Journal: Legacy Systems

I've spent a number of full weekends (you know the type - Friday after work 'til 8AM, Sat Noon-5AM, Sunday 10AM to 3AM) updating some legacy DOS software.

The software is a modem server that allows remote devices to dialin & send data using a proprietary format. The data is saved in a local, proprietary flat-file database and reports are dumped to a printer on the parallel port.

The new requirements were:

1. It needed to continue to run under DOS (there were specific val/ver/cert issues that prevented extensive changes)
2. It needed to work with current UARTs & modems.
3. It needed to be network/web enabled.

#3 was addressed by buying an Ultra 10 & installing a Sun PC/PCI card (700MHz AMD PC on a card). This allowed the existing system to run in DOS (FreeDOS) while giving Unix access to its data files. On the Unix side, cron jobs run to grab data updates & ship them off to the web/database system.

This left 1&2. The software was extremely flakely with the latest com hardware & modems. In addition, #3 still had some problems for unknown reasons...

So step 1 was to get the source code. After much prodding & pushing this was accomplished. I had already figured out from running 'strings' on the .exe that it was Borland C based & I have TurboC 2.0 & BC5 already - that shouldn't be a problem. Once I got the source code in my hands, I realized I needed Borland C++ 3.1 [unless I wanted to go through a lot of pain getting it to work with BC5]. So where in the hell was I going to get that?!

Fortunately, it seems to be pretty readily available as bc31.zip...

After a lot of dicking around, I finally got a stable build of the original source. It took me all of an hour of code changes to break the damn thing. More digging around & I remembered those nasty old Borland OVERLAYS...yuck. Okay, another couple of hours & I had the thing working again.

I ripped out all of the printer support (predominately because it used a proprietary set of libraries that didn't come with the source & that I couldn't dig up).

Step 2 - get the com ports working. Fortunately, a nice package came to the rescue in the form of pmcom. Works with Borland & DJGPP. A bunch of changes to hook into the new libs & I've got basic modem com working!! Another big WhoHoo!

Step 3 - get this thing working properly on the Sun PC/PCI card. This mostly consisted of cleaning up some file system call problems in the original code. I also dropped support for writing to that big, nasty flat-file system. Each report is instead written into an ASCII file that is parsed on the web/database side & shoved into a MySQL database.

The rest of the work was cleanup, bug fixes, and the website itself. Did all of the website with Apache+PHP & MySQL. All of the data import was handled with Perl.

In addition to the fact that the pay was good, this was really a fun project - kind've like taking an old car & getting it running again...with the constraint that it has to be someone's primary car when it's done & time is of the essence. Not often I get to really put those old 'get it done right AND get it done right NOW' skills to work...

I suppose it would be nice to finish this up with a fine treatise on how I did all of the development using Linux + a PC Emulator running FreeDOS or VirtualPC on my Mac...

In reality, I booted my Linux box to Win2k & did everything with GVim, edit, and CMD. Although I did install the cygnus utils so I'd have a reasonable grep...