It sounds like excuses I hear from junior devs that don’t want to take time to learn existing frameworks and systems.

Juniors just label anything that requires putting effort to understand as „bad”.

There was insane growth in junior dev numbers last 3 decades. It is not like „no one knows”, it is more like there is much more people who don’t know.

I disagree. I think that after 1992, we got memory safe languages that brought a meaningful improvement to the status quo. And after 2015, we've got low-level memory safe languages (Rust, as the major example. There are others, more experimental.)

The average programmer doesn't get better – if anything, we might be getting worse, because the tools allow us to, and the capitalist reality doesn't optimize for great programs or programmers but for more money.

But, at least, the tools are way better than in 1992, and I think we, as a collective profession, have learned a thing or two.

IBM has historically been heavily influenced by petty politics. Legacy programs like DOORS (an acquisition, but developed largely afterwards) continue to use UI patterns unlike any other Windows products, which I attribute to their legendary humiliations by Gates.

Letting pride outvote usability is an insane business decision.

I think significant improvements are

- not writing compilers in assembly

- not requiring overlays

- knowing how previous compilers produced fast code (Web search doesn’t give me conclusive answers, but that Fortran compiler may have been the first to do loop unrolling and common subexpression elimination)

- having way more memory, CPU and disk available

- possibly: spending less time looking at optimizations. I expect IBM tried hard to make the output of their compiler to match the performance of hand-written assembly

The best link I could find is https://en.wikipedia.org/wiki/Fortran#FORTRAN_IV:

“In particular, the FORTRAN H compiler played an important role in the development of certain kinds of optimization approaches, such as allocating a specific set of registers to hold the values of variables while in a loop. Overall, the compiler had three levels of possible optimization, as Fortran compiler developers had learned early on that the ability to turn off optimization was a necessity, since it drove up compilation times considerability for program runs that often were not going to work anyway. Even with the larger amount of main memory available to it, the FORTRAN H compiler was still organized via a number of overlays.”

The author is comparing a 1990 hypothetical compiler to a 1970-ish compiler. The late 1960s and early 1970s are essentially when all of the foundational parser theory gets laid down. By the late 1970s, we're getting into autoparallelization and autovectorization research. Monotone dataflow analysis was developed in the 1970s as well. To be a little bit glib, basically what happened is compiler theory is really birthed starting in the 1970s; if you wanted to track down most of the techniques in the Dragon book, I suspect the vast majority of them originate in that timeframe.

There is a second shift that occurs around 2000-2005-ish, which is the transition of optimizing compilers from an instruction-based semantics to a more value-based semantics, in that modern optimizers make no real attempt or guarantee to preserve the structure of code. For example, an if statement may happily be converted into an expression lacking an if entirely.

I think the reason writing a compiler is easy today is the theory I learned in compilers class. How to do context free grammars, the concept of abstract syntax trees, the pattern of writing a recursive descent parser and having a lexer that only looks one symbol ahead and has a peek function. On top of that we have experience with lots of languages and type systems to draw from when constructing a new one.

I was just doing some research and apparently all of this stuff was invented around the late 60s and so in the 70s it was still new and by the 90s it was standard practice. The dragon book came out in 1986 and spelled it all out in one place.

Today we have the benefit of knowing the right ideas to use from the start and confidence that if you follow the formula it will all work out.

Was going to make a similar comment…most systems programming was in PL/S or PL/X on 370/390 architecture (regardless of the O/S). AIX and OS/2 were mostly in C. AS/400 in RPG. There were some oddball programs in APL. And thousands of internal "tools" in Rexx.

> That bottleneck is unfortunately long gone

? You are pro-bottleneck?

1. For the most common things people will write, we have a plethora of asymptotically optimal choices that have been discovered.

2. Consider any algorithm of roughly linear complexity (this probably applies to N lg N as well): the only way to make it significantly faster now is to improve parallelism (whether making it so the CPU can exploit ILP, running multicore, or running on a GPU). In 1992, you could make things run significantly faster by just waiting until it was 1994.

    In the late 90s a couple of companies, including Microsoft and Apple, noticed (just a little bit sooner than anyone
    else) that Moore's Law meant that they shouldn't think too hard about performance and memory usage... just build
    cool stuff, and wait for the hardware to catch up. Microsoft first shipped Excel for Windows when 80386s were too
    expensive to buy, but they were patient. Within a couple of years, the 80386SX came out, and anybody who could
    afford a $1500 clone could run Excel.

    As a programmer, thanks to plummeting memory prices, and CPU speeds doubling every year, you had a choice. You
    could spend six months rewriting your inner loops in Assembler, or take six months off to play drums in a rock and
    roll band, and in either case, your program would run faster. Assembler programmers don't have groupies.

    So, we don't care about performance or optimization much anymore.

                                                                             — Joel Spolsky, "Strategy Letter VI" (2007)

I still think about it sometimes, but I don’t think it matters nearly as much nowadays as it did back in the day. Oftentimes I’m working in situations where the all the complicating factors that big O explicitly excludes matter much more now than they used to. Partially because they’re relatively larger (e.g., cost of memory access) and partially because they’ve become variable in a way that they weren’t 30+ years ago (e.g., the cost of a conditional branch instruction).

Also to some extent it’s just that we’ve pretty well standardized on algorithm implementations. Thinking about the relative merits of a BST vs a red-black tree vs a hash table with bucketing or open addressing or whatever just doesn’t happen as often when the standard library has one implementation and not choosing it would cost you a week of implementing testing and justifying the decision to your colleagues.

It is fun when debugging slow code and you see a for loop in a for loop. Feels rewarding to sort that out.

Big O doesn't capture parallelism well enough and that's really been the push since Moore's Law started to hit diminishing returns on single threaded perf.

for advanced undergrads, the university of michigan (i'm speaking from my own experience here) has a course progression of (1) programming languages for 1st year grad students, (2) compilers for undergraduates, (3) compiler optimization for 1st year grad students. the expectation is that you can write a fast implementation of a simple language (~scheme) on your own, and you have the knowledge to write a simple optimizing compiler for e.g. c89/99 if you were to take it on as a longer project. any compiler written in a semester probably doesn't have enough manpower to be optimized to a point of usefulness, and you'll learn some things only via implementation struggles, but you'll have the capability to try, search for literature to get unstuck, etc. which is cool!

Languages were simpler (except for certain ones, like C++ which was a beast even in 1992), and incredibly complex and magical optimizers weren’t yet a thing, never mind a feature expected of a "passable" compiler. One could still write a reasonable non-optimizing Pascal or C89 compiler in a weekend more or less, and it would be both faster to write (thanks to more expressive languages) and faster at compiling (thanks to itself being compiled by an optimizing compiler) than in 1992!

> The definition of "passable compiler" in 1992 must have been very different from what it is today;

It was.

But they did invest billions in a super-Opus, which they called Mythos.

This article is fake.

Intel Fortran Compiler and IBM XL Fortran compilers are still developed and very well funded

This submission was made two days earlier than the one you linked to, it just came back through the second chance pool. Not a dupe.