It depends on the size of the program. I work on a a program which used to have a million lines of source code and on a 2GHz Intel MacBook Pro takes around 10 minutes to compile for debugging. Building the final version on our PowerPC server for two target architectures take a good half hour. You can imagine how long it would take to compile on my 1301 where the shortest instruction (NOP) is 12 microseconds and the longest (multiply with both operands of 12 non zero digits) is 3784 microseconds. I have recently constructed an RS232 interface for my 1301 (Just a level changer and one bit input the processor can sample, no fancy UART or SCC) and using a USB to serial converter on the Mac have connected the two. A six instruction bootstrap hand keyed into Flossie (my 1301) and a download and go high level assembler in the Mac mean I can write software whenever and wherever I happen to have a few minutes free. Back at home I can power up Flossie and usually find the core memory still contains my bootstrap, then assemble and squirt the program down the serial port at 9600 baud, which wouldn't take long, even to load the entire 2000 words (48 bits) core store. So far my biggest program is only about two hundred words. When I fill the core I'll have to start overlaying off drum or tape, or even Mac when I get serial output working too. I am hoping to retrieve software from about 100,000 punched cards via the check reading brushes of the card punch, which is much more gentle on old cards than the real card reader. Then it will be time to start on the programs on magnetic tape, but first I have to get a higher speed parallel interface working so I can read the tapes without stopping between blocks and risk stretching/breaking 40 year old tapes. I have been looking into getting the line printer working. There are 120 off 100 microfarad 100v capacitors which hold the energy required to fire the print hammer solenoids. Each on is about an inch diameter and three inches long. The modern equivalent is tiny and will not even reach between the holes in the PCB, but the leads can be extended. Of course they don't look right, maybe I should think about opening the old cans and putting the new components inside. Maybe there are harmful chemicals inside. When a transistor or diode fails I don't have the option of buying new ones, but fortunately I have ample spares of used ones of most types. Most times I can replace the whole card (four AND gates, one flip flop, one inverter and re-inverter, a clock gate or dozens of types of amplifiers/attenuators). Occasionally the fault is with the wiring or wire wrap or soldered joint. All these are very reliable but there are literally millions of them so the odd one failing is not really surprising. Core failures are a totally different problem. The official method is to use a spare row or column which has worked fine so far though long term I am worried about running out. I don't think my eyesight would be up to remaking a whole core plane but one day it might be necessary to get it done. I enjoy the intellectual task of finding the location of a logic fault but not mechanical problems, they are just something which has to be done, a pain in the backside.