Is Fortran “A Dead Language”? (2022)

Original Article ↗ Hacker News Discussion ↗

Fortran’s status as a “dead language” is contested: while largely absent from mainstream software jobs and general-purpose development, it remains heavily used and actively maintained in high‑performance scientific and engineering computing, especially in fields like fluid dynamics, climate modeling, and nuclear simulation. Commenters note that modern Fortran offers powerful native support for numerical arrays and long-term code stability, but lacks the rich standard libraries, tooling, and ecosystem found in languages like Python and C++. Many argue it now functions as a de facto domain‑specific language for HPC, with new large projects increasingly gravitating toward C++ or other languages for better interoperability and GPU support, even as vast Fortran codebases continue to underpin critical infrastructure.

Is Fortran a “dead language”?

  • Many argue it is not dead but highly niche: very active in scientific and engineering HPC, mostly absent elsewhere.
  • A proposed definition of “dead” is “no new non‑hobby codebases”; by that standard, several posters say Fortran is still alive, others say outside scientific computing it is effectively dead.
  • Betteridge’s law is invoked: a headline asking if it’s dead implies “no”.

Primary Domains and Legacy Code

  • Widely reported use in: weather and climate models, CFD, nuclear engineering, materials, electromagnetics, crash simulation, combustion, finance-related numerics (via BLAS/LAPACK), and many other simulation codes.
  • BLAS/LAPACK and many foundational numeric libraries were written in Fortran and are still central to modern stacks (often via C wrappers).
  • Large legacy codes with decades of validation make rewrites risky and expensive; long‑term reproducibility and bitwise‑stable outputs matter in aerospace, nuclear, and other high‑consequence domains.

Language Features, Strengths, and Weaknesses

  • Strengths: native multidimensional arrays, vectorized operations, complex numbers, strong compiler optimizations, OpenMP support, and long‑term stability. Some compare array ergonomics favorably to NumPy.
  • Modern Fortran (2003–2018/2023) adds many conveniences; slicing is supported, limited “broadcasting” via elemental functions and array/scalar ops.
  • Weaknesses: extremely thin standard library (no built‑in networking, threads, rich data structures), poor string handling, sparse higher‑level abstractions; many tasks feel like “C without libc”.
  • Some say overall developer experience is “awful” compared to Python; others strongly disagree and find Fortran simple and pleasant for numerics.

Tooling, Compilers, and Ecosystem

  • Multiple active compilers (gfortran, Intel, LLVM flang, vendor compilers); some commercial compilers and small vendors have disappeared.
  • Complaints about buggy free compilers and 1990s‑era tooling; others highlight ongoing work to improve compilers and tests.

Jobs, Education, and Maintainability

  • General job boards show almost no Fortran roles compared to mainstream languages; many Fortran jobs are in research labs or specialized engineering firms and recruit via academic networks.
  • Some national HPC centers still teach modern Fortran; elsewhere, Python/C++ dominate training and new projects, with Fortran gradually phased out in some labs for maintainability and hiring reasons.

GPUs and Future Direction

  • Porting Fortran codes to accelerators is described as hard and fragmented (CUDA Fortran, partial OpenMP 5 support); this pushes new HPC projects toward C++.
  • Some optimism that OpenMP, unified memory on new hardware, and maturing flang will improve accelerator portability, but current state is mixed.