Gobán Saor

In the past I’ve talked about Excel as a tool to develop Really Simple Systems. Such “systems” usually occupy the middle ground between continuing to do a task by hand or  investing time/money in using a packaged/bespoke “proper system”.

When such systems are primarily reporting in emphasis, the justification for using Excel is usually straight forward and compelling (and getting even more compelling with the appearance on the scene of Excel 2010′s  PowerPivot.) But, alarm bells sound across the world of professional IT when Excel is proposed as a “book of record” i.e. when it’s to be used to store and maintain a business critical dataset. And, with some considerable justification, the nightmare that is linked (or even worse, shared) workbooks is very real indeed. But yet, businesses continue to do so, and do so quite successfully.

I myself record my business as a series of Excel documents (Invoices, Timesheets, Expenses) in a set of folders (Financial Years subdivided into major customers).  Essentially a document-oriented database.

In the past I simply then used a VBA powered workbook to open the required “documents” and extracted the data necessary for whatever report I required (VAT, year-end etc.).  To better automate (i.e. less bespoke VBA) this task I’ve have made changes to HAMMER to help with this and with similar tasks for clients.

The following list of commands will be added to the next release of HAMMER. (In the meantime these new functions can be previewed here …)

LISTOFFILES

This command takes a single argument, the folder to search, and will return a table of files in that folder and in any sub-folders. The result can then be used to select a list of files for further processing.

Example:

=HAMMER(“C:\a\rss”,”LISTOFFILES”)

_XLTOJSONDOC

This command takes a list of workbooks, opens each one, checks for a list of named ranges and generates a JSON document. The command is intended to be called from within a VBA macro (as opening and closing workbook breaks the “no side effects” rule of UDFs). Most “_” commands such as “_MD5″ etc. are likewise intended for “programming use”, but any command beginning with “_XL” must be restricted to macro (i.e. non-UDF) use.

See the example workbook FetchInventory for an example of this function in action. The function takes one argument (the name of the document to load) and expects a table where the last column is the full name of the workbook to open. Any columns in the source table will be copied to the new “JSON document” table with an additional column called “Document” which will hold a JSON document representing the key-name pairs and table(s) extracted from the workbook.

On opening a workbook, it is searched for a named range with the name of document concatenated with “_fields” (e.g. INVENTORY_fields). The value of this range is expected to be a CSV list of fields and tables to load. A single Excel “document” could contain multiple logical documents (each specified by its own “_fields” list) .

See the PartsInventory_bin4 for an example of a multi-document workbook (INVENTORY and EXAMPLE). The EXAMPLE document in this workbook also demonstrates the various types of tables handled.

Example:

lJSONObjects = oHammer.HAMMER(“C:\a\rss\StockTake1″,”LISTOFFILES”,”Select name,fullname from table2 limit 1″,”SQL”,”INVENTORY”,”_XLTOJSONDOC”)

lReturn = oHammer.HAMMERToRange(lJSONObjects,”Sheet2!A27″)

… will output

JSONDOCVIEW

This command is where the previous commands are leading to, i.e. extracting some real information value from your documents. It converts JSON documents into Excel friendly tables. It is, in essence, a Map function as in MapReduce. In a previous example I used a Python Map and a SQL Reduce, here, both Map and Reduce are via SQL (the command uses a series of SQL commands to perform its task).

Before I describe the function let me explain why I use an intermediate JSON format. I could just extract the data directly from each document and either store directly in Excel or create tables in SQLite of Access to hold this data. And in fact, that’s what I would have done in the past (seeExcel as a document-oriented NoSQL database). Now , however, I tend to favour using a free-format (i.e. no need for a fixed database schema) structure like a JSON document, so as the source documents evolve over time (which tends to happen not just during design stages but as the system matures) this will not break older documents.

So, for example, original Invoice workbooks might not have a backing time-sheet while newer Invoices do. As long as new and old documents share a core sub-set of data fields they can continue to be analysed together.

The command takes 5 arguments and a driving table (a record so far for HAMMER commands, most have a max of two arguments). The driving table’s last column is assumed to contain the JSON document to process, columns prior to this (if any) will be output unchanged for each resulting row.

The first argument specifies the name of the “inner” table to fetch (if any). Most real life documents consist of header details (the “outer”document) and one or more tables (“inner” details). Invoices, time-sheets,  stock-takes, all tend to follow this pattern. This command will effectively join each document’s outer details to a single inner table (if more than 1 inner table, a call for each one is required).

The second (field list in SQL format) and third (SQL where predicate format) arguments specify what inner fields to extract (if blank, then all) and what restrictions to impose (if any). So “InvNo, Date”,”InvNo > 12″ would only fetch documents where the InvNo > 12 and only include the InvNo and Date fields.

The fourth and fifth arguments do the same for the outer table (i.e. Header data).

If any of the columns specified  (inner or outer) can not be found, or if the predicates (inner or outer) result in no selection, no error is returned, the document simply returns no rows. Likewise if an inner table is specified and no such table exists, then no rows are returned for that document – in other words this is not an outer join, which is not usually a problem as in most cases a “header” without detail lines is meaningless. If an outer join is required, then extract the headers (outers) and details (inner tables(s)) separately and join using SQL.

Example:

=HAMMER(“Select Name,FullName,Document from invoice_docs”,”SQL”,”table_2″, “[PART NUMBER],QTY”, “QTY >30″, “Bin_Number”, “Bin_Number > 1″, “JSONDOCVIEW”)

would result in:

For more complex JSON objects use the JSON command to incrementally parse the text or use the VBA JSON module within microETL. But for most situations (especially if you control the expected format) JSONDOCVIEW should handle it.

As JSON is fast becoming the preferred transport format for web and mobile applications having the ability to parse and produce JSON form within Excel is very useful. It is possible, for example, to use a simple web technology such as http://robla.net/jsonwidget/ to craft another type of Really Simple System. This time with the collection happening on the web (most likely using AWS S3 pre-signed forms, so no HTML server required – keep it simple) but with the control and reporting remaining within Excel (a variation on my Steam Powered Server idea).

For an example of a really simple system  download this.

Latest version of HAMMER including the above commands now released …