Author Archive

First bug of the year

And it’s a funny one. Save all your work, then make up a simple datasource that has a formula in the header:

Now, turn the Macro Recorder on, and while it’s recording, turn that datasource into an Excel Table. (I use the Ctrl T shortcut for that)
Create Table
Excel will warn you that the formula will be converted to static values. Click No.

Happens for me using Office 365. Pretty obscure, granted.

Evaluate Mid

In my last post, I created an array from formula text by using VBA’s Evaluate method, in order to roll my own FormulaArray function that displays the array returned by a formula, for documentation purposes.

In the course of this, I’ve discovered something a bit weird about how this method evaluates the arrays returned by a MID function.
Let’s use this snippet:

Sub EvaluateThis()
Dim var As Variant
var = ActiveSheet.Evaluate(ActiveCell.Formula)
End Sub

First, let’s look at how it handles an array generated by the COLUMN() function:
Evaluate COLUMN array2

Now watch what happens when we use that array to split apart a string using the MID() function:
Evaluate MID array2

So if you push F9, you get an array, but if you use the Evaluate method you don’t…you just get the first letter. Is this weird, or am I missing the point?


Since Excel 2013, Microsoft has given us a FORMULATEXT function, which if you point at a cell will do just that. Here’s a naughty snapshot of FORMULATEXT playing with itself in the corner:

I thought I’d have a go at writing a FormulaArray function to complement it, because when I’m building up a complicated formula that uses lots of array manipulation, then I like to document how all the different arrays within it fit together. (I was going to say “come together” there, but after that crack about FORMULATEXT playing with itself, I thought better of it. But now that I’ve said crack, I’m gonna throw caution to the wind and say wind too.)

Currently I document my formula beasts by either either array-enter a sub-part in the sheet with some notes, like this:
Documentation 1
…which shows how my ExtractNumber formula works, or I enter the desired formula in one cell with a ShowFormula to the left and a hand-rolled hard-coded array to the right, like in this table where I’m documenting a few ways to dynamically generate consecutive integers:
Documentation 2
I get that ResultArray manually, by clicking in the formula bar, pushing F9, copying the resulting evaluated array, then pasting it in another cell. Tedious. Especially when I later make a change to that sub-part, because then I get to do those steps all over.

So I started to roll my own FormulaArray function. I got a bit bogged down in the joining bit, but after about an hour of Googling, I rediscovered Nigel Heffernan’s code for joining two dimensional arrays. Which is very concerning, because I discovered it like just 10 days ago, and even wrote an extensive blog post on it right here. Senility is obviously setting in. If I start saying the same thing over and over like my mother does, just shoot me. If I start saying the same thing over and over like my mother does, just shoot me.

Anyways, Nigel’s function needs a 2D array. You can create an array from formula text by using VBA’s Evaluate method. If the formula returns a Row vector or a 2D vector, then Evaluate nicely turns it into a 2D vector. But here’s the rub: if the formula returns a Column vector, then Evaluate only gives us a 1D vector, which ain’t gonna wash with Nigel’s function:
Row vs Column
So what we need to do is TRANSPOSE any formulas that would return Column vectors, because chucking a TRANSPOSE into the mix has the desired effect:
Row vs Column2
Note that I’m using the square brackets [ ] shortcut for Evaluate. I could just have easily done it like this:
Row vs Column3

Okay, so we know that if our formula string returns a Column vector, we’ve got to transpose it. But how can we tell that ahead of time? I can’t think of a way. So I just do this:

Function FormulaArray(Target As Range) As String
Dim strInput As String
Dim var2 As Variant
Dim lb As Long

strInput = Mid$(Target.Formula, 2)
var2 = ActiveSheet.Evaluate(strInput)
On Error Resume Next
lb = LBound(var2, 2)
If Err.Number <> 0 Then
    var2 = Application.Transpose(ActiveSheet.Evaluate(strInput))
    FormulaArray = Join2d(var2, ",", ";")
    FormulaArray = Join2d(var2, ";", ",")
End If
End Function

So I evaluate the formula as if it’s a Row vector, then check if I’ve got 2 dimensions as a result. If not, it must have been a column vector, in which case I transpose it, then reevaluate it. Shame about the double evaluation, but I can’t think of a foolproof way to do it differenty, other than perhaps array entering the formula into a 2D range on the worksheet and looking at where the #N/A! errors fall.

Anyway, it seems to work just fine:

..unless you happen to be using Structured Table References, and your arguments happen to use the @ table notation to point at something on the same row:
…or unless you happen to have a formula with the INDIRECT function in it:
With the ThisTableRow thing, I guess I can just replace the @[SomeColumn] bit with the actual address, but I can’t think of easy ways around the INDIRECT thing. Anyone got any ideas?

Edit: Thinking about this some more, all I need to do is substitute the INDIRECT(SomeExpression) with whatever gets returned by RANGE(SomeExpression).value

Sample workbook:


There’s a handy post over at Charles Williams’ site that talks about some other quirks of Evaluate that’s worth checking out:

Out, damn’d gridlines! Out, I say!

So after doing some incredibly complicated formula challenges and ninja-grade VBA, I thought I’d turn my hand to something simple: finally changing the default template that loads when Excel starts or when I create a new workbook or worksheet so that gridlines are turned off. Because if there’s one thing I hate about Excel, it’s those gridlines: they make everything look like it was done in Excel. And if there’s one thing I do as soon as I open a new file, it’s turn those gridlines off. Tables and PivotTables give me all the borders I need, thank you very much.

Boy, what a battle. I have saved my changed book as an XLTM to just about everywhere I can think of.

  • I’ve tried saving it to C:\Users\Samsung\AppData\Roaming\Microsoft\Excel\XLSTART but when I start Excel, I have gridlines.
  • I’ve tried creating a new file at C:\xlStart with the template in it, and told Excel via Options>Advanced to open files in that folder but when I start Excel, I have gridlines:


  • I’ve saved it to C:\Users\Samsung\Documents\Custom Office Templates but when I start Excel, I have gridlines

I’ve gone from feeling like I’ve mastered to Excel, to feeling like I’m a complete idiot. Anyone care to tell me that I’m not?

How the heck the average user is supposed to know how to do this stuff is beyond me. Why isn’t there simply a button on the ribbon or backstage that says:
Give all future workbooks the settings of this one.

I’m using Excel 2013 365. But I’m thinking of doing a complete 180.

Dynamically extracting the nth numerical element without delimiters

ExcelXOR has a great post on using formulas to extract all numbers from a string where:

  • The string in question consists of a mixture of numbers, letters and special characters
  • The numbers may appear anywhere within that string
  • Decimals within the string are to be returned as such
  • The desired result is to have all numbers returned to separate cells

That’s a tall order with formulas. Here’s what ExcelXOR came up with:


…where s is the string you want to break apart, and e the element you want returned.

That fatboy runs to 415 characters. Which is a heck of a lot less than my first effort:

*COLUMN(OFFSET($A$1,,,,LEN(s)))^0) = SMALL(IF((ISNUMBER(-MID(s&"|",ROW(OFFSET($A$1,,,LEN(s)))*COLUMN(OFFSET($A$1,,,,LEN(s)))^0,

…although there is some pretty nifty stuff going on in there, that I’ll bore you with at a later date…including a MID that breaks a string down into ALL possible slices of text:


I couldn’t rest with just that. Literally. It was too heavy. So I had another crack. The result: Here’s a generic ExtractNumbers formula I just put together. And by ‘just’ I mean I only just managed it, and it took an entire weekend, I ignored the kids, and forgot to bathe. (That last one is pretty much a given, and I can’t really blame Excel).

=MID(s,SMALL( IF( ISERROR( -MID( TEXT( MID("||"&s,ROW( OFFSET( $A$1,,,LEN( s))),2),"|"),2,1)),FALSE,ROW( OFFSET( $A$1,,,LEN( s)))),e)
-1,SUM( SMALL(IF( ISERROR( -MID( TEXT( MID("|"&s&"|",ROW( OFFSET( $A$1,,,LEN( s)+1)),2),"|"),{1,2},1)),FALSE,ROW( OFFSET( $A$1,,,LEN(

…again, where:
S = the string you want to break apart
E = the number element you want to return

It will handle numbers with decimal places provided there is a digit to the left of the decimal place e.g. SomeText5.745 and NOT SomeText.745, and it’s a much more svelte 277 characters in length. Isn’t she a beauty? A lot of the inspiration for the approach came from Excel Ninja Sajan, over at the awesome Formula Challenges section of Chandoo’s forum.

In that incarnation, you can use it to extract just one element of a specific number:

Or if you prefer, you can use this version:


…you can use it to extract numbers into separate columns, where $A28 holds the string to be split, and B28 holds the first column that you want to extract a number to. Like so:

I don’t know how either of these perform against a UDF, let alone each other. Anyone got a lean, mean, UDF-based splitting machine that we can test it against?

Here’s a sample file:

—Edit 21 Nov 2014—
It turns out that my above formula fails for a few specific special character & number combinations. Here’s a table, showing in which cases Excel will still treat a number as a number when you pad it out with a non number. For completeness I do three tests:

  • Special character before the number
  • Special character after the number
  • Special character on either side of the number

special character matches

—Edit 8 Dec 2014—
Crikey…after chaining myself to the computer since the last update, I finally managed to cobble together a formula that will extract all numbers in practically any shape or form that the local version of Excel deems as a number. That is, given a string like this:
Jeff Weir Age: 43 DOB: 25/4/71 Salary: $100,000 StartTime: 8:30
…It returns this

It ain’t pretty. If you like UDFs, then you’ll agree with my pal Gareth Hayter that this is pure formulabation. Unless you like formulas, in which case it’s orgasmic. (You sick, sick analyst, you. Don’t let your boss catch you playing with it in the office…)

=MID( String,SMALL( IF( ( FREQUENCY( ISNUMBER( -MID( String,ROW( OFFSET( $A$1,,,LEN( String)))*COLUMN( OFFSET( $A$1,,,,LEN( String)))^0,COLUMN( OFFSET( $A$1,,,,LEN( String)))*ROW( OFFSET( $A$1,,,LEN( String)))^0))*( COLUMN( OFFSET( $A$1,,,,LEN( String)))+ROW( OFFSET( $A$1,,,LEN( String)))-1),ROW( OFFSET( $A$1,,,LEN( String)+1))-1)=1)*ISNUMBER( -MID( "|"&String&"|",ROW( OFFSET( $A$1,,,LEN( String)+2)),1))=1,ROW( OFFSET( $A$1,,,LEN( String)+2))-1,FALSE),ROW( OFFSET( $A$1,,,LEN( String)))),SMALL( IF( ( FREQUENCY( ISNUMBER( -MID( String&"|",ROW( OFFSET( $A$1,,,LEN( String)))*COLUMN( OFFSET( $A$1,,,,LEN( String)))^0,COLUMN( OFFSET( $A$1,,,,LEN( String)))*ROW( OFFSET( $A$1,,,LEN( String)))^0))*( LEN( String)+1-ROW( OFFSET( $A$1,,,LEN( String)))),MOD( LEN( String)+2-ROW( OFFSET( $A$1,,,LEN( String)+1)),LEN( String)+1))=1)*ISNUMBER( -MID( "|"&String,ROW( OFFSET( $A$1,,,LEN( String)+2)),1))=1,ROW( OFFSET( $A$1,,,LEN( String)+2))-1,FALSE),ROW( OFFSET( $A$1,,,LEN( String))))-SMALL( IF( ( FREQUENCY( ISNUMBER( -MID( String,ROW( OFFSET( $A$1,,,LEN( String)))*COLUMN( OFFSET( $A$1,,,,LEN( String)))^0,COLUMN( OFFSET( $A$1,,,,LEN( String)))*ROW( OFFSET( $A$1,,,LEN( String)))^0))*( COLUMN( OFFSET( $A$1,,,,LEN( String)))+ROW( OFFSET( $A$1,,,LEN( String)))-1),ROW( OFFSET( $A$1,,,LEN( String)+1))-1)=1)*ISNUMBER( -MID( "|"&String&"|",ROW( OFFSET( $A$1,,,LEN( String)+2)),1))=1,ROW( OFFSET( $A$1,,,LEN( String)+2))-1,FALSE),ROW( OFFSET( $A$1,,,LEN( String))))+1)

Here’ a file showing how this bit of formulabation comes together: Dynamic Split Numbers_20141208 v6

An INDEX of insights from ExcelXOR

Everybody knows that you can’t do some things with the result of an INDEX function unless you return it to the worksheet first. For instance, I can return the 2nd and 3rd members of the array {10,20,30} to the worksheet with an array-entered INDEX function like this:

…but if I wrap a SUM around that INDEX and array enter it, I get the wrong answer:
…meaning I have to return that INDEX function to the grid and then point my SUM function at it:
Everybody knows that, right? Well, everybody’s wrong. Everybody but the author over at the new ExcelXOR blog, that is. At this earth-shattering post the Author shows several intriguing ways to get INDEX to play nicely with outer functions:


Far out! That’s the kind of trick I would have sold my soul to the devil to learn.

There’s a lot to learn at that site: go to and scroll most of the way down until the Archives section, and then get clicking. The earliest post is practically from yesterday: August 2014. But there’s a lifetime of learnings there already…both in the posts, and in the comments. Plus the most wicked formula challenges you will find in one place.

Cult of the Flying Spaghetti VLOOKUP


Sumit Bansal’s post VLOOKUP Vs. INDEX/MATCH – The Debate Ends Here! sparked some great discussion on the merits of VLOOKUP vs INDEX/MATCH, including at Oz du Soleil’s lighthearted rebuttal at The Anti-VLOOKUP Crowd Is Out In The Streets Again!

I especially love Peter B’s comment at Sumit’ post:

My opinion is that VLOOKUP and HLOOKUP are simply over-specialised legacy functions and Excel would be all the better for ‘pruning’ them out. I do use VLOOKUP occasionally when I have a 2-D range; the search array happens to be on the left; I only wish to return a single field; I am sure the data is clean and the match will always succeed. Despite that, I think the value they bring to the bloated zoo of Excel functions is not worth their keep.

Of course, NOTHING can ever be cleaned out of Excel, for good reason…otherwise all the millions of complex black-box spreadsheets that continue to function just fine long after the person who constructed them moved on to another task, job, or incarnation will break. Not to mention all those fantasy football spreadsheets. MS has backwards-compatibility issues that are beyond belief really.

At the same time I agree with Bob Phillips’ point at Sumit’s post:

The biggest selling point to me is that VLOOKUP is easy to teach to people, and it sticks, INDEX/MATCH less so.

But I disagree with Bob’s point that VLOOKUP can be/is just as flexible as INDEX/MATCH, merely because we can do stuff like this with it:

Just as flexible? Maybe, if you bend it double with brute force. Just as fast after you’ve made it just as flexible? Not likely. Any more understandable than the INDEX/MATCH equivalent? Not in my experience.

In fact, I feel a rude joke coming on:

Young analyst with unlit cigarette in mouth, having just consummated his first VLOOKUP: Has anyone got a match?
Analyst of distinguished years: Yes. Your VLOOKUP and my arse.

If MS were designing Excel from scratch – and I was on the committee that was deciding whether to include a dumbed-down function to do a subset of lookups based on hard-coded input parameters and a fixed data layout – then I’d make a case for not including it. Not just because of those quite reasonable complaints, either. But also because of evolution. A user that is forced to learn INDEX and MATCH due to lack of suitable alternatives is be better placed to evolve into a higher Excel lifeform than one that hasn’t looked beyond VLOOKUP.

(I’d make an exception if a major competitor – say Lotus – had a VLOOKUP function in their beast. But only in that specific case.)

Formulas remind me a bit like DNA: just by stringing a few different base-pairs together in the right order, you can build a mouse. Or a Human, with a few extra tweaks. Similarly, with a few good formula combinations under your belt, you can conquer most problems you’re likely to come across. INDEX and MATCH are not just formulas in their own right, but are the formula equivalent of DNA basepairs: they give users a peek into other formula ecosystems that they can gradually spread into and colonize. VLOOKUP ain’t one of those base pairs. It’s Neanderthal.

Hey, don’t get me wrong: I’m fine that it’s in the fossil record. I’m happy enough to have one in my spreadsheet, just as I’m happy enough to have an appendix that doesn’t burst.

String concatenation is like the weather…

…everyone complains about it, but nobody does anything about it. Well, certainly not Microsoft, anyhows. But back in 2012 Nigel Heffernan at did: he put up some nifty code for joining and splitting two dimensional arrays that I just stumbled across. I thought I’d have a go at turning the join one into a function that can be called from the worksheet, and add a few more options while I’m at it. More butchering than tweaking, as you’re about to see.

My revision can be called from the worksheet, and has the following arguments:

Yes, more arguments than at my last social outing. Most are optional and have defaults. Take that how you will. The default delimiter is a comma. The Field Delimiter is a separate Delimiter that gets added if your input array is 2D, and the default is also a comma. EndDelimiter puts an extra Delimiter of your choice on the end if you want one. Aesthetics only, really. The rest are explained below.

First, the result:
JoinText 20141115

  • That orange block is my data.
  • Column D shows the result if you point the function at each respective row
  • Row 8 shows the result of pointing the function at each respective column
  • In rows 12 to 15 you see the result of pointing it at the entire 2D block of data, under different settings.

Those last two results are what happens if the data is laid out by row and then by column, and you’ve incorrectly told the UDF to transpose the input array. If your data happenned to be laid out like this, you wouldn’t need that Transpose argument:
JoinText 20141115 2

The DelimitEnd argument does something pretty minor, really. If we include it, the end of the string gets padded with it – in this case an Exclamation Mark . If we exclude it, the string doesn’t get padded with any extra delimiters:
JoinText 20141115 3

You might notice it skips blanks. It doesn’t have to, if you don’t want it to:
JoinText 20141115 4

And it doesn’t need your two arrays to be the same size:

JoinText 20141115 5

A real-world example where this might be useful is when concatenating lists of names, where some may have more parts than others:
JoinText 20141115 6

Both the last two screenshots show examples of using three different delimiters…a space between words, a comma between columns, and something different on the end.

Here’s the code and workbook:
Join Function_20141115

Public Function JoinText(target As Range, _
                           Optional Delimiter As String = ",", _
                           Optional FieldDelimiter As String = ",", _
                           Optional EndDelimiter As String = "", _
                           Optional SkipBlanks As Boolean = False, _
                           Optional Transpose As Boolean = False) As String

'Based on code from Nigel Heffernan at

' Join up a 2-dimensional array into a string.

'   ####################
'   # Revision history #
'   ####################

'   Date (YYYYMMDD)     Revised by:         Changes:
'   20141114            Jeff Weir           Turned into worksheet function, added FinalDelimiter and Transpose options
'   20141115            Jeff Weir           Changed FinalDelimiter to EndDelimiter that accepts string, with default of ""

Dim InputArray As Variant
Dim i As Long
Dim j As Long
Dim k As Long
Dim lngNext As Long
Dim i_lBound As Long
Dim i_uBound As Long
Dim j_lBound As Long
Dim j_uBound As Long
Dim arrTemp1() As String
Dim arrTemp2() As String

    If target.Rows.Count = 1 Then
        If target.Columns.Count = 1 Then
            GoTo errhandler 'Target is a single cell
            ' Selection is a Row Vector
            InputArray = Application.Transpose(target)
            Transpose = True
        End If
        If target.Columns.Count = 1 Then
            ' Selection is a Column Vecton
            InputArray = target
            'Selection is 2D range. Transpose it if that's what the user has asked for
            If Transpose Then
                InputArray = Application.Transpose(target)
                Transpose = True
            Else: InputArray = target
            End If
        End If
    End If

    i_lBound = LBound(InputArray, 1)
    i_uBound = UBound(InputArray, 1)
    j_lBound = LBound(InputArray, 2)
    j_uBound = UBound(InputArray, 2)
    ReDim arrTemp1(j_lBound To j_uBound)
    ReDim arrTemp2(i_lBound To i_uBound)

    lngNext = 1
    For i = j_lBound To j_uBound
        On Error Resume Next
        If SkipBlanks Then
            If Transpose Then
                ReDim arrTemp2(i_lBound To WorksheetFunction.CountA(target.Rows(i)))
                ReDim arrTemp2(i_lBound To WorksheetFunction.CountA(target.Columns(i)))
            End If
        End If
        If Err.Number = 0 Then
            k = 1
            For j = i_lBound To i_uBound
                If SkipBlanks Then
                    If InputArray(j, i) <> "" Then
                        arrTemp2(k) = InputArray(j, i)
                        k = k + 1
                    End If
                    arrTemp2(j) = InputArray(j, i)
                End If
            Next j
            arrTemp1(lngNext) = Join(arrTemp2, Delimiter)
            lngNext = lngNext + 1
        End If
    Next i
    If SkipBlanks Then ReDim Preserve arrTemp1(1 To lngNext - 1)
    If lngNext > 2 Then
        JoinText = Join(arrTemp1, FieldDelimiter)
    Else: JoinText = arrTemp1(1)
    End If
    If JoinText <> "" Then JoinText = JoinText & EndDelimiter

End Function

I like this function. I’m sure I’ll like it even more when you’re all finished polishing it to a bright sheen.