Skip to content

ngpepin/mdexplore

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

114 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mdexplore

Fast Markdown explorer for Ubuntu/Linux desktop: browse .md files in a directory tree and preview fully rendered output instantly.

User Guide

For a use-case/tutorial walkthrough, start with USER-GUIDE.md.

Code Layout

The application still uses mdexplore.py as the main entrypoint and primary UI module, but low-risk support code is now split into mdexplore_app/:

  • mdexplore.py: main window, markdown renderer, preview/PDF orchestration, and app entrypoint.
  • mdexplore_app/constants.py: shared runtime/render constants.
  • mdexplore_app/runtime.py: runtime environment helpers (config path, default root, GPU/software fallback, print layout knobs).
  • mdexplore_app/search.py: extracted search tokenization, boolean/NEAR parsing, and per-file hit counting.
  • mdexplore_app/templates.py: extracted HTML template-asset registry/renderer for preview document shells.
  • mdexplore_app/pdf.py: PDF footer stamping, blank-page suppression, and PlantUML stderr formatting.
  • mdexplore_app/icons.py: icon loading/recoloring helpers.
  • mdexplore_app/tree.py: markdown tree model and delegate (ColorizedMarkdownModel, MarkdownTreeItemDelegate).
  • mdexplore_app/tabs.py: custom multi-view tab bar (ViewTabBar).
  • mdexplore_app/workers.py: background worker classes for preview render, PlantUML, and PDF write/stamp steps.
  • mdexplore_app/fast_base64.py: shared BASE64 helpers with vendor fastbase64 + pybase64 acceleration and stdlib fallback.

This is intentionally a first-stage modularization. Most behavior and call flow still lives in mdexplore.py so feature risk stays low while the file is gradually decomposed.

Features

  • Expandable left-pane directory tree rooted at a chosen folder.
  • Shows Markdown files only (*.md), while still allowing folder navigation.
  • Large right-pane rendered preview with scroll support.
  • PlantUML diagram renders run in background workers so markdown preview stays responsive.
  • Supports:
    • CommonMark + tables + strikethrough.
    • Markdown rendering defaults to cmarkgfm for speed, with automatic fallback to markdown-it-py when unavailable/incompatible.
    • TeX/LaTeX math via MathJax.
    • Mermaid diagrams with improved dark-theme contrast.
    • Optional Rust Mermaid backend via mmdr (--mermaid-backend rust).
    • PlantUML diagrams (asynchronous local render with placeholders).
    • Markdown callouts (> [!NOTE], > [!TIP], > [!IMPORTANT], > [!WARNING], > [!CAUTION]).
  • Top actions:
    • Recent opens a dropdown of up to 35 retained root directories: first 10 shown most-recent-first, then a separator, then up to 25 remaining entries sorted alphabetically. The list refreshes from disk each time the menu is opened so multiple running instances stay in sync. A root is added only after it has been active for at least 30 seconds and then you navigate to another root.
    • ^ moves root up one directory level.
    • Refresh rescans the current directory view to pick up new/deleted files.
    • PDF exports the current preview to <filename>.pdf with centered page numbering (N of M). During export, retrievable image sources are inlined as BASE64 data URIs so images render in the PDF instead of relying on external links; broken links are left unchanged.
    • Add View creates another tabbed view of the same document at the current top visible line.
    • Edit opens the selected file in MarkText (/usr/bin/marktext).
  • Window title shows the current effective root path.
  • Effective-root directory row is always bold:
    • aqua-blue (#7fdfe8) when no active search hits are under that scope,
    • yellow with an appended hit-count pill when active search has hits under that scope (1..99, then ++).
  • Preview cache keyed by file timestamp and size for fast re-open.
  • Inline preview BASE64 images are materialized in parallel per document (deduped by payload hash) to reduce first-load latency on image-heavy files.
  • Copy-time BASE64 image conversion warms image targets in parallel before rewrite for faster large-file copy/export workflows.
  • Mermaid SVGs are cached in-memory per app run to avoid re-rendering when returning to previously viewed files.
  • Mermaid keeps separate in-memory cache modes for GUI (auto) and PDF (pdf) rendering.
  • Navigating back to a cached file still performs a fresh stat check; changed files re-render automatically.
  • Startup-generated icon assets are persisted under ~/.cache/mdexplore/icon-cache:
    • app icon normalization output (from mdexplor-icon.png/fallback icon files),
    • two-tone icon recolor outputs used by the tab/tree UI. These are regenerated automatically when source asset identity (path/mtime/size) or render parameters change.
  • F5 refresh shortcut for directory view rescan (same behavior as Refresh button).
  • Ctrl++ / Ctrl+- / Ctrl+0 zoom only the preview pane content in/out/reset.
  • Preview-only zoom changes briefly show a percentage badge at the top of the preview pane.
  • If the currently previewed markdown file changes on disk, preview auto-refreshes and shows a status bar message.
  • Status bar reports active long-running work (preview load/render, PlantUML progress, PDF export) and returns to Ready instead of staying blank.
  • Manual tree/preview pane resizing is preserved across ^ root navigation for the current app run.
  • Right-click a Markdown file to assign a highlight color in the tree.
  • Highlight colors persist per directory in .mdexplore-colors.json files.
  • Available file-highlight colors are: Yellow, Green, Blue, Orange, Purple, Light Gray, Medium Gray, and Red.
  • View-tab state persists per directory in .mdexplore-views.json files for documents that have more than one saved view or a custom tab label.
  • Persistent preview text highlights are stored per directory in .mdexplore-highlighting.json.
  • Files in the tree show a small light-gray tab badge when they currently have more than the default single view.
  • Files with persistent preview text highlights show a separate purple marker badge in the tree.
  • Markdown tree gutter badges are packed together in a fixed-width strip so the markdown file icon and filename stay aligned even when only some badges apply.
  • Right-click tree menu includes Clear in Directory (non-recursive) and Clear All (recursive) to remove persisted file-highlight colors, and both actions prompt for confirmation before clearing.
  • Top-right copy controls are labeled Copy to: () Clipboard () Directory, with Clipboard selected by default.
  • A BASE64 image toggle button sits immediately to the left of Copy to::
    • off tooltip: Turn BASE64 image encoding on
    • on tooltip: Turn BASE64 image encoding off
    • toggle state persists in ~/.mdexplore.cfg across restarts (copy_base64_images_enabled).
    • when enabled, copied markdown (clipboard staging or directory copy) converts retrievable image links (local paths and URLs) into embedded BASE64 data: URIs.
  • In Clipboard mode, color buttons copy matching highlighted files and the pin button copies the currently previewed markdown file.
  • In Directory mode, pin/color actions open a target-folder picker, then copy the file(s) into that folder.
  • Directory copy also writes merged metadata entries for copied files into the target folder sidecars: .mdexplore-colors.json, .mdexplore-views.json, and .mdexplore-highlighting.json.
  • Search box includes an explicit X clear control that clears the query and removes match styling.
  • Active search evaluates the markdown files currently visible in the tree (root + expanded branches), and reruns automatically when tree visibility changes (expand/collapse/root/scope navigation).
  • Search-hit files in the tree show a yellow hit-count pill in the left gutter.
  • Matching filenames are styled bold+italic, and filename-term matches are rendered in yellow text.
  • When search is active and a matched file is opened, preview matches are highlighted in yellow and the view scrolls to the first match.
  • While dragging the preview scrollbar, mdexplore shows an approximate current line / total lines indicator beside the scrollbar handle.
  • When search is active, preview scrollbar markers show the vertical positions of highlighted hits; clicking a marker jumps to the nearest hit in that cluster.
  • Preview scroll position is remembered per markdown file for the current app session.
  • Right-click selected text in the preview pane to use:
    • Copy Rendered Text for plain rendered selection text.
    • Copy Source Markdown for markdown source content. Copies matching source markdown using direct range mapping first, then selected-text/fuzzy line matching as fallback, and finally the full source file if no match is possible.
  • Clipboard copy uses file URI MIME formats compatible with Nemo/Nautilus paste.
  • Last effective root plus recent-root history are persisted to ~/.mdexplore.cfg on root navigation and on exit.
    • Payload format is JSON with keys default_root, recent_roots, and copy_base64_images_enabled (legacy plain-text config is still accepted on read).
    • Recent roots are capped at 35 entries in rolling newest-first storage. Menu presentation is split: 10 most-recent-first, separator, then remaining entries alphabetically.
    • Config writes use a lock file (~/.mdexplore.cfg.lock) with non-blocking, momentary locking.
    • If another instance holds the lock during a save attempt, that save is skipped silently.
    • Lock files older than 2 minutes are cleaned up automatically (silently).
    • If no directory is selected at quit time, the most recently selected/expanded directory is used for default_root.

Requirements

  • Ubuntu/Linux desktop with GUI.
  • Python 3.10+.
  • python3-venv package available.
  • Internet access for Mermaid only when no local Mermaid bundle is available.
  • Internet access for MathJax only when no local MathJax bundle is available.
  • Java runtime (java in PATH) for local PlantUML rendering.
  • plantuml.jar available (vendored path by default, or set PLANTUML_JAR).
  • Optional: MarkText installed at /usr/bin/marktext for Edit.
  • Optional: mmdr in PATH (or MDEXPLORE_MERMAID_RS_BIN) for Rust Mermaid backend.

Quick Start

From any directory:

/path/to/mdexplore/setup-mdexplore.sh
/path/to/mdexplore/mdexplore.sh

setup-mdexplore.sh is the full bootstrap path. It:

  • creates or updates .venv
  • installs Python dependencies from requirements.txt
  • verifies tracked UI assets under assets/ui
  • downloads vendored MathJax, Mermaid, and PlantUML runtime assets if they are missing
  • downloads/builds the Rust Mermaid renderer under vendor/mermaid-rs-renderer

It is safe to rerun. After bootstrap, use mdexplore.sh for normal launches.

When no PATH is supplied, the app opens:

  1. default_root from ~/.mdexplore.cfg (if valid), otherwise
  2. your home directory.

To open a specific root directory:

/path/to/mdexplore/mdexplore.sh /path/to/notes

What the launcher does:

  • Creates .venv inside the project if missing.
  • Uses .venv/bin/python directly (does not alter your current shell session).
  • Installs dependencies when requirements.txt changes.
  • Stores a runtime import-check stamp in .venv/.runtime-import-check.sha256.
  • Runs runtime import verification when one of these is true:
    • .venv is newly created,
    • requirements.txt hash changed,
    • runtime import-check stamp is missing or does not match requirements hash.
  • If runtime import verification fails, self-heals by reinstalling dependencies.
  • Runs the app.

Usage

Wrapper script

mdexplore.sh [--mermaid-backend js|rust] [PATH]
  • PATH is optional.
  • --mermaid-backend is optional. For mdexplore.sh, the launcher defaults to rust when not specified. (mdexplore.py direct runs default to js.)
  • Supports plain paths and file:// URIs (for .desktop %u launches).
  • If a file path is passed, mdexplore opens its parent directory.
  • If omitted, default_root in ~/.mdexplore.cfg is used (falling back to home directory).
  • --help prints usage.

Direct Python run

python3 -m pip install -r /path/to/mdexplore/requirements.txt
python3 /path/to/mdexplore/mdexplore.py [--mermaid-backend js|rust] [PATH]

If PATH is omitted for direct run, the same config/home default rule applies.

hfind CLI

hfind reuses mdexplore search syntax (AND/OR/NOT, quoted terms, NEAR(...)) for file discovery.

Run via wrapper:

./hfind.sh [--query QUERY|-q QUERY] [--content|-c] [--recursive|-r] [--verbose|-v] [--pdf|-p] [--sort|-s] [--sort-case-sensitive|-S] PATTERN [PATTERN ...]

Run via Python directly:

python3 /path/to/mdexplore/hfind.py [--query QUERY|-q QUERY] [--content|-c] [--recursive|-r] [--verbose|-v] [--pdf|-p] [--sort|-s] [--sort-case-sensitive|-S] PATTERN [PATTERN ...]

Notes:

  • --content / -c is optional (default is filename-only search).
  • --recursive / -r is optional.
  • --verbose / -v prints matching line(s) under each matched file and highlights hit text in yellow.
  • --pdf / -p enables searching extracted text inside .pdf files.
  • Default output is progressive: matches print as they are found.
  • --sort / -s waits for full scan, then prints results sorted case-insensitively.
  • --sort-case-sensitive / -S waits for full scan, then prints results sorted case-sensitively.
    • when either sort mode is enabled, hfind prints: One moment please... finding all matches to sort them
  • NEAR(...) is strict in both mdexplore and hfind: a hit is only produced when all NEAR terms occur within the configured 50-word window.
  • For content search, inline data:image/...;base64,... payloads are ignored in both mdexplore and hfind to avoid false positives from embedded image data.
  • In verbose NEAR(...) output, numbering reflects match semantics:
    • lines that independently satisfy NEAR(...) are each numbered,
    • lines shown only as contiguous context for a cross-line NEAR window are grouped and only the first line is numbered.
  • Quoted terms with exactly one leading and/or trailing literal space use boundary-aware matching at that edge (word boundary, punctuation, or line break can satisfy it).
    • Example: 'Nico ' can match Nico., Nico), Nico;, or Nico followed by a line break.
    • Example: ' Nico ' can match \nNico\n and His name is Nico.
    • Exception: this boundary rule applies only to a single leading/trailing space. Terms like ' Nico' or 'Nico ' require those two literal spaces.
  • Short flags are stackable in any order (for example -cr, -rc).
  • If -q / --query is omitted, the first positional string is treated as the query.
  • Filename-only mode checks the full filename (including extension, no path).

Examples:

./hfind.sh --query "OR(fred, paul)" --content --recursive *.txt

Recursively searches from current directory for readable .txt files whose filename or content contains (case-insensitive) fred or paul.

./hfind.sh -q "OR(fred, paul)" -cr *.txt
./hfind.sh -cr "OR(fred, paul)" *.txt
./hfind.sh --recursive -c "OR(fred, paul)" *.txt
./hfind.sh -crvp "OR(fred, paul)" *.txt
./hfind.sh -crvps "OR(fred, paul)" *.txt
./hfind.sh -crvpS "OR(fred, paul)" *.txt

All above are valid shorthand forms.

./hfind.sh -rs "conflicted" "./**/*"
./hfind.sh -rS "conflicted" "./**/*"

Shows the two sort modes over recursive filename matches:

  • -rs sorts case-insensitively.
  • -rS sorts case-sensitively.
./hfind.sh --query "OR(fred, paul)" *.txt

Searches filenames only (current directory, non-recursive).

./hfind.sh -q "AND('Fred',paul)" /path/to/directory/*.md

Searches markdown filenames in /path/to/directory/ (non-recursive) where query requires case-sensitive Fred and case-insensitive paul.

./hfind.sh -rc "NEAR(\"Fred is my friend\", \"is a nice guy\")" "/path/to my/directory/*.md"

Uses escaped double quotes inside a NEAR(...) query while recursively searching matching markdown files.

./hfind.sh -crv "NEAR(alpha,beta)" "./**/*.md"

For each matching file, prints matching line numbers and highlights the triggering terms in yellow. For NEAR(...), this may include multiple nearby lines so both terms are visible.

./hfind.sh -cv "'Nico '" "./**/*.md"
./hfind.sh -cv "' Nico '" "./**/*.md"

Demonstrates the single-space boundary behavior for quoted terms (including punctuation and line-break boundaries).

Full Bootstrap Script

setup-mdexplore.sh [--skip-python] [--skip-assets] [--skip-rust] [--rebuild-rust]
  • --skip-python leaves the existing .venv untouched.
  • --skip-assets skips vendored MathJax / Mermaid JS / PlantUML runtime asset checks.
  • --skip-rust skips Rust Mermaid bootstrap/build.
  • --rebuild-rust forces a fresh cargo build --release --locked for mmdr.

By default the script prefers the vendored vendor/mermaid-rs-renderer/ checkout if it already exists. If it does not exist, the script clones it from:

https://github.com/1jehuang/mermaid-rs-renderer.git

If cargo is missing, the setup script installs a minimal Rust toolchain through rustup so it can build mmdr locally.

Headless Regression Tests

Run the GUI regression suite under xvfb so QWebEngineView has a display:

xvfb-run -a .venv/bin/python -m unittest discover -s tests -v

The current suites in tests/test_preview_regressions.py, tests/test_search_query_syntax.py, tests/test_pdf_layout_hints.py, tests/test_template_assets.py, tests/test_tab_bar_layout.py, and tests/test_window_layout.py cover:

  • saved view-tab round trips on the test/testdoc.md fixture,
  • left-gutter named-view markers restoring the same saved positions as tabs,
  • left-gutter persistent highlight markers navigating to visible highlights,
  • right-gutter search markers navigating to visible search hits,
  • same-document live preview search coverage for unquoted terms, double/single quotes, trailing-space phrases, implicit AND, function-style AND/OR, NOT, and NEAR(...) variants,
  • search parser/matcher handling for unquoted terms, double-quoted phrases, single-quoted case-sensitive phrases, and literal apostrophes inside double-quoted phrases,
  • PDF post-pass TOC detection and landscape/diagram page-flag classification,
  • HTML preview-template asset loading and MarkdownRenderer.render_document() integration,
  • custom tab-bar label budgeting and stale close-button geometry fallback.
  • top-bar path-label layout protection so long document paths do not force the main window to stay overly wide.

File Highlights

  • In the file tree, right-click any .md file.
  • Choose a highlight color (Highlight Yellow, then ... <Color>), or clear it.
  • Colors are persisted in .mdexplore-colors.json in each affected directory.
  • Available colors include Light Gray and Medium Gray in addition to the original color set.
  • If a directory is not writable, color persistence fails quietly by design.
  • Use top-right Copy to: () Clipboard () Directory controls to choose copy destination mode.
  • In Clipboard mode, color buttons copy files with a given highlight color and the pin button copies the currently previewed markdown file path payload.
  • In Directory mode, pin/color actions open a folder chooser that defaults to: previously selected destination folder, else current effective root.
  • Directory copy writes file content and merges copied-file metadata into target .mdexplore-colors.json, .mdexplore-views.json, and .mdexplore-highlighting.json (creating sidecars when missing, updating existing sidecars by filename when present).
  • Right-click a directory (or file row) and use Clear in Directory for non-recursive clear in that directory, with confirmation.
  • Right-click a directory (or file row) and use Clear All for recursive clear under that scope, with confirmation.
  • Color-copy match collection is recursive and uses scope in this order: selected directory, else most recently selected/expanded directory, else root.

Preview Text Highlights

  • In the preview pane, right-click selected text and choose either Highlight or Highlight Important to add a persistent text highlight behind that rendered text.
  • Normal highlights use a darker subdued purple. Important highlights use a lighter purple so the two categories are visually distinct on the dark theme.
  • Selecting part or all of an existing highlighted block and applying the other action converts just that selected range to the chosen highlight type.
  • Highlighted preview text also appears as left-side navigation markers in the preview gutter. Important-highlight markers are lighter than normal-highlight markers, and longer markers indicate highlights that span more lines.
  • Clicking a highlight marker jumps to the corresponding highlighted block.
  • Named views with a saved Return to beginning anchor also appear as color-matched left-gutter markers in the preview, layered above highlight markers when the two coincide.
  • Right-click inside an existing highlighted block to remove it with Remove Highlight.
  • If the current selection overlaps existing highlighted text and also includes unhighlighted text, both highlight actions and Remove Highlight remain available so the block can be extended, converted, or removed.
  • Preview text highlights persist per directory in .mdexplore-highlighting.json.
  • The tree also mirrors persisted preview highlights with a marker badge so highlighted documents remain easy to spot while browsing directories.
  • The tree gutter badge order is: hit-count pill, highlight marker, views badge, markdown file icon.
  • Highlight persistence is best-effort: if the directory is not writable, mdexplore fails quietly rather than interrupting preview use.

Preview Zoom

  • Ctrl++ zooms the preview pane in.
  • Ctrl+- zooms the preview pane out.
  • Ctrl+0 resets preview zoom to 100%.
  • These shortcuts affect only the preview content, not the tree pane, toolbar, or overall window layout.
  • Each zoom change briefly shows a percentage badge at the top-center of the preview pane and also reports the same value in the status bar.

PDF Export

  • Click PDF (between Refresh and Add View) to export the currently previewed file.
  • Output path is the previewed file path with .pdf extension in the same directory.
  • Export is based on the active document content and print-prepared preview state (markdown + math + Mermaid + PlantUML).
  • Export waits briefly for math/diagram/font readiness to reduce rendering artifacts.
  • Mermaid PDF behavior is backend-specific:
    • JS backend: Mermaid is rendered through a print-safe monochrome/grayscale path.
    • Rust backend: Mermaid starts from a dedicated PDF SVG set rendered by mmdr (default theming), then applies print-safe grayscale normalization (multi-shade, with readable dark text).
  • Export auto-scales page content into a print-style layout with top/side margins and an uncluttered footer band.
  • Landscape diagram sections are isolated onto dedicated print blocks and use a tighter horizontal margin budget so wide diagrams can make use of the rotated page.
  • Headed landscape sections are anchored to the page that actually contains the diagram heading/content, avoiding earlier prose pages being rotated instead.
  • Pages that look like a table of contents are never promoted to landscape just because they mention a landscape section heading in the TOC text.
  • Footer number font size is matched to the document's dominant scaled body text size.
  • Pages are stamped with centered footer numbering as 1 of N, 2 of N, etc.

PDF Layout Tuning

PDF pagination and diagram sizing are controlled by a small group of constants near the top of mdexplore.py. These are the first settings to adjust if PDF output starts making poor keep/spill/landscape choices.

Most important knobs:

  • MIN_PRINT_DIAGRAM_FONT_PT
    • Lower bound for the largest font in a diagram when mdexplore tries to keep that diagram on a single page.
    • Lower it if you want tall diagrams to stay on one page more often.
    • Raise it if you prefer multi-page spill over small text.
  • MAX_PRINT_DIAGRAM_FONT_PT
    • Upper bound for diagram enlargement in PDF output.
    • Lower it if diagrams are printing too large.
    • Raise it cautiously if diagrams look too small even when there is space.
  • PDF_PRINT_WIDE_DIAGRAM_ASPECT_RATIO
    • Aspect-ratio threshold for promoting a diagram to landscape.
    • Lower it if wide diagrams are not rotating when they should.
    • Raise it if too many diagrams are switching to landscape.
  • PDF_PRINT_WIDE_DIAGRAM_LANDSCAPE_GAIN
    • Minimum width gain required before landscape is preferred.
    • Lower it if landscape should be chosen more aggressively.
    • Raise it if portrait should remain the default more often.
  • PDF_PRINT_PLANTUML_LANDSCAPE_ASPECT_RATIO
    • Separate PlantUML-specific landscape trigger.
    • Useful because PlantUML often benefits from landscape earlier than Mermaid.
  • PDF_PRINT_HORIZONTAL_MARGIN_PX
  • PDF_PRINT_VERTICAL_MARGIN_PX
    • Effective printable margin budget used by the hidden print-layout solver.
    • Lower them to let diagrams use more page area.
    • Raise them if output feels cramped against the page edges.
  • PDF_PRINT_HEADING_TO_DIAGRAM_GAP_PX
    • Vertical spacing reserved between a heading cluster and the diagram it governs.
    • Raise it if headed sections feel visually crowded.
    • Lower it if headed diagrams are spilling to the next page too eagerly.
  • PDF_PRINT_LAYOUT_SAFETY_PX
    • Extra safety padding used to avoid borderline fits that clip in Chromium.
    • Lower it if diagrams are being pushed into spill mode too early.
    • Raise it if “just barely fits” cases are clipping or producing unstable output.

Heuristic guidance:

  • If a diagram is being split across pages but you would accept smaller text, lower MIN_PRINT_DIAGRAM_FONT_PT.
  • If a diagram is tiny but should clearly be landscape, first adjust PDF_PRINT_WIDE_DIAGRAM_ASPECT_RATIO or PDF_PRINT_WIDE_DIAGRAM_LANDSCAPE_GAIN.
  • If headings are being orphaned above diagrams, leave the font knobs alone and look instead at the page-geometry and spacing constants.
  • If many diagrams are near the right answer but consistently a little too big or too small, adjust margins before changing font caps.

Recommended workflow when tuning:

  1. Change one constant at a time.
  2. Regenerate the specific problematic PDF.
  3. Recheck both the target document and one known-good document.
  4. Prefer small adjustments; many of these settings interact.

The hidden PDF preflight logic consumes these constants through a Python-to-JS configuration handoff, so changing the constants in mdexplore.py is the intended way to tune print behavior.

Known TODOs

  • Diagram interaction state restore is not yet reliable across document switches:
    • Mermaid zoom/pan state may not restore consistently when returning to a file.
    • PlantUML zoom/pan state may not restore consistently when returning to a file.
  • This is tracked as an explicit TODO for future maintenance; core markdown rendering, search, highlighting, and PDF export remain functional.

Multiple Views

  • Click Add View to create another tab for the same currently previewed document.
  • New tabs inherit the current view's top visible line/scroll position.
  • By default, tab labels show the current top-most visible source line number for that tab.
  • Tabs show a small left-side position bargraph indicating where that view sits within the document.
  • Tabs use a fixed soft-pastel color sequence based on the order each view was opened.
  • Tabs can be dragged to reorder without changing each tab's assigned color.
  • Right-click a tab to assign a custom label (including spaces) up to 48 characters.
  • If a longer label is entered, mdexplore truncates it to the first 48 characters.
  • Entering a blank custom label restores the default dynamic line-number label for that tab.
  • When a tab receives a custom label, mdexplore stores that tab's current scroll position and top visible source line as the tab's saved beginning.
  • Right-click a custom-labeled tab to use Return to beginning, which jumps that tab back to the stored label-time location.
  • Custom-labeled tabs also show a refresh icon beside the close button; clicking it resets that tab's saved beginning to the current scroll position/top line, the same way relabeling the tab does.
  • Relabeling a custom-labeled tab with different text resets the saved beginning to the scroll position at the time of relabeling.
  • When a new view is added and the palette wraps, mdexplore skips any color already used by open tabs.
  • If you switch to another markdown file and later return in the same app run, that file's tabs restore with their prior order and selected tab.
  • View-tab state also persists across app restarts in .mdexplore-views.json beside the document directory, keyed by markdown filename.
  • For custom-labeled tabs, .mdexplore-views.json also persists the stored label-time beginning location used by Return to beginning.
  • If custom labels make the tab strip too wide for the window, tab scrolling is enabled through the tab bar's built-in scroll buttons.
  • Only documents with more than one saved view, or with a custom tab label, are written to .mdexplore-views.json; untouched single-view documents continue to use the default one-tab state.
  • The tab strip is hidden when there is only one unlabeled default view.
  • If only one view remains and it has a custom label, its tab stays visible so the custom label and Return to beginning action remain available.
  • Closing that sole remaining custom-labeled tab clears the custom label and bookmark, then returns the document to the hidden default single-view state.
  • Tabs are closeable with X; at least one tab is always kept open.
  • Maximum views per document: 8.
  • Named-view left-gutter markers use the same restore path as selecting the corresponding tab, so marker navigation lands at the same saved location.

Markdown Callouts

  • mdexplore supports GitHub/Obsidian-style callouts written as blockquotes.
  • Supported types: NOTE, TIP, IMPORTANT, WARNING, CAUTION (case-insensitive).
  • INFO is accepted and styled as a NOTE callout.
  • Custom titles are supported: > [!WARNING] Custom title.
  • +/- markers are accepted in syntax, but callouts are rendered as non-collapsible boxes.

Example:

> [!NOTE]
> This is a note callout with **markdown** content.

Search and Match Highlighting

  • Use Search and highlight: to match markdown files currently visible in the tree (root + expanded directories).
  • While search is active, expand/collapse and directory/root scope changes automatically rerun the search against the newly visible set.
  • Matching files are shown bold+italic in the tree.
  • If filename terms match, filename text is rendered in yellow.
  • The effective-root directory label is bold aqua when idle, and becomes yellow with an appended hit-count pill when active search has hits under that scope.
  • Press Enter in the search field to run search immediately (skip debounce).
  • Clicking the X in the search field clears search text and removes match styling.
  • Opening a matched file while search is active highlights matching text in yellow in the preview and scrolls to the first highlighted match.
  • Preview scrollbar markers show where highlighted hits occur within the document; clicking a marker jumps to the nearest hit in that marker cluster.
  • For NEAR(...) queries, preview highlighting is constrained to qualifying NEAR windows rather than every standalone occurrence of those terms elsewhere in the document.
  • Non-quoted terms are case-insensitive.
  • Double-quoted terms are case-insensitive and preserve spaces.
  • Single-quoted terms are case-sensitive and preserve spaces.
  • Only the quote character that opens a phrase closes it; the other quote character remains literal inside the phrase.
  • Function-style operators accept both no-space and spaced forms before (: NEAR(...)/NEAR (...), OR(...)/OR (...), AND(...)/AND (...), and NOT(...)/NOT (...).
  • Legacy CLOSE(...) remains accepted for backward compatibility, but is normalized internally to canonical NEAR(...).
  • AND(...), OR(...), and NEAR(...) accept comma-delimited, space-delimited, or mixed argument lists.
  • AND(...) and OR(...) are variadic and can take 2+ arguments.
  • NEAR(...) requires 2+ terms and matches only when all terms appear within 50 words of each other in file content.
  • NEAR(...) requires distinct qualifying occurrences for each listed term; one text start cannot satisfy two different required terms.
  • For single-word NEAR(...) terms, proximity matching uses word boundaries, so the does not qualify via the The in They.
  • For NEAR(...) queries, the tree hit-count pill counts qualifying NEAR windows in the file, not the number of individual highlighted term spans.

Search examples:

joe

Matches joe, JOE, JoE, etc. (filename or content).

"Anne Smith"

Case-insensitive exact phrase match.

'Anne Smith'

Case-sensitive exact phrase match.

"Program Director's RAG"

Case-insensitive exact phrase match, with the apostrophe treated as literal text.

OR (Joe, "Fred")

Matches files containing either Joe (any case) or phrase "Fred" in any case.

NEAR(Fred "Anne Smith" Joe)

Matches only if all listed terms occur within 50 words of each other.

NEAR('The ', the)

Requires a distinct later/earlier standalone the; the The in They does not qualify.

Joe "Anne Smith" NOT draft

Implicit AND: equivalent to Joe AND "Anne Smith" AND NOT draft.

AND(alpha beta gamma)

Requires all three terms to match (equivalent to alpha AND beta AND gamma).

PlantUML Local Configuration

The app renders PlantUML blocks locally with plantuml.jar.

  • Default jar path: /path/to/mdexplore/vendor/plantuml/plantuml.jar
  • Override jar path with PLANTUML_JAR:
PLANTUML_JAR=/path/to/plantuml.jar /path/to/mdexplore/mdexplore.sh

Behavior details:

  • Markdown preview loads immediately with PlantUML rendering... placeholders.
  • Each diagram is replaced automatically as soon as it completes.
  • Diagram replacements are applied in-place so scroll position is preserved.
  • PlantUML SVG data: URIs are BASE64-encoded through guarded helpers; if vendored fastbase64 encode output is malformed, mdexplore automatically falls back to pybase64/stdlib encode so diagrams still render.
  • Failed diagrams show PlantUML render failed with error ... plus detailed stderr context (including line number when PlantUML provides one).
  • Diagram progress continues while you browse other files; returning shows completed progress.

MathJax Local-First Configuration

Math rendering is local-first:

  • mdexplore first tries a local tex-svg.js bundle (best rendering quality).
  • If no local SVG bundle is found, it falls back to local tex-mml-chtml.js.
  • If no local bundle is found, it falls back to CDN.

Local lookup order:

  1. MDEXPLORE_MATHJAX_JS (explicit file path)
  2. mathjax/es5/tex-svg.js under the repo
  3. mathjax/tex-svg.js under the repo
  4. assets/mathjax/es5/tex-svg.js under the repo
  5. vendor/mathjax/es5/tex-svg.js under the repo
  6. System paths such as /usr/share/javascript/mathjax/es5/tex-svg.js
  7. Local CHTML bundle paths (tex-mml-chtml.js) as fallback

Example override:

MDEXPLORE_MATHJAX_JS=/absolute/path/to/tex-svg.js /path/to/mdexplore/mdexplore.sh

Mermaid Local-First Configuration

Mermaid rendering is local-first:

  • mdexplore first tries a local mermaid.min.js bundle.
  • If no local bundle is found, it falls back to CDN.
  • If --mermaid-backend rust is used, mdexplore uses mmdr instead of Mermaid JS for diagram SVG generation.

Local lookup order:

  1. MDEXPLORE_MERMAID_JS (explicit file path)
  2. mermaid/mermaid.min.js under the repo
  3. mermaid/dist/mermaid.min.js under the repo
  4. assets/mermaid/mermaid.min.js under the repo
  5. assets/mermaid/dist/mermaid.min.js under the repo
  6. vendor/mermaid/mermaid.min.js under the repo
  7. vendor/mermaid/dist/mermaid.min.js under the repo
  8. System paths such as /usr/share/javascript/mermaid/mermaid.min.js

Example override:

MDEXPLORE_MERMAID_JS=/absolute/path/to/mermaid.min.js /path/to/mdexplore/mdexplore.sh

Rust backend executable override:

MDEXPLORE_MERMAID_RS_BIN=/absolute/path/to/mmdr /path/to/mdexplore/mdexplore.sh --mermaid-backend rust

Markdown Engine and BASE64 Performance Knobs

  • Markdown engine selection:
    • MDEXPLORE_MARKDOWN_ENGINE=cmark (default): uses cmarkgfm.
    • MDEXPLORE_MARKDOWN_ENGINE=markdown-it: forces markdown-it-py.
    • If cmarkgfm is unavailable at runtime, mdexplore automatically falls back to markdown-it-py.

Example:

MDEXPLORE_MARKDOWN_ENGINE=markdown-it /path/to/mdexplore/mdexplore.sh
  • BASE64 image worker threads (preview data-URI materialization + copy-time prefetch):
    • MDEXPLORE_BASE64_IMAGE_THREADS=<N>
    • Default is CPU-based (max(2, min(24, cpu_count * 2))).

Example:

MDEXPLORE_BASE64_IMAGE_THREADS=24 /path/to/mdexplore/mdexplore.sh
  • BASE64 codec path:
    • Decode path prefers pybase64, with adaptive routing to vendored vendor/fastbase64 for payload-size sweet spots.
    • Encode path may use vendored fastbase64, but output is validated (length/padding/alphabet). Invalid vendor output is discarded and mdexplore falls back to pybase64/stdlib automatically.
    • Adaptive benchmark state persists in fastbase64-adaptive.json.
    • Optional vendor-vs-fallback benchmark telemetry is written to fastbase64-benchmark.log when the debug dual-path switch is enabled in mdexplore_app/fast_base64.py.
    • If accelerated backends are unavailable, helpers fall back to Python stdlib base64.

Debug Logging

  • mdexplore.py --debug enables verbose debug logging to mdexplore.log in the project directory.
  • mdexplore.sh exposes the same behavior through the DEBUG_MODE variable near the top of the script. Set DEBUG_MODE=true to pass --debug into the app.
  • When debug logging is disabled, mdexplore.log is not written.
  • The application debug log trims itself to the most recent 10,000 lines.
  • The non-interactive launcher log remains separate at ~/.cache/mdexplore/launcher.log and trims to 1,000 lines.

Render and Rules Maps

  • See DEVELOPERS-AGENTS.md for the consolidated deep maps:
    • formal behavior/rule hierarchy,
    • render/caching and PDF/export control flow,
    • decision tables and invariants,
    • agent-facing maintenance workflow.
  • See UML.md for subsystem-level architecture, class relationships, and activity diagrams.

Project Structure

mdexplore.py                  # Qt application, file tree, renderer integration
mdexplore.sh                  # launcher (venv create/install/run)
setup-mdexplore.sh            # full bootstrap script (venv/assets/mmdr)
mdexplore.desktop.sample      # sample desktop launcher entry for user customization
mdexplore_app/                # extracted support modules for search, tree, tabs, icons, PDF, runtime, template/JS assets, and workers
tests/                        # headless/unit regressions for preview, search, template assets, and tab layout
assets/js/                    # externalized preview/PDF JavaScript templates loaded into a startup registry at startup
assets/js/pdf/                # PDF export/preflight/restore JavaScript templates
assets/js/preview/            # preview search/highlight/selection/context-menu JavaScript templates
assets/templates/             # externalized HTML preview document templates loaded into a startup registry
assets/templates/preview/     # preview HTML shell/template assets
assets/ui/                    # local UI icons/font assets used by tree, tabs, copy controls, and search pills
mdexplor-icon.png             # primary app icon asset kept at repo root for desktop launchers
requirements.txt              # Python runtime dependencies
README.md                     # project docs
DEVELOPERS-AGENTS.md          # developer and coding-agent maintenance guide
UML.md                        # subsystem-level architecture/class/activity diagrams
LICENSE                       # MIT license

Development

Basic local checks:

bash -n mdexplore.sh
python3 -m py_compile mdexplore.py

Troubleshooting

If you see ModuleNotFoundError: No module named 'PySide6.QtWebEngineWidgets':

  • Re-run mdexplore.sh; if runtime import verification is triggered and fails, it auto-reinstalls dependencies.
  • If verification was previously stamped but the venv later drifted, force a recheck by removing:
    • /path/to/mdexplore/.venv/.runtime-import-check.sha256 then run:
    • /path/to/mdexplore/mdexplore.sh
  • If it still fails, run:
    • rm -rf /path/to/mdexplore/.venv
    • /path/to/mdexplore/mdexplore.sh

If Edit does nothing:

  • Ensure MarkText is installed.
  • Verify /usr/bin/marktext exists and is executable.

If the dock/menu still shows an old app icon:

  • Ensure your launcher contains:
    • Icon=/absolute/path/to/mdexplor-icon.png
    • StartupWMClass=mdexplore
  • Refresh desktop entries:
    • update-desktop-database ~/.local/share/applications
  • Remove old pinned mdexplore favorites from the dock and pin it again.
  • Log out/in if the shell still shows the previous cached icon.

If running the launcher appears to do nothing:

  • Watch terminal output; the launcher now prints setup/launch status.
  • First dependency install can take time because Qt packages are large.
  • Ensure you are in a GUI session (DISPLAY or WAYLAND_DISPLAY must be set).
  • For desktop/dock launches (Terminal=false), check launcher log:
    • ~/.cache/mdexplore/launcher.log
    • log is auto-trimmed to the most recent 1000 lines

Security Notes

  • Markdown HTML is enabled (html=True), so preview untrusted Markdown with care.
  • Mermaid uses local-first loading with CDN fallback.
  • MathJax uses local-first loading with CDN fallback.

Contributing

  1. Keep changes focused and small.
  2. Run syntax checks before submitting.
  3. Update docs (README.md, DEVELOPERS-AGENTS.md, and UML.md when relevant) when behavior changes.

License

This project is licensed under the MIT License. See LICENSE.

About

A high-speed desktop Markdown browser suited for RAG document curation that uses Python and PySide6 and allows highlighting and bulk copying. Rapidly renders Mermaid, PlantUML, MathJax and BASE64 images with extensive caching. Like File Explorer for heavy duty Markdown file previewing.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors