Authentication-Results: minnie.tuhs.org; dkim=fail reason="signature verification failed" (2048-bit key; unprotected) header.d=dartmouth.edu header.i=@dartmouth.edu header.b="WXhavXnT"; dkim-atps=neutral From: Douglas McIlroy Date: Wed, 12 Jan 2022 23:23:23 -0500 To: TUHS main list Subject: [TUHS] Fwd: struct(1) revived! And a request for help Cc: Brenda Baker Errors-To: tuhs-bounces@minnie.tuhs.org Sender: "TUHS" Arnold, I'm very glad you have revived struct. It is an important historical artifact, and it's nice to have it freed from distracting obsolete usage. To my mind, struct ranks among the top accomplishments of our department at Bell Labs. It was also a lesson in humility to me. When Brenda proposed struct, it was obvious that it could be built, but I advised against doing so. I thought it would take an endless pile of special cases to generate respectable Ratfor. She demurred, saying that she thought she could do better than that. And she was right. She produced not only working Ratfor programs, but a canonical-form theorem that distinguished those programs from all the ad hoc alternatives I had imagined. The value of the theorem was attested by users' reports that the derived Ratfort was easier to understand than the Fortran they had written themselves. Doug ---------- Forwarded message --------- From: Date: Wed, Jan 12, 2022 at 7:58 AM Subject: [TUHS] struct(1) revived! And a request for help To: Hello All. We recently discussed Brenda Baker's struct program, that read Fortran and generated Ratfor. Many of us remarked as to what a really cool program it was and how much we admired it, myself included. For fun (for some definition of "fun") I decided to try to bring the code into the present. I set up a GitHub repo with the V7, V8 and V10 code, and then started work in a separate branch. (https://github.com/arnoldrobbins/struct, branch "modernize".) The program has three main parts: - structure, which reads Fortran and outputs something that is almost Ratfor on standard output. - beautify, which reads the output of structure and finishes the job, primarily making conditions readable (.not. --> !, removing double negatives, etc.) - struct.sh - a simple shell script that runs the above two components. This is what the user invokes. The code was written in 1974. As such, it is rife with "type punning" between int, int *, int **, and char *. These produce a lot of warnings from a modern day C compiler. The code also uses a, er, "unique" bracing style, making it nearly illegible to my stuck-in-a-rut programming brain. Here is what I've accomplished so far: * Converted every function definition and declaration to use modern (ANSI) C style, adding a header file with function declarations that is included everywhere. * Run all the code through the indent program, formatting it as traditional K&R bracing style, with tabs. * Fixed some macros to use modern style for getting parameter values as strings into the macros. * Fixed a few small bugs here and there. * Fixed beautify to work with modern byacc/bison (%union) and to work with flex instead of lex. This latter was a challenge. In structure, only three files still generate warnings, but they all relate to integer <--> pointer assignment / use as. However, when compiled in 32 bit mode (gcc -m32), where sizeof(int) is the same as sizeof(pointer), despite the warnings, structure works!! Beautify works, whether compiled in 32 or 64 bit mode. What I've done so far has been mostly mechanical. I hereby request help from anyone who's interested in making progress on "the hard part" --- those three files that still generate warnings. I think the right way to go is to replace int's with a union that holds and int, a char* and an int*. But I have not had the quiet time to dive into the code to see if this can be done. Anyone who has some time to devote to this and is interested, please drop me a note off-list. Thanks, Arnold Robbins