Welcome to the Datatron 205 and 220 Blog

This blog is a companion to T J Sawyer's Web Page that outlines the history of the Burroughs Datatron 205 and Paul Kimpel's incredible 205 and 220 emulators. Please visit those sites and return here to post any comments.

Tuesday, February 11, 2020

retro-205 Emulator Version 1.02 Released

The retro-205 emulator for the ElectroData/Burroughs Datatron 205 has been stable and operating nicely now for almost three years. I occasionally find small problems to fix, or come across a reference that sheds new light on the operation of the system that prompts a programming change. Enough of these have accumulated that I recently pushed out a new release, version 1.02.

This post describes the changes in 1.02. These changes have been committed to the project's source repository on GitHub and uploaded to the project's hosting site from which you can run the emulator. The project wiki also has a few minor updates.

Application Cache Feature Removed


Support for the browser Application Cache ("appcache") feature has been removed from the emulator. This feature allowed the browser to load the entire emulator into local storage, and once the code was loaded, run it without requiring access to the host web site, or even to a network connection. Alas, this feature was not well-accepted by web developers and was deprecated as a standard a few years ago. Some browsers may no longer support it.

The impact of this on most users should range from minor to none. The browser will still require network access to the host web site, either to load the code or verify that the copies in its cache are still current. Once loaded into the browser and running, the emulator requires no further access to the host site, as before. If you want to run the emulator from a location that cannot access your host site, a possible option is to set up a web server on your local workstation. Only a very simple web server is needed. See "Setting Up a Web Server" in the Getting Started wiki for more information.

Floating Point Issues Corrected


I came across a set of floating-point diagnostic test routines at the Charles Babbage Institute (CBI) during a visit in October 2019. These looked interesting, so I transcribed them and got them to run in the retro-205 emulator. That exercise uncovered one bug in the emulator's implementation of floating-point arithmetic, and one aspect of the implementation that was incorrect, but relatively benign.
  • Floating Divide (FDIV, 83) should set the Overflow Toggle when dividing by zero. It was not doing this, however, when the mantissas of both the divisor and dividend were zero. This bug was due to testing the mantissas of the two operands for zero in the wrong order.
  • The emulator included code to pre-normalize the operands, i.e., shift the mantissa left and decrement the exponent until the high-order digit was non-zero. This was code carried over from the retro-b5500 emulator, which I used as a base for the implementation of floating point in retro-205. I discovered that the 205 always assumed the operands were normalized and did not adjust them before commencing an operation. All pre-normalization code has therefore been removed from the emulator.
The floating-point diagnostic routines are interesting, and I plan a future blog post to discuss them in more detail.

Breakpoint Switch on Control Console Corrected


In the process of getting the floating-point diagnostic routines to work, I discovered that the Breakpoint switch on the Control Console was not working correctly. Setting the switch to the "4" position did not cause the processor to halt after it executed an instruction having the 4-bit set in the fourth digit of the instruction word. Having the switch in the "2" position worked correctly. Having it in the "1" position also worked, but only accidentally -- a case of two errors canceling each other out. This bug was introduced version 0.04e (July 2015) as part of the change that caused the Breakpoint switch to reverse when it was clicked at either end of its travel. It is now fixed.

Rogue Paper-Tape Reader Brought to Heel


During the floating-point routine testing, I also discovered that the paper-tape reader stayed "hot" after clearing the system. If the device was hanging on a read command, waiting for a paper-tape image file to be loaded, then even though the processor had been halted and the system cleared, the reader would immediately start reading when a paper-tape file was loaded into it. It now detects the system clear operation and does not start reading the tape until specifically commanded to do so.

External Control Instruction Corrected


The External Control instruction (EXC, 71) was implemented in the 205 as a potential means for the system to control external equipment. There were eight external switches in the Model 421 External Switching Unit that could be set from the high-order eight digits (not including sign) of the memory word fetched by the instruction's operand address. See the DATATRON Command List (Bulletin 3030C, which follows the DATATRON 204 Central Computer Handbook at the link) for details.

The change in this release is a detail that Bulletin 3030C did not discuss, viz, if the low-order bit of the operand word's sign bit is set, the computer will halt in the same manner as for the Breakpoint switch, i.e., at the end of the instruction. This detail was discovered in a copy of Bulletin 3031, External Switching Unit Model 421 and Output Selector Unit Model 420 (Preliminary Edition), ElectroData Division of Burroughs Corporation, 1956.

Note that although the emulator implements the EXC instruction, it presently doesn't do anything except set flags internally in the Processor corresponding to the switches. In the future it could be possible to use these flags to control some new mechanism.

Cardatron Reader Format Lockout Corrected


The Cardatron interface to IBM punched-card equipment had a feature when reading cards termed "format lockout" that would prohibit the next card from being read automatically. It was triggered by an 8-punch on the card in the column used to determine which format band would convert the card punches to 205 memory words. Since the emulator uses ordinary ASCII text files as card decks, the concept of an "8 punch" does not translate well, and instead the emulator uses specific ASCII characters to encode the 8-punch along with another punch in rows 1-7 that indicates the format band to be selected.

Two of those ASCII characters were not represented properly in the code, the grave accent (`) for the 1-8 punch and the apostrophe (') for the 5-8 punch. This problem has been corrected. See the Cardatron wiki for the full list of format lockout codes.

No comments: