. But if it's false, then the GROUP BY section is empty and the query returns only the first row.
I basically tried to reorganize the quesry in every way that came to my mind, nothing seemed to work, and I didn't find anything conclusive on internet.
I'm using MySql 5.7.
SELECT
{element I want to select}
FROM
{tables}
WHERE
{conditions}
GROUP BY
CASE WHEN (condition) THEN [table].[column] END,
CASE WHEN (condition) THEN [table].[column] END,
CASE WHEN (condition) THEN [table].[column] END
ORDER BY
{...}
Full query :
SELECT
tx.code,
IFNULL(hr.label,'') AS rh_label,
IFNULL(cli.label,'') AS client_label,
DATE(FROM_UNIXTIME(created.value / 1000)) AS Created,
IFNULL(item_enfant.label,'') As Parasite,
IFNULL(item_parent.label,'') As Zone,
CASE
WHEN :perWeek = 'week' THEN SUM(qte.value)
ELSE qte.value
END AS Quantite,
CEILING(DATEDIFF(DATE(FROM_UNIXTIME(created.value / 1000)), DATE(FROM_UNIXTIME(:from / 1000))) / 7) AS Weeks
FROM tx
LEFT JOIN tx_type AS tt ON tt.id = tx.tx_type_id
LEFT JOIN human_resource AS hr ON hr.id = tx.human_resource_id
LEFT JOIN client AS cli ON cli.id = tx.client_id
LEFT JOIN tx_state AS ts ON ts.id = tx.current_tx_state_id
LEFT JOIN workflow_step AS ws ON ws.id = ts.workflow_step_id
LEFT JOIN item AS item_enfant ON item_enfant.item_list_id = tx.item_list_id
JOIN item_type AS ite ON ite.id = item_enfant.item_type_id
LEFT JOIN item_meta AS qte ON qte.item_id = item_enfant.id AND qte.name = 'qtePourRapport'
LEFT JOIN item_prop AS created ON created.item_id = item_enfant.id AND created.name = 'visite.timestamp'
JOIN item AS item_parent ON item_parent.id = item_enfant.parent_item_id
JOIN item_type AS itp ON itp.id = item_parent.item_type_id
WHERE
ite.name = 'parasite' AND
item_enfant.product_id IN (:parasiteIds) AND
itp.name = 'zone' AND
item_parent.product_id IN (:zoneIds) AND
cli.id = (:clientId) AND
ws.logic_id = 600 AND
created.value BETWEEN :from AND :to AND
created.value IS NOT NULL AND qte.value IS NOT NULL
GROUP BY
CASE WHEN :perWeek = 'week' THEN item_enfant.label END, #Parasite
CASE WHEN :perWeek = 'week' THEN item_parent.label END, #Zone
CASE WHEN :perWeek = 'week' THEN CEILING(DATEDIFF(DATE(FROM_UNIXTIME(created.value / 1000)), DATE(FROM_UNIXTIME(:from / 1000))) / 7) END #Weeks
ORDER BY
Created;
I'm getting the datas of the first row alone. And I actually have no idea how to get it just not to group if the condition is not met.

Answer 1

You need a unique value for the aggregation or two separate queries. The simplest method might be union all:
select . . .
from t
where <conditions not to group by>
union all
select . . .
from t
where <conditions to group by>
group by . . .;
You need to be sure that each subquery returns compatible columns.

Answer 2

SELECT
{element I want to select}
FROM
{tables}
WHERE
{conditions}
GROUP BY
CASE WHEN (condition) THEN [table].[column] ELSE [some unique value of same data-type as column] END,
CASE WHEN (condition) THEN [table].[column] ELSE [some unique value of same data-type as column] END,
CASE WHEN (condition) THEN [table].[column] ELSE [some unique value of same data-type as column] END
ORDER BY
{...}
I guess the missing ELSE clause will evaluate to NULL. This is constant, thus all rows will be in the same group, thus there will be only one row returned for this group. To avoid grouping you need unique values over all returned rows in the combination of the grouping-elements (not in every single grouping-element as stated erlier).
EDIT
Thus the soultion from the comment might be easier: Just add another grouping-element CASE WHEN !(condition) THEN CONCAT([different elements making it unique]) END