Repository navigation
Remove QuestPDF — its Community licence is pay-to-use above a revenue threshold #1230
Description
MeshWeaver ships dual-licensed Apache-2.0 / MIT. QuestPDF (Directory.Packages.props, 2026.7.2) is dual-licensed: the Community tier is free only below a revenue threshold; above it a paid commercial licence is required. MarkdownExportExtensions.cs:60 accepts that Community licence programmatically at mesh-builder time, and MeshWeaver.Blazor pulls it transitively, so QuestPDF ships in every deployed portal image.
The user's decision (2026-08-11) is to dereference it — not to buy a licence and not to swap in another restrictively-licensed package. PDF rendering moves to the browser via the existing Pixel/HeadlessChromiumPdfRenderer.cs (Playwright, MIT).
Footprint (verified on main)
src/MeshWeaver.Markdown.Export/Pdf/PdfDocumentRenderer.cs— 368 lines, 100% of the QuestPDF rendering, the only file withusing QuestPDF.src/MeshWeaver.Markdown.Export/Configuration/MarkdownExportExtensions.cs:60—Settings.License = LicenseType.Community.src/MeshWeaver.Markdown.Export/MeshWeaver.Markdown.Export.csproj:8— the onlyPackageReference.- Tests using it:
PdfHeaderLogoTests.cs,RendererOutputTests.cs. ExportFidelity.cs,ExportFormat.cs,Branding/ExportTemplateResolver.csmention it in comments only — no compile dependency.
Why it is not a one-line deletion
Fork point Templates/ExportPdf.csx:126: Pixel fidelity goes to Chromium (Deck only); ExportPdf.csx:193-196 warns and falls back to QuestPDF for every non-Deck node. Per PixelFaithfulExport.md:70 the Chromium path today has no cover page, no TOC, no running header/footer and no page-break rules — those exist only in PdfDocumentRenderer.cs. README.md:23 also states headless Chromium is deliberately not shipped in the portal image.
So the deletion must land with the document-export migration onto Chromium, otherwise it is a functional regression rather than a cleanup.
Done when
- Markdown→PDF, cover page, TOC and running header/footer render through the browser path
-
PdfDocumentRenderer.cs, the licence line and thePackageReferenceare deleted -
PdfHeaderLogoTests/RendererOutputTestsassert the same behaviour against browser-rendered output (do not weaken them) -
<PackageVersion Include="QuestPDF" …>removed fromDirectory.Packages.props - the
questpdfentry is deleted fromEXCEPTIONSin.github/scripts/check-licenses.py— the licence gate then enforces this permanently
Activity
rbuergi commentedon Aug 11, 2026
Hard prerequisite: Chromium is not in the deployed image
Confirmed while auditing dependency licences (#1233):
deploy/base-images/portal-ai/Dockerfile— the base ofmemex-portal-aipermain-cd.yml— does not include a browser, and that is a stated design choice rather than an oversight.That changes this issue's shape:
Sequencing
#1234 fixed the real defect underneath this at its root:
Ast/DocumentBuilder.csbuilt its own Markdig pipeline that omittedLayoutAreaMarkdownExtension, so@@(…)parsed as a paragraph and printed as raw source; the pixel path emitted the empty layout-area anchor, but its print document isfile://underdefault-src 'none'with--host-resolver-rules=MAP * ~NOTFOUND, so the browser could never fill it and printed blank. It also lifted the resolver/renderer/table-sizer/sanitizer out ofEmail/into a format-neutralHtml/.This issue stays open and blocked on that chain. Meanwhile QuestPDF is carried as a dated, reasoned exception in
EXCEPTIONSin.github/scripts/check-licenses.py(added by #1233), so the licence gate enforces everything else today while this stays visible instead of silently permanent. Closing this issue includes deleting that entry.fix(ci): give the headless browser back its sandbox on the runner
docs(export): the portal-ai image ships the browser now
review: resolve the browser binary without walking the whole tree
fix(export): restore PDF contents page numbers by measuring the print