vite.config: optimize memory and fix symlink loop by pietrushnic · Pull Request #23 · 3mdeb/slidev-template

pietrushnic · 2025-10-12T21:31:31Z

Disable HMR error overlay to reduce memory consumption
Add file watching optimizations (disable polling, 1s interval)
Ignore slides/tools/** and slides/slidev-template/** patterns to prevent infinite symlink loop in file watcher (ELOOP errors)
Configure vendor chunk splitting to reduce memory pressure
Exclude @slidev/cli from pre-bundling optimization
Add build-time rollup optimizations for chunking

scripts/render-slides: expose GC API for better memory management

Add --expose-gc to NODE_OPTIONS to allow manual garbage collection
Helps Vite/Slidev plugins manage memory more effectively

These changes address memory exhaustion issues when running Slidev dev server with Vite 7, which has known HMR memory leaks.

Add test infrastructure for validating slidev-template features: - playwright.config.ts: Playwright test configuration - tests/smoke.spec.ts: Validates render-slides.sh and gen_slides.sh - tests/visual-regression.spec.ts: Screenshot comparison for layouts Tests cover PR #22 (docker helpers) and PR #23 (vite memory fixes): - Dev server startup validation - Symlink creation for /slides/ paths - Memory limit handling - Visual regression for slide layouts Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

pietrushnic · 2026-01-23T10:44:16Z

Those are partially recommended partially AI-researched changes. I was using those changes for quite some time without any memory issues or loops for various sizes of presentations. So question what should be an evidence here to merge this?

m-iwanicki · 2026-01-23T10:54:21Z

Does polling has to be disabled? I need to restart whole slidev to see updated slides, unless there is better method?

pietrushnic · 2026-01-23T13:07:14Z

Does polling has to be disabled? I need to restart whole slidev to see updated slides, unless there is better method?

It really depends on the change, methods I use for refresh:

for serious CSS changes or if I really can't see any update (e.g. change of image and cache reset doesn't work) I just disable
I hit r in slidev render-slides.sh terminal
hit ctrl+r in web browser
do nothing - it works very good in most cases

I'm not sure in what part of developer workflow you see the effect you mention.

m-iwanicki · 2026-01-23T13:10:42Z

I'm not sure in what part of developer workflow you see the effect you mention.

Even simple letter change requires me to restart slidev (point 2)

Add Playwright test that modifies slide content on disk and verifies the browser picks up the change without a manual page reload. This addresses m-iwanicki's review on PR #23 where usePolling: false broke HMR. The previous commit removes the explicit usePolling setting, letting Vite auto-detect. This test ensures HMR keeps working. Mount test-repo volume into Playwright container so the HMR test can write to the slide file that the dev server watches. Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

pietrushnic · 2026-02-24T11:30:11Z

@m-iwanicki I tried to fix and validated what you see. I added tests to make sure feature will not regress:

% SLIDEV_PORT=8002 ./scripts/run-tests.sh
Setting up test presentation repo...
HEAD is now at 4a079d568381 test: add HMR broken fixture for regression proof
Test repo ready
Starting dev server on port 8002...
Waiting for dev server to be ready...
Dev server is ready
Running Playwright tests...

Running 21 tests using 1 worker

  ✓   1 [chromium] › tests/hmr.spec.ts:23:3 › Hot Module Replacement › slide content updates after file change without reload (6.9s)
  ✓   2 [chromium] › tests/smoke.spec.ts:17:3 › Dev Server › responds on configured port (461ms)
  ✓   3 [chromium] › tests/smoke.spec.ts:22:3 › Dev Server › theme loads without console errors (2.6s)
  ✓   4 [chromium] › tests/smoke.spec.ts:44:3 › Dev Server › src: directive renders content (1.5s)
  ✓   5 [chromium] › tests/smoke.spec.ts:55:3 › Navigation › all slides load without OOM (15.2s)
  ✓   6 [chromium] › tests/smoke.spec.ts:64:3 › Navigation › presenter mode accessible (467ms)
  ✓   7 [chromium] › tests/smoke.spec.ts:69:3 › Navigation › overview mode accessible (443ms)
  ✓   8 [chromium] › tests/smoke.spec.ts:76:3 › Assets › images load without errors (2.0s)
  ✓   9 [chromium] › tests/smoke.spec.ts:92:3 › Keyboard Navigation › vim keys navigate slides (h/l) (2.8s)
  ✓  10 [chromium] › tests/smoke.spec.ts:110:3 › Keyboard Navigation › vim keys navigate slides (j/k) (2.9s)
  ✓  11 [chromium] › tests/visual-regression.spec.ts:13:3 › Layouts › cover (2.3s)
  ✓  12 [chromium] › tests/visual-regression.spec.ts:23:3 › Layouts › two-cols (2.2s)
  ✓  13 [chromium] › tests/visual-regression.spec.ts:33:3 › Layouts › two-cols-header (2.1s)
  ✓  14 [chromium] › tests/visual-regression.spec.ts:43:3 › Layouts › quote (2.1s)
  ✓  15 [chromium] › tests/visual-regression.spec.ts:55:3 › Components › figure with figcaption (2.1s)
  ✓  16 [chromium] › tests/visual-regression.spec.ts:67:3 › Components › Footnotes (2.2s)
  ✓  17 [chromium] › tests/visual-regression.spec.ts:78:3 › Components › table (2.1s)
  ✓  18 [chromium] › tests/visual-regression.spec.ts:91:3 › Footer › visible on content slides (1.5s)
  ✓  19 [chromium] › tests/visual-regression.spec.ts:99:3 › Footer › hidden on cover slides (1.5s)
  ✓  20 [chromium] › tests/visual-regression.spec.ts:107:3 › Presenter Mode › renders with notes panel (2.5s)
  ✓  21 [chromium] › tests/visual-regression.spec.ts:119:3 › Font Size Classes › code-10px and code-12px apply correctly (2.2s)

  21 passed (59.4s)
All tests passed!
Cleaning up...
slidev-test-server-1070693

Add Playwright test that modifies slide content on disk and verifies the browser picks up the change without a manual page reload. This addresses m-iwanicki's review on PR #23 where usePolling: false broke HMR. The previous commit removes the explicit usePolling setting, letting Vite auto-detect. This test ensures HMR keeps working. Mount test-repo volume into Playwright container so the HMR test can write to the slide file that the dev server watches. Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

m-iwanicki · 2026-02-24T11:41:04Z

Hmmm, still doesn't work for me, even opened slide in new browser just to make sure it's not cache fault. How I tested it:

Clone zarhus/presentations and it's submodules
Checkout feature/vite-memory-fixes (in slidev-template)
Render slide e.g.: ./slidev-template/scripts/render-slides.sh pages/zdm_3/dts.md
Open slides in your browser: http://localhost:8000/
Modify text in any slide in dts.md and check if it'll change in browser

Results:

No changes, not even when refreshing or opening slides in new browser.

- Disable HMR error overlay to reduce memory consumption - Add file watching optimizations (disable polling, 1s interval) - Ignore slides/tools/** and slides/slidev-template/** patterns to prevent infinite symlink loop in file watcher (ELOOP errors) - Configure vendor chunk splitting to reduce memory pressure - Exclude @slidev/cli from pre-bundling optimization - Add build-time rollup optimizations for chunking scripts/render-slides: expose GC API for better memory management - Add --expose-gc to NODE_OPTIONS to allow manual garbage collection - Helps Vite/Slidev plugins manage memory more effectively These changes address memory exhaustion issues when running Slidev dev server with Vite 7, which has known HMR memory leaks. Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

Add Playwright test that modifies slide content on disk and verifies the browser picks up the change without a manual page reload. This addresses m-iwanicki's review on PR #23 where usePolling: false broke HMR. The previous commit removes the explicit usePolling setting, letting Vite auto-detect. This test ensures HMR keeps working. Mount test-repo volume into Playwright container so the HMR test can write to the slide file that the dev server watches. Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

Add vite-config-hmr-disabled.ts fixture that sets hmr: false in vite.config.ts. When applied, the HMR test correctly fails, proving it detects when file changes stop propagating to the browser. Verified: ./scripts/run-tests.sh broken hmr → PROOF COMPLETE Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

The `slides -> ..` symlink causes chokidar to report file changes through recursive symlink-traversed paths instead of canonical paths. Setting `followSymlinks: false` ensures Slidev's watchFiles map matches the reported paths, fixing HMR for external files. Also adds test-hmr-external.sh that reproduces the real developer workflow (host file edit -> Docker dev server -> browser HMR) and reduces broken fixture test verbosity (quiet on pass, verbose on fail). Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

sed -i and git checkout replace file inodes, causing chokidar's inotify watches to break on subsequent edits. Polling via fs.stat() is immune to inode changes and is the standard Docker workaround. Signed-off-by: Piotr Król <piotr.krol@3mdeb.com>

pietrushnic · 2026-03-03T12:40:08Z

@m-iwanicki @macpijan I did it one more time, tested carefully:

in zarhus/presentation: SLIDEV_PORT=8011 ./slidev-template/scripts/render-slides.sh pages/zdm_3/1-greetings-agenda.md
Open localhost:8011/2 - check content

In other terminal: sed -i 's/Welcome/CHANGED Welcome/' ~/src/3mdeb/zarhus/presentations/pages/zdm_3/1-greetings-agenda.md

git checkout -f .

Test results:

% SLIDEV_PORT=8082 ./scripts/run-tests.sh
Setting up test presentation repo...
HEAD is now at 4627c876183f fix: use polling for reliable HMR in Docker bind mounts
Test repo ready
Starting dev server on port 8082...
Waiting for dev server to be ready...
Dev server is ready
Running Playwright tests...

Running 21 tests using 1 worker

  ✓   1 [chromium] › tests/hmr.spec.ts:23:3 › Hot Module Replacement › slide content updates after file change without reload (6.2s)
  ✓   2 [chromium] › tests/smoke.spec.ts:17:3 › Dev Server › responds on configured port (535ms)
  ✓   3 [chromium] › tests/smoke.spec.ts:22:3 › Dev Server › theme loads without console errors (3.6s)
  ✓   4 [chromium] › tests/smoke.spec.ts:44:3 › Dev Server › src: directive renders content (1.5s)
  ✓   5 [chromium] › tests/smoke.spec.ts:55:3 › Navigation › all slides load without OOM (16.8s)
  ✓   6 [chromium] › tests/smoke.spec.ts:64:3 › Navigation › presenter mode accessible (531ms)
  ✓   7 [chromium] › tests/smoke.spec.ts:69:3 › Navigation › overview mode accessible (1.1s)
  ✓   8 [chromium] › tests/smoke.spec.ts:76:3 › Assets › images load without errors (2.1s)
  ✓   9 [chromium] › tests/smoke.spec.ts:92:3 › Keyboard Navigation › vim keys navigate slides (h/l) (2.9s)
  ✓  10 [chromium] › tests/smoke.spec.ts:110:3 › Keyboard Navigation › vim keys navigate slides (j/k) (2.9s)
  ✓  11 [chromium] › tests/visual-regression.spec.ts:13:3 › Layouts › cover (2.2s)
  ✓  12 [chromium] › tests/visual-regression.spec.ts:23:3 › Layouts › two-cols (3.4s)
  ✓  13 [chromium] › tests/visual-regression.spec.ts:33:3 › Layouts › two-cols-header (2.7s)
  ✓  14 [chromium] › tests/visual-regression.spec.ts:43:3 › Layouts › quote (2.2s)
  ✓  15 [chromium] › tests/visual-regression.spec.ts:55:3 › Components › figure with figcaption (2.2s)
  ✓  16 [chromium] › tests/visual-regression.spec.ts:67:3 › Components › Footnotes (2.2s)
  ✓  17 [chromium] › tests/visual-regression.spec.ts:78:3 › Components › table (2.2s)
  ✓  18 [chromium] › tests/visual-regression.spec.ts:91:3 › Footer › visible on content slides (1.4s)
  ✓  19 [chromium] › tests/visual-regression.spec.ts:99:3 › Footer › hidden on cover slides (1.6s)
  ✓  20 [chromium] › tests/visual-regression.spec.ts:107:3 › Presenter Mode › renders with notes panel (2.6s)
  ✓  21 [chromium] › tests/visual-regression.spec.ts:119:3 › Font Size Classes › code-10px and code-12px apply correctly (2.2s)

  21 passed (1.1m)
All tests passed!
Cleaning up...
slidev-test-server-3316854
[13:37:27 pietrushnic@ollama ~/src/3mdeb/zarhus/presentations/slidev-template(feature)]
% SLIDEV_PORT=8082 ./scripts/run-tests.sh broken
Running 12 broken fixture tests...

✓ src-directive
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ images
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ cover
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ two-cols
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ two-cols-header
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ quote
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ figure
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ footnotes
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ table
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ footer-visible
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ footer-hidden
slidev-test-server-3318059
Cleaning up...
Cleanup complete
✓ hmr
slidev-test-server-3318059
Cleaning up...
Cleanup complete

All 12 broken fixture tests passed ✓
Cleaning up...

macpijan

@pietrushnic It works the same for me. I get instant change in Firefox browser when changing sources as you do in your scenario.

macpijan · 2026-03-03T13:07:13Z

Tets pass for me as well just fine.

m-iwanicki · 2026-03-04T08:23:22Z

@pietrushnic

I did it one more time, tested carefully

Yes, now it works 👍

pietrushnic requested a review from m-iwanicki October 12, 2025 21:31

pietrushnic force-pushed the scripts-improve-docker-helpers branch from c3707bf to ca45381 Compare November 22, 2025 00:03

pietrushnic force-pushed the feature/vite-memory-fixes branch from ae74262 to 718bc58 Compare November 22, 2025 00:05

pietrushnic force-pushed the scripts-improve-docker-helpers branch from ca45381 to afa8537 Compare January 21, 2026 14:29

pietrushnic force-pushed the scripts-improve-docker-helpers branch 2 times, most recently from f491e8a to ae8bb22 Compare January 21, 2026 14:47

Base automatically changed from scripts-improve-docker-helpers to main January 21, 2026 15:40

pietrushnic force-pushed the feature/vite-memory-fixes branch from 718bc58 to 6ec86f7 Compare January 23, 2026 09:31

pietrushnic requested a review from macpijan January 23, 2026 09:32

pietrushnic force-pushed the feature/vite-memory-fixes branch from 6ec86f7 to 2112c55 Compare February 24, 2026 08:38

pietrushnic changed the base branch from main to fix/visual-regression-baselines February 24, 2026 09:18

pietrushnic force-pushed the feature/vite-memory-fixes branch from 2112c55 to 4a079d5 Compare February 24, 2026 11:30

Base automatically changed from fix/visual-regression-baselines to main March 2, 2026 19:38

pietrushnic mentioned this pull request Mar 3, 2026

Uml rendering #33

Merged

pietrushnic added 5 commits March 3, 2026 12:42

pietrushnic force-pushed the feature/vite-memory-fixes branch from 4a079d5 to a17cdee Compare March 3, 2026 12:51

macpijan approved these changes Mar 3, 2026

View reviewed changes

pietrushnic merged commit 9f92c1c into main Mar 3, 2026

pietrushnic deleted the feature/vite-memory-fixes branch March 3, 2026 14:31

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

vite.config: optimize memory and fix symlink loop#23

vite.config: optimize memory and fix symlink loop#23
pietrushnic merged 5 commits intomainfrom
feature/vite-memory-fixes

pietrushnic commented Oct 12, 2025

Uh oh!

pietrushnic commented Jan 23, 2026

Uh oh!

m-iwanicki commented Jan 23, 2026

Uh oh!

pietrushnic commented Jan 23, 2026

Uh oh!

m-iwanicki commented Jan 23, 2026

Uh oh!

pietrushnic commented Feb 24, 2026

Uh oh!

m-iwanicki commented Feb 24, 2026

Uh oh!

pietrushnic commented Mar 3, 2026 •

edited

Loading

Uh oh!

macpijan left a comment

Uh oh!

macpijan commented Mar 3, 2026

Uh oh!

m-iwanicki commented Mar 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

pietrushnic commented Oct 12, 2025

Uh oh!

pietrushnic commented Jan 23, 2026

Uh oh!

m-iwanicki commented Jan 23, 2026

Uh oh!

pietrushnic commented Jan 23, 2026

Uh oh!

m-iwanicki commented Jan 23, 2026

Uh oh!

pietrushnic commented Feb 24, 2026

Uh oh!

m-iwanicki commented Feb 24, 2026

Uh oh!

pietrushnic commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

macpijan left a comment

Choose a reason for hiding this comment

Uh oh!

macpijan commented Mar 3, 2026

Uh oh!

m-iwanicki commented Mar 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

pietrushnic commented Mar 3, 2026 •

edited

Loading