Derek Atkins <email@example.com> 2004-01-07
A work in progress....
The existing qif importer in src/import-export/qif-import is both hard to maintain and hard to re-integrate into the shared import architecture. Similarly, the half-completed re-write in qif-io-core is similarly hard to maintain (although it is arguably easier to integrate). One problem with both of these solutions is that they are written in Scheme, a language that many gnucash developers just don't understand well. Another issue is that the code is not commented and no documentation exists to help future developers track down bugs or extend the importer as QIF changes over time (c.f. Memorized Transaction import).
As much as "complete rewrite" tends to be a lot of work for little gain, when few (if any) developers can understand the implementation well enough to make changes, a complete re-write may make sense. This document is an attempt to describe the architecture of the new importer, implemented in C, and how it interfaces to the rest of the import infrastructure.
The importer is a multi-step, staged system that should implement a read, parse, convert, combine, filter, finish process. The importer starts with a clean import context and then each processing step modifies it as per the user's requirements. A small set of APIs allow the user to progress along the processing steps (and an internal state machine makes sure the caller proceeds in the proper order).
The importer is driven by the UI code; the importer itself is just a multi-stage worker. The UI code calls each step in the process. For long-running operations the UI can provide a callback mechanism for a progress bar of completion.
Each stage of the import process may require some user input. What input is required depends on the stage of the process and what the last stage returned. In some cases stages can be skipped. For example, during the conversion phase if the date format is unambigious then no user input would be required and the "ask for date format disamiguation" input can be skipped.
QUESTION: How does the importer relate the processing state back to the UI? Simiarly, how does it pass back specific disambiguating questions to ask the user (and how are those responses returned to the importer)?
The import process starts when the UI creates a new import context. All of a single import is performed within that context. The context model allows multiple import processes to take place simultaneously.
The first step in the import process is selecting the file (or files) to be imported. The UI passes each filename to the importer which reads the file and performs a quick parse process to break the file down into its component QIF parts. While the importer should allow the user to iteratively add more and more files to the import context, it should also allow the user to select multiple files at once (e.g. *.qif) to reduce the user workload.
Each imported file may be a complete QIF file or it may be a single QIF account file. In the latter case the UI needs to ask the user for the actual QIF account name for the file. Similarly, each file may need user intervention to disambiguate various data, like the date or number formats.
QUESTION: If the user provides multiple files at once and each file has internal ambiguities (e.g. the date format), should the user be asked for each file, or can we assume that all the files have the same format? Perhaps the UI should allow the user to "make this choice for all files"?
Once the user chooses all their files (they can also remove files during the process) the importer will combine the files into a common import, trying to match QIF accounts and transactions from different files. Part of this is duplicate detection, because QIF only includes half a transaction (for any QIF transaction you only know the local account, not necessarily the "far" account). If the importer sees multiple parts of the same transaction it can (and should) combine them into a single transaction, thereby pinning down the near and far accounts.
The next series of steps maps QIF data objects to GnuCash data objects. In particular, the importer needs the help of the UI to map unknown QIF Accounts and Categories to GnuCash Accounts (the latter to Income and Expense Accounts) and QIF Securities to GnuCash Commodities. Finally the importer can use the generic transaction matcher to map the existing transactions to potential duplicates and also Payee/Memo fields to "destination accounts".
At the end of this process the accounts, commodities, and transactions are merged into the existing account tree, and the import context is freed.
QifContext QifError QifFile QifObject +-QifAccount +-QifCategory +-QifClass +-QifSecurity +-QifTxn +-QifInvstTxn
/** Create and destroy an import context */ QifContext qif_context_create(void) void qif_context_destroy(QifContext ctx) /** return the list of QifFiles in the context. */ GList *qif_context_get_files(QifContext ctx) /** merge all the files in the context up into the context, finding * matched accounts and transactions, so everything is working off the * same set of objects within the context. */ void qif_context_merge_files(QifContext ctx);
/** * Open, read, and minimally parse the QIF file, filename. * If progress is non-NULL, will call progress with pg_arg and a value from * 0.0 to 1.0 that indicates the percentage of the file read and parsed. * Returns the new QifFile or NULL if there was some failure during the process. */ QifFile qif_file_new(QifContext ctx, const char* filename, void(*progress)(gpointer, double), gpointer pg_arg) /** removes file from the list of active files in the import context. */ void qif_file_remove(QifContext ctx, QifFile file) /** Return the filename of the QIF file */ const char * qif_file_filename(QifFile file); /** Does a file need a default QIF account? */ gboolean qif_file_needs_account(QifFile file); /** Provide a default QIF Account-name for the QIF File */ void qif_file_set_default_account(QifFile file, const char *acct_name); /** Parse the qif file values; may require some callbacks to let the * user choose from ambigious data formats * XXX: Is there a better way to callback from here? * Do we need progress-bar info here? */ QifError qif_file_parse(QifFile ctx, gpointer ui_arg)
The state machine has the following structure. Named states (and substates) must proceed in order. Some states (e.g. state b) have multiple choices. For example you could enter substates b1-b2, or you can run qif_file_remove.
a. Create the context