Feature/dataframe cache refactor

.github/workflows/codestyle-python-flake8.yml

@@ -12,7 +12,7 @@ jobs:
       - uses: actions/setup-python@v4
         with:
           python-version: "3.11"
-      - name: Install Flake8
-        run: pip install flake8
+      - name: Install dependencies
+        run: pip install -r dev-requirements.txt
       - name: Run Flake8
         run: flake8 python tests/python tests/fixtures

.github/workflows/unittests-python.yml

@@ -13,6 +13,6 @@ jobs:
         with:
           python-version: "3.11"
       - name: Install dependencies
-        run: pip install pytest
+        run: pip install -r dev-requirements.txt
       - name: Run pytest
         run: pytest tests/python

.github/workflows/unittests-vim.yml

@@ -12,10 +12,16 @@ jobs:
       - name: Install Vim and Neovim
         run: |
           sudo apt-get update
-          sudo apt-get install -y vim neovim
+          sudo apt-get install -y vim neovim python3-venv
+      - name: Install Python dependencies
+        run: |
+          python3 -m venv .venv
+          .venv/bin/pip install -r dev-requirements.txt
       - name: Run Vimscript tests (Neovim then Vim)
         run: |
           set +e
+          export VIRTUAL_ENV="$PWD/.venv"
+          export PATH="$VIRTUAL_ENV/bin:$PATH"
           rm -f nvim-test.log vim-test.log
 
           echo

CHANGELOG.md

@@ -0,0 +1,41 @@
+# Changelog
+
+## [1.1.0] - 2026-06-20
+
+### Added
+- Added a Python session cache for experiments, run summaries, run details,
+  metric histories, and artifact trees to avoid re-querying MLflow on every
+  redraw.
+- Added `g:vim_mlflow_runs_cache_mode` with `selected_expt` and `all_expts`
+  modes for controlling run-summary cache scope.
+- Added `g:vim_mlflow_plotpane_pct` to control the plot-versus-artifact height
+  split when both panes are visible.
+- Added `pandas` as an explicit dependency and expanded automated coverage for
+  cache behavior, viewer layout, and artifact rendering.
+- Added pretty-printed display for valid `.json` artifacts in the scratch
+  viewer buffer, with raw-text fallback for invalid JSON.
+
+### Changed
+- Refactored the main `__MLflow__` sidebar and `__MLflowRuns__` comparison view
+  to render from cached in-memory data instead of repeatedly rebuilding from
+  fresh MLflow API calls.
+- Changed the viewer layout to a shared opposite-side viewer column with plots
+  stacked above artifacts.
+- Changed the runs-window undo-layout mapping from `<C-u>` to `u`, with a
+  confirmation prompt before clearing all column layout tweaks.
+- Made vim-mlflow-managed MLflow, runs, plot, and artifact buffers explicitly
+  use `nowrap` and `noswapfile` instead of depending on editor defaults.
+- Updated README and Vim help documentation for caching, viewer layout, JSON
+  display, and new configuration settings.
+
+### Fixed
+- Fixed repeated redraws so section toggles, artifact directory expansion, and
+  experiment/run selection preserve the visible scroll position much more
+  reliably.
+- Fixed artifact buffer identity so similarly named artifacts from different
+  runs no longer collide.
+- Fixed artifact and plot viewer placement so the main `__MLflow__` buffer stays
+  pinned while viewer panes are reused predictably.
+- Fixed noisy artifact-open status messages that previously triggered hit-enter
+  prompts in Vim.
+- Fixed Neovim wrapping issues that broke table and artifact display layouts.

README.md

@@ -4,24 +4,23 @@
 [![codestyle-python-flake8](https://github.com/aganse/vim-mlflow/workflows/codestyle-python-flake8/badge.svg)](https://github.com/aganse/vim-mlflow/actions/workflows/codestyle-python-flake8.yml)
 [![codestyle-vimscript-vint](https://github.com/aganse/vim-mlflow/workflows/codestyle-vimscript-vint/badge.svg)](https://github.com/aganse/vim-mlflow/actions/workflows/codestyle-vimscript-vint.yml)
 ![licence](https://img.shields.io/badge/license-MIT-blue.svg)
-![version](https://img.shields.io/badge/version-1.0.2-blue.svg)
+![version](https://img.shields.io/badge/version-1.1.0-blue.svg)
 
 
-`Vim‑mlflow` is a lightweight Vim/NVim plugin that lets you browse and interact
-with MLflow experiments, runs, metrics, parameters, tags, and artifacts directly
-in your Vim editor.  It opens a dedicated sidebar and a detail pane so you can
-explore data without leaving the terminal, even allowing you to plot metric
-histories and browse non-graphical artifacts.  The plugin is written in
-Vimscript with embedded Python and talks to MLflow through its Python API.
-It works with both MLflow3.x and MLflow2.x ML tracking servers (but not the
-GenAI traces/etc in MLflow3.x currently; feedback/demand can guide such future
-steps).
+`Vim‑mlflow` is a Vim/NVim plugin that lets you browse and interact with MLflow
+experiments, runs, metrics, parameters, tags, and artifacts directly in your
+Vim editor.  It opens a dedicated sidebar and a detail pane so you can explore
+data without leaving the terminal, even allowing you to plot metric histories
+and browse non-graphical artifacts.  The plugin is written in Vimscript with
+embedded Python and talks to MLflow through its Python API.  It works with both
+MLflow3.x and MLflow2.x ML tracking servers (but not the GenAI traces/etc in
+MLflow3.x currently).
 
-[![example vim-mlflow screenshot](doc/demo_1.0.0_light.gif)](doc/demo_1.0.0_light.gif)
+[![example vim-mlflow screenshot](doc/demo_1.1.0_light.gif)](doc/demo_1.1.0_light.gif)
 
 
 ## TL;DR
-* Must run Vim/NVim in a python environment with `mlflow` installed.
+* Must run Vim/NVim in a python environment with `mlflow` and `pandas` installed.
 * Vim must be a compiled-with-python version (check `vim --verison` for `+python3`);
   or for NVim just install the `pynvim` package in that python environment as well.
 * Put `Plugin 'aganse/vim-mlflow'` or your package manager equivalent in your
@@ -46,8 +45,8 @@ steps).
 - `python3 -m venv .venv`
 - `source .venv/bin/activate  # syntax for linux/mac`
 
-#### 3. Install the `mlflow` Python package (and also `pynvim` for Nvim):
-- `pip install mlflow` (in both Vim and NVim)
+#### 3. Install the required Python packages (`mlflow`, `pandas`, and `pynvim` for Nvim):
+- `pip install mlflow pandas` (in both Vim and NVim)
 - In NVim you also need this package in your env to support the python:
   `pip install pynvim`
 
@@ -67,13 +66,12 @@ steps).
 - The Configuration section has quite a list of settings (colors, characters,
   sizing, etc) that can be customized.
 
-- For NVim you may need to set `setlocal nowrap` in your resource file - see
-  last Troubleshooting tip below regarding line-wrap default in NVim affecting
-  content layout.
+- vim-mlflow manages `nowrap` itself for its scratch buffers, so no extra NVim
+  line-wrap setting should normally be needed.
 
 
 ## Usage
-* Ensure you're in your python environment with MLflow before starting Vim.
+* Ensure you're in your python environment with `mlflow` and `pandas` before starting Vim.
 * Press `\m` to start vim-mlflow (default setting, ie leader-key and m. or can
   use `:call RunMLflow()`).  You can update that leader/key mapping via
   `nnoremap <leader>m :call RunMLflow()<CR>`.
@@ -91,14 +89,16 @@ steps).
 
 ## Configuration
 Only `g:mlflow_tracking_uri` is required to be set by user (e.g. in resource file).
-But a typical small set of vim-mlflow config variables that one might set is:
+But a typical small set of vim-mlflow config variables that one would usually set is:
 ```vim
 " Vim-mlflow settings
 let g:mlflow_tracking_uri = "http://localhost:5000"  " running locally or via ssh-tunnel
 let g:vim_mlflow_icon_useunicode = 1  " default 0 value uses ascii chars instead
 let g:vim_mlflow_width = get(g:, 'vim_mlflow_width', 50)  " width of mlflow window
 let g:vim_mlflow_expts_length = 10  " experiments to show at a time
 let g:vim_mlflow_runs_length = 15   " runs to show at a time
+let g:vim_mlflow_plotpane_pct = 66  " plot pane height % when artifact pane is also open
+let g:vim_mlflow_runs_cache_mode = 'selected_expt'  " (or 'all_expts') how much to query at once 
 ```
 
 By default Vim-mlflow uses standard color groups like "Comment" and "Statement"
@@ -107,47 +107,18 @@ work" in vim-mlflow.  (E.g. the animated GIF above used
 [PaperColor](https://github.com/vim-scripts/PaperColor.vim) colorscheme;
 see also its [dark-mode equivalent animated GIF](doc/demo_1.0.0_dark.gif)).
 But all details can be changed, per listing below.
-With no configuration parameters set, ascii characters with no color are used.
+With no configuration parameters set, ASCII characters with no color are used.
 
 See the [full listing of vim-mlflow config variables](doc/configuration_params.md)
 that may be of interest to set in your resource file.
 
 
-## Troubleshooting
-- The sidebar may be slow on high-latency MLflow connections because each
-  refresh starts a short-lived Python process and re-queries MLflow.
-  Performance seems fine when Vim runs close to the tracking server; running
-  all components in AWS within same region, it has worked well for our team.
-  On slower links, increasing `g:vim_mlflow_timeout` may help.  A future version
-  could use a persistent Python process to reduce queries if necessary, but so
-  far this has not been a common enough concern.
-- Unicode icons require a font that includes box-drawing characters.  Set
-  `g:vim_mlflow_icon_useunicode = 0` if glyphs look broken as the simple quick
-  fix, and also note there are config vars to change individual icon characters.
-- Text artifacts (`*.txt`, `*.json`, `*.yaml`, `MLmodel`) open directly in the
-  plugin.  Binary artifacts are listed but cannot be opened in the plugin.
-- If the plugin fails to load in classic Vim, verify that Vim supports Python
-  (vim --version) and that mlflow is importable in Vim’s Python environment
-  (:py3 import mlflow). In Neovim, also ensure pynvim is installed.
-- In NVim if the layout seems screwy, check step 5 above in Installation
-  regarding `nowrap`.
-- Neovim enables line wrapping by default, which can break table layouts in
-  this plugin. Adding `setlocal nowrap` fixes this globally. The issue is most
-  noticeable in the MLflowRuns window (opened with R), which displays many
-  columns.
-
-
 ## Contributing
-Contributions are welcome; just note this project is maintained on a best-effort
-basis by a single maintainer (Andy Ganse) alongside other commitments, and so
-focused on long-term maintainability more than rapid feature growth.  Bug fixes,
-documentation improvements, and small, well-scoped enhancements are the most
-likely to be accepted.  Feature requests may be declined if they significantly
-increase complexity, maintenance burden, or diverge from the project’s stated
-scope.  Please read [CONTRIBUTING.md](CONTRIBUTING.md) before opening issues or
-pull requests, and note that response and review times may not be fast.
-
-#### Dev tools to be aware of when contributing
+Contributions are welcome. This project is maintained on a best-effort basis,
+so bug fixes, documentation improvements, and small, well-scoped enhancements
+are the most likely to be accepted. Please read [CONTRIBUTING.md](CONTRIBUTING.md)
+before opening issues or pull requests.
+
 This repo has unittests and codestyle checks, implemented in both CI workflows
 and also available locally via the following Makefile calls.  To run these you
 need a few more packages in your Python environment than when just using the
@@ -161,41 +132,16 @@ CI workflows):
 - `make codestyle` lints the Vimscript in `plugin/` with `vint` and the
   Python in `python/` with `flake8`.
 
-#### Related repos you may find useful to populate an MLflow installation with test data
-The following tools are useful in their own right to get an MLflow instance up
-and running quickly and to get some modeling up and running quickly.
-But in this case you may find them convenient to populate test contents into
-a temporary MLflow tracking server for dev/test purposes - they created the
-contents seen in screencast above:
-* [aganse/docker_mlflow_db](https://github.com/aganse/docker_mlflow_db):
-    ready-to-run MLflow server with PostgreSQL, AWS S3, Nginx
-* [aganse/py_torch_gpu_dock_mlflow](https://github.com/aganse/py_torch_gpu_dock_mlflow):
-    ready-to-run Python/PyTorch/MLflow-Projects setup to train models on GPU
-
-
-## Legacy/older versions
-Legacy/older versions of this plugin can be accessed by git checking out an
-earlier version locally, and then referencing it in your .vimrc (for classic
-Vim like `Plugin 'file:///my/path/to/python/vim-mlflow'`.
-Or similarly you can set that path in your runtimepath in NVim (without the
-`file://`).
-That said, it's recommended to use >= v1.0.0 - that's the first "official"
-release (we'll just keep adding to this table as more releases come out).
-
-|  vim-mlflow git tag  | tested with mlflow version | tested with vim version |
-| ---------------------| -------------------------- | ----------------------- |
-| v0.8                 |  1.26.1                    | vim 8.2                 |
-| v0.9                 |  1.30.0, 2.7.1             | vim 8.2                 |
-| v1.0.0 (this version)|  2.12.0, 2.19.0, 3.6.0     | vim 9.1, nvim v0.11.5   |
-
 
 ## Making the animated screen-shot gif
+Just so I remember for next time:
 * Install [rust](https://rust-lang.org/tools/install) (`curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh`)
 * Install [asciinema](https://github.com/asciinema/asciinema) (`cargo install --locked --git https://github.com/asciinema/asciinema`)
 * `~/.cargo/bin/asciinema rec demo.cast  # start recording terminal screen to file`
 * Manually conduct the usage sequence to record, which gets saved to file; ctrl-D to exit/end when done.
 * Install [agg](https://github.com/asciinema/agg) (`cargo install --git https://github.com/asciinema/agg`)
-* `~/.cargo/bin/agg --speed 2 demo.cast demo.gif  # convert the asciinema cast to animated gif`
+* `~/.cargo/bin/agg --font-size 24 --speed 2 --theme github-light --font-family "SauceCodePro Nerd Font" demo.cast demo.gif  # convert the asciinema cast to animated gif`
+  (the "SauceCodePro Nerd Font" is what my system uses but that arg can be left out)
 
 
 ## Acknowledgements

VERSION

@@ -1 +1 @@
-1.0.2
+1.1.0

dev-requirements.txt

@@ -1,4 +1,5 @@
 mlflow
+pandas
 pynvim
 pytest
 flake8

doc/configuration_params.md

@@ -5,13 +5,14 @@ Full list of vim-mlflow config variables that may be of interest to set in resou
 | `g:mlflow_tracking_uri` _(required)_ | The MLFLOW_TRACKING_URI of the MLflow tracking server to connect to (default is `"http://localhost:5000"`)|
 | `g:vim_mlflow_timeout`           | Timeout in float seconds if cannot access MLflow tracking server (default is 0.5)|
 | `g:vim_mlflow_buffername`        | Buffername of the MLflow side pane (default is `__MLflow__`)|
-| `g:vim_mlflow_runs_buffername`   | Buffername of the MLflowRuns side pane (default is `__MLflow__`)|
-| `g:vim_mlflow_vside`             | Which side to open the MLflow pane on: 'left' or 'right' (default is `right`)|
+| `g:vim_mlflow_runs_buffername`   | Buffername of the MLflowRuns side pane (default is `__MLflowRuns__`)|
+| `g:vim_mlflow_vside`             | Which side to open the MLflow pane on: 'left' or 'right' (default is `left`)|
 | `g:vim_mlflow_hside`             | Whether to open the MLflowRuns pane 'below' or 'above' (default is `below`)|
 | `g:vim_mlflow_width`             | Width of the vim-mlflow window in chars (default is 70)|
-| `g:vim_mlflow_height`            | Width of the vim-mlflow window in chars (default is 10)|
+| `g:vim_mlflow_height`            | Height of the MLflowRuns window in rows (default is `10`)|
 | `g:vim_mlflow_expts_length`      | Number of expts to show in list (default is 8)|
 | `g:vim_mlflow_runs_length`       | Number of runs to show in list (default is 8)|
+| `g:vim_mlflow_runs_cache_mode`   | Cache run summaries for `'selected_expt'` (default) or eagerly for `'all_expts'`|
 | `g:vim_mlflow_viewtype`          | Show 1:activeonly, 2:deletedonly, or 3:all expts and runs (default is 1)|
 | `g:vim_mlflow_show_scrollicons`  | Show the little up/down scroll arrows on expt/run lists, 1 or 0 (default is 1, ie yes show them)|
 | `g:vim_mlflow_icon_useunicode`   | Allow unicode vs just ascii chars in UI, 1 or 0 (default is 0, ascii)|
@@ -37,7 +38,7 @@ Full list of vim-mlflow config variables that may be of interest to set in resou
 | `g:vim_mlflow_color_between_plotpts` | Highlight group for line segments between points (default `'Comment'`)|
 | `g:vim_mlflow_plot_height`       | ASCII plot height in rows when graphing metric history (default `25`)|
 | `g:vim_mlflow_plot_width`        | ASCII plot width in columns (default `70`)|
+| `g:vim_mlflow_plotpane_pct`      | Plot pane height percent when both plot and artifact panes are visible (default `66`)|
 | `g:vim_mlflow_plot_xaxis`        | `'step'` or `'timestamp'` for metric plot x-axis (default `'step'`)|
 | `g:vim_mlflow_plot_reuse_buffer` | If `1`, reuse a single `__MLflowMetricPlot__` buffer; if `0`, create sequential plot buffers (default `1`)|
 | `g:vim_mlflow_artifacts_max_depth` | Maximum artifact directory depth shown when expanding folders (default `3`)|
-

doc/description.txt

@@ -1,16 +1,15 @@
-`Vim‑mlflow` is a lightweight Vim/NVim plugin that lets you browse and interact
-with MLflow experiments, runs, metrics, parameters, tags, and artifacts directly
-in your Vim editor.  It opens a dedicated sidebar and a detail pane so you can
-explore data without leaving the terminal, even allowing you to plot metric
-histories and browse non-graphical artifacts.  The plugin is written in
-Vimscript with embedded Python and talks to MLflow through its Python API.
-It works with both MLflow3.x and MLflow2.x ML tracking servers (but not the
-GenAI traces/etc in MLflow3.x currently; feedback/demand can guide such future
-steps).
+`Vim‑mlflow` is a Vim/NVim plugin that lets you browse and interact with MLflow
+experiments, runs, metrics, parameters, tags, and artifacts directly in your
+Vim editor.  It opens a dedicated sidebar and a detail pane so you can explore
+data without leaving the terminal, even allowing you to plot metric histories
+and browse non-graphical artifacts.  The plugin is written in Vimscript with
+embedded Python and talks to MLflow through its Python API.  It works with both
+MLflow3.x and MLflow2.x ML tracking servers (but not the GenAI traces/etc in
+MLflow3.x currently).
 
 
 ## TL;DR
-* Must run Vim/NVim in a python environment with `mlflow` installed.
+* Must run Vim/NVim in a python environment with `mlflow` and `pandas` installed.
 * Vim must be a compiled-with-python version (check `vim --verison` for `+python3`);
   or for NVim just install the `pynvim` package in that python environment as well.
 * Put `Plugin 'aganse/vim-mlflow'` or your package manager equivalent in your
@@ -23,7 +22,7 @@ steps).
 
 
 ## Usage
-* Ensure you're in your python environment with MLflow before starting Vim.
+* Ensure you're in your python environment with `mlflow` and `pandas` before starting Vim.
 * Press `\m` to start vim-mlflow (default setting, ie leader-key and m. or can
   use `:call RunMLflow()`).  You can update that leader/key mapping via
   `nnoremap <leader>m :call RunMLflow()<CR>`.

doc/installdetails.txt

@@ -11,8 +11,8 @@
 - `python3 -m venv .venv`
 - `source .venv/bin/activate  # syntax for linux/mac`
 
-#### 3. Install the `mlflow` Python package (and also `pynvim` for Nvim):
-- `pip install mlflow` (in both Vim and NVim)
+#### 3. Install the required Python packages (`mlflow`, `pandas`, and `pynvim` for Nvim):
+- `pip install mlflow pandas` (in both Vim and NVim)
 - In NVim you also need this package in your env to support the python:
   `pip install pynvim`
 
@@ -32,6 +32,5 @@
 - The Configuration section has quite a list of settings (colors, characters,
   sizing, etc) that can be customized.
 
-- For NVim you may need to set `setlocal nowrap` in your resource file - see
-  last Troubleshooting tip below regarding line-wrap default in NVim affecting
-  content layout.
+- vim-mlflow manages `nowrap` itself for its scratch buffers, so no extra NVim
+  line-wrap setting should normally be needed.