# Combined Call 2025-12-03 ## Agenda * Design docs and/or RFCs * [\[RFC\] How to inform users why an executable stack is required](https://discourse.llvm.org/t/rfc-how-to-inform-users-why-an-executable-stack-is-required/89007) * Currently trampolines are implemented on the stack requiring the stack to be executable and the linker emits a message about it, but this may be confusing to users * There are a few options to explore for how to improve this implementation (see RFC) * Reorganizing the table of contents for the documentation \- seems a good idea and at least one person volunteered to review PRs related to this * Should trampolines be implemented on the stack or on the heap? * Kiran: Implementing on the heap is better, but do people have the bandwidth to re-implement on the heap? * This RFC is about improving the explanation to users, not necessarily about re-implementing trampolines in an improved fashion * [\`-ffp-contract=fast\` Violates the Fortran Standard](https://discourse.llvm.org/t/ffp-contract-fast-violates-the-fortran-standard/88897) * Themos Tsikas: currently there is a disagreement about parentheses in rounding. We may need an interp request from the J3 committee. * Jean Perier: we need to account for behavior that apps rely on and provide a fallback mechanism if we make a change * [\[RFC\] Command-line compatibility with gfortran](https://discourse.llvm.org/t/rfc-command-line-compatibility-with-gfortran/88961) * Related PR: [https://github.com/llvm/llvm-project/pull/165579](https://github.com/llvm/llvm-project/pull/165579) * Consensus: we should not try to aim for command-line compatibility with gfortran, and a list exists of specific options that are wanted by users for their specific cases. Support for options will be considered on a case-by-case basis. * Compatibility with Clang \> with gfortran * PRs of Note * [\[Flang\]\[FIR\] Introduce FIRToCoreMLIR pass. \#168703](https://github.com/llvm/llvm-project/pull/168703) * Issues of Note * [Invalid code for 'where'](https://urldefense.com/v3/__https://discourse.llvm.org/t/invalid-code-for-where/88881/2__;!!Bt8fGhp8LhKGRg!AzRYeczr2-nZWMBB2mVGUwRR3M58rt_SIF-svPI6IVhIyIiSgTTDssvRaU8H0UI4kGx_9LKg4f2gkD4HaosWtAksKlal0j8r$) * FYI * CFP FOSDEM 2026 LLVM dev room * extended the submission deadline with 1 week until December 7th * [Discourse post](https://urldefense.com/v3/__https://discourse.llvm.org/t/cfp-fosdem-2026-llvm-dev-room-deadline-extended-until-december-7th/88746/2__;!!Bt8fGhp8LhKGRg!CN-bph7losNBxOURN0VbmhDaHVSVDiQSqbr3oJqWjE6O7tA4Q3yebG-T8B6JQFkQc97cWmtY5XlHyqO9n44zxlCjPvgysnnW$) * [Flang Liaison Report ot J3](https://j3-fortran.org/doc/year/25/25-184r1.txt) * [Operational Maturity Round Table Notes](https://urldefense.com/v3/__https://discourse.llvm.org/t/operational-maturity-round-table-notes/88898/4__;!!Bt8fGhp8LhKGRg!DBCD7MXLIlKDQMxYC24V2tf9xKpQF7PD3s92rD_AqokuSQOdNeDC7vxesj3En2jeZJlK5nCu2wRFTbe3WtjJCi-o4hGTHJtt$) * Possible switch to PR-only mechanism * Other topics as time allows * Proposing an Interactive Fortran Workflow with Flang using Jupyter Notebooks * Anutosh Bhat \- Fortran ecosystem lacks a modern interactive workflow * Something is available using LFortran, but it’s not perfect: [https://github.com/lfortran/lfortran/blob/main/share/lfortran/nb/Demo1.ipynb](https://github.com/lfortran/lfortran/blob/main/share/lfortran/nb/Demo1.ipynb) * Flang seems a natural place to bring this into being * [https://compiler-research.org/xeus-cpp-wasm/lab/index.html](https://compiler-research.org/xeus-cpp-wasm/lab/index.html) * There is clang-repl for this with C++ * RFC coming * [OpenMP notebook demo](https://github.com/compiler-research/xeus-cpp/blob/da0cc730b175c2dd880d16845187e68917afc9f5/notebooks/openmp-notebooks/openmp-demo.ipynb) * [CUDA notebook demo](https://github.com/compiler-research/xeus-clang-repl/blob/main/notebooks/kalman_CUDA_demo/run_kf.ipynb) * [prototype-debug.mp4](https://drive.google.com/file/d/1KxMpHz7njRTb2d1FSTPumZXfgoNRHhbM/view) * [FOSDEM 2025 \- O\_o \[ Flang \+ WASM \] o\_O](https://archive.fosdem.org/2025/schedule/event/fosdem-2025-5202-oo-flang-wasm-oo/) * Given the issues with the Call Notes document, this may be a good time to migrate to a Markdown document located in the repo * ## Details * Consists of over **704,000** lines of code, documentation, build files, and test * To date, over **11,430** commits have been made to Flang