Re: [TUHS] Recreation of the PDP-7 UNIX TMG compiler compiler

I was deeply motivated by TMG. The good news is that you could say what you wanted and it just did it. The bad news was the error handling. Because it was recursive if you made a syntax error the program backtracked and tried again. And again. And again. And eventually was back at the first character of the input with nowhere to go. So it issued one of its two messages -- "Syntax Error". No hint as to where it was. The other message was something like "Internal Error: call X1234". It very much encouraged a "write one line and compile" approach. It was my moaning about this in Aho's presence that got us started on LR parsing and led to Yacc. There was actually a similar problem with Yacc. I made an attempt to recover from syntax errors so I could give more than one useful message from Yacc. But there was a lot of cross-checking in the various modules, and the result was usually an internal error that I detected and reported. I naively assumed that if there was a syntax error the user would fix that and rerun. But instead, there was a steady stream of people coming in to tell me about these fatal internal errors in Yacc. When I realized that 95% of those internal errors were triggered by syntax errors, I set a flag when there was a syntax error and if the flag was on I changed the internal error message to "Cannot recover from earlier errors." About a week later, I gave a talk on Yacc and several people mentioned that Yacc used to "crash" all the time, but in the last week it had become much more stable... --- On 2021-10-14 09:53, Douglas McIlroy wrote: