Editing EDIF (section)

=== Problems with 2 0 0 ===

{{Unreferenced section|date=October 2007}}

To understand the problems users and vendors encountered with EDIF 2 0 0, one first has to picture all the elements and dynamics of the electronics industry. The people who needed this standard were mainly design engineers, who worked for companies whose size ranged from a house garage to multi-billion dollar facilities with thousands of engineers. These engineers worked mainly from schematics and netlists in the late 1980s, and the big push was to generate the netlists from the schematics automatically. The first suppliers were Electronic Design Automation vendors (e.g., Daisy, Mentor, and Valid formed the earliest predominating set). These companies competed vigorously for their shares of this market.

One of the tactics used by these companies to "capture" their customers was their proprietary databases. Each had special features that the others did not. Once a decision was made to use a particular vendor's software to enter a design, the customer was ever after constrained to use no other software. To move from vendor A's to vendor B's systems usually meant a very expensive re-entry of almost all design data by hand into the new system. This expense of "migration" was the main factor that locked design engineers into using a single vendor.

But the "customers" had a different desire. They saw immediately that while vendor A might have a really nice analog simulation environment, vendor B had a much better PCB or silicon layout auto-router. And they wished that they could pick and choose amongst the different vendors.

EDIF was mainly supported by the electronics design end-users, and their companies. The EDA vendors were involved also, but their motivation was more along the lines of wanting to not alienate their customers. Most of the EDA vendors produced EDIF 2 0 0 translators, but they were definitely more interested in generating high-quality EDIF readers, and they had absolutely no motivation at all to write any software that generated EDIF (an EDIF Writer), beyond threats from customers of mass migration to another vendor's software.

The result was rather interesting. Hardly any software vendor wrote EDIF 2 0 0 output that did not have severe violations of syntax or semantics. The semantics were just loose enough that there might be several ways to describe the same data. This began to be known as "flavors" of EDIF. The vendor companies did not always feel it important to allocate many resources to EDIF products, even if they sold a large number of them. There were several stories of active products with virtually no-one to maintain them for years. User complaints were merely gathered and prioritized. The harder it became to export customer data to EDIF, the more the vendors seemed to like it. Those who did write EDIF translators found they spent a huge amount of time and effort on generating sufficiently powerful, forgiving, artificially intelligent readers, that could handle and piece together the poor-quality code produced by the extant EDIF 2 0 0 writers of the day.

In designing EDIF 3 0 0, the committees were well aware of the faults of the language, the calumny heaped on EDIF 2 0 0 by the vendors and the frustration of the end users. So, to tighten the semantics of the language, and provide a more formal description of the standard, the revolutionary approach was taken to provide an information model for EDIF, in the information modeling language [[EXPRESS (data modeling language)|EXPRESS]]. This helped to better document the standard, but was done more as an afterthought, as the syntax crafting was done independently of the model, instead of being generated from the model. Also, even though the standard says that if the syntax and model disagree, the model is the standard, this is not the case in practice. The [[Backus-Naur form|BNF]] description of the syntax is the foundation of the language inasmuch as the software that does the day-to-day work of producing design descriptions is based on a fixed syntax. The information model also suffered from the fact that it was not (and is not) ideally suited to describing EDIF. It does not describe such concepts as name spaces very well at all, and the differences between a definition and a reference is not clearly describable either. Also, the constructs in EXPRESS for describing constraints might be formal, but constraint description is a fairly complicated matter at times. So, most constraints ended up just being described as comments. Most of the others became elaborate formal descriptions which most readers will never be able to decipher, and therefore may not stand up to automated debugging/compiling, just as a program might look good in review, but a compiler might find some interesting errors, and actually running the program written might find even more interesting errors. (Additionally, analogous EXPRESS compilers/executors didn't exist when the standard was written, and may not still exist today!)