tdf#133858 opening spreadsheet with large array formula takes 10 mins

the spreadsheet has lots of SUM(IF over a whole column.

Which results in us allocating a matrix with 1 million rows, which is
rather slow to process.

So reduce the matrix to the data that is actually there.

We can only do this for some opcodes, because other opcodes act
differently when referencing empty space, so for now I only perform this
optimisation for the opcode in use in this spreadsheet.

This takes the load time from 5m to 3s on my machine.

Change-Id: I41fe9afcb0fbdf2a928a19c44a0f291a1247a41c
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96022
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
(cherry picked from commit 9a5f2961b085ce2f23ecdf0a03d1114bacac8e2c)
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96041
Tested-by: Jenkins CollaboraOffice <jenkinscollaboraoffice@gmail.com>

1 files changed, 9 insertions, 0 deletions

diff --git a/sc/source/core/tool/interpr5.cxx b/sc/source/core/tool/interpr5.cxx
index aefb26c920e3..5de62ce07bfe 100644
--- a/sc/source/core/tool/interpr5.cxx
+++ b/sc/source/core/tool/interpr5.cxx
@@ -325,6 +325,15 @@ ScMatrixRef ScInterpreter::CreateMatrixFromDoubleRef( const FormulaToken* pToken
         return nullptr;
     }
 
+    // Clamp the size of the matrix to rows which actually contain data.
+    // For e.g. SUM(IF over an entire column, this can make a big difference.
+    // Limit to ocEqual opcode for now, some opcodes behaviour differently if the
+    // input has empty space.
+    if (nTab1 == nTab2 && pCur->GetOpCode() == ocEqual)
+    {
+        pDok->ShrinkToDataArea(nTab1, nCol1, nRow1, nCol2, nRow2);
+    }
+
     SCSIZE nMatCols = static_cast<SCSIZE>(nCol2 - nCol1 + 1);
     SCSIZE nMatRows = static_cast<SCSIZE>(nRow2 - nRow1 + 1);