How did people code entire machine language programs in hex?

You get a piece of graph paper and do exactly what the assembler would do by hand.

First you write the program out in assembly language instructions then you convert individual instructions and keep track of the program counter on the left hand side, you will have to leave space for the operands of forward jumps because you don't know what that address will be be. Then you go back and fill them in. Easy-peasy.

There is a nice image of such a notebook here

https://www.quora.com/What-is-the-real-purpose-of-graph-pape...

which looks exactly like the listing output of an assembler because it is.

As someone who did this a lot on various machines back in the day, I'm tempted to say "very carefully and methodically" (which is the truth). The more complete answer is what PaulHoule said. It's called "hand assembly".

In any case, none of this was particularly difficult or unusual, just a bit tedious. At least, the part where you enter hex codes is.

Even more fun was before microcomputers. On a few minicomputers and mainframes, you would occasionally have to enter short programs in binary using switches on the front panel.

I started my career as a field engineer (basically a computer mechanic) on mainframes and minicomputers 47 years ago. For the first 13 years, almost every computer I worked on had front panel switches and lights which was great for writing and debugging tiny diagnostic programs. I wrote those diags directly in machine code because I had the CPU instruction manuals, no access to an assembler, and a fair amount of free time when I worked a midnight shift on call. I'd sit at a desk, writing the machine code on lined paper because the sites I worked at rarely had graph paper in the supply cabinet. At first I had to look up the op codes for the instructions I wanted but I was able to memorize the ones I used most often which made it a quicker process. The fun part was using the toggle switches to enter those instructions and finally to start it running.

After 6 years of being a field engineer and writing exclusively in machine language, I made the switch to being a systems analyst which gave me access to an assembler. That seemed easy after having to calculate relative jumps/subroutine calls for so long.

This is what I loved about the PDP-11 series. The instruction opcodes were so painfully obvious you could read them in octal.

I would use a hand punch to create holes in paper tape that was the program. (Just a metal block with holes drilled and a metal dowel to punch the tape).

"Patching the tape" involved finding a section of code that could be "overpunched" to be a jump instruction to the current end of tape where you could punch in a subroutine and a return statement.

If all else failed you'd create a "patch tape" that was read after the original tape and overwrote memory locations.

It was all very convenient since you could "single step" the program from the front panel and watch the registers change.

Life got really sweet when the new teletype had a reader/punch.

Good times.