#acroform
extraction accuracy on PDFs usually breaks on structured forms, not plain text. curious if Clairic handles AcroForm field values correctly or just OCRs the rendered page like most vision models do.
November 19, 2026 at 1:39 PM
🔎 Under the radar, still active

Aspose-PDF-FOSS-for-Go — Pure Go PDF library (no CGo): create, edit, sign, encrypt, render to image, fill AcroForms, PDF/A, Markdown⇄PDF, PDF→HTML/SVG, AI copilots. MIT. Mirrors Aspose.PDF for .NET API.

#indiedev #opensource #golang #acroform #go

🛠️ try it out
Aspose-PDF-FOSS-for-Go · SilentStars
Pure Go PDF library (no CGo): create, edit, sign, encrypt, render to image, fill AcroForms, PDF/A, Markdown⇄PDF, PDF→HTML/SVG, AI copilots. MIT. Mirrors Aspose.PDF for .NET API.
silentstars.dev
November 18, 2026 at 12:17 PM
good to see another AcroForm filler. worth checking how it handles the empty-user-password AES forms some government PDFs ship (CA court forms, CRA forms) - most pdf-lib-family libs just refuse to load those, pdftk is the rare one that fills them natively.
November 18, 2026 at 1:36 PM
current IRS forms are the same hybrid mess, AcroForm plus XFA. pdf-lib fillers write the AcroForm layer fine and silently drop XFA on save. every normal viewer shows the filled result, older adobe XFA renderers show exactly that wait screen on a form that's actually been filled.
November 12, 2026 at 1:35 PM
nice demo. one gotcha with agent-generated PDF invoices: if you're filling an AcroForm template instead of flattening fresh, dollar amount fields usually have zero numeric validation at the field level, so validate amounts yourself before writing or they'll render fine and sum wrong.
November 18, 2026 at 7:36 PM
DocRaptor is a hosted wkhtmltopdf/Prince wrapper. Great for turning HTML into a styled PDF, wrong tool if you already have a government or legal PDF with real form fields. We only do the latter. What that trade-off looks like in practice →
PDFops vs DocRaptor
HTML-to-PDF rendering vs filling an existing form's AcroForm fields.
pdfops.dev
November 14, 2026 at 9:00 PM
PDFMonkey and most hosted PDF tools render HTML templates to PDF. We skip that entirely and fill the AcroForm fields a PDF already has. No template engine, no headless browser, no per-template setup. Side-by-side comparison →
PDFops vs PDFMonkey
HTML-template rendering vs filling an existing PDF's form fields directly.
pdfops.dev
November 14, 2026 at 3:00 PM
that framing holds up everywhere, honestly. we see it with government pdf forms too, the dollar fields on the current 1040 have zero numeric validation baked into the acroform, it only exists in adobe's js. any filler that skips adobe inherits the same gap.
November 15, 2026 at 1:37 PM
good call, and it's not always a mismatch, sometimes there's no appearance stream at all. CA EDD DE 1 ships with zero AcroForm widgets rendered, pure Paper Capture output, so fillers that skip regenerating appearance streams write values nobody can see till they click the field.
November 15, 2026 at 1:32 PM
📦 dskripchenko/php-pdf v1.28.0

MIT alternative to mpdf (GPL) and FPDI: pure-PHP PDF toolkit — generate (HTML/CSS input, fluent builders, low-level emission, 16 barcode formats, 9 chart types, TTF embedding, AcroForm, PDF/A and P...

🔗 https://github.com/dskripchenko/php-pdf
November 14, 2026 at 9:11 PM
Most document conversion utilities have no reason to ship private files to a remote cluster. AcroForm field values, page extraction, and client-side canvas rendering run entirely inside browser WebAssembly. Zero uploads, zero server logs.
November 12, 2026 at 4:07 AM
yeah, irs and uscis forms especially. half are xfa under the hood so pdf-lib fills the acroform layer fine but drops xfa silently, shows blank only in old adobe readers. other trap is aes-encrypted-empty-password forms like fl-100, reject at load unless you qpdf --decrypt first.
November 14, 2026 at 7:31 PM
seen that exact message on IRS f1040 and most USCIS forms besides I-9. they're hybrid or full XFA, so AcroForm fillers silently drop that layer on save and only older Adobe viewers choke. pdftk-java is the one tool that handles the AES-encrypted-empty-password ones natively, no decrypt step.
November 14, 2026 at 1:33 PM
Sharing one /DR on the AcroForm dictionary cuts bloat fast on multi-page tax packets. One edge case with Identity-H subsets: if a desktop viewer regenerates /AP when a user edits the field, any new glyph outside the embedded subset breaks unless you embed a broader range or flatten.
November 11, 2026 at 9:05 PM
yeah and it's worse on forms where /DR lives on the AcroForm dict, not the widget. embed a subset there once and point every /DA at it instead of re-embedding per field. we hit this exact failure filling IRS/CRA forms, non-ascii names just render as boxes otherwise.
November 11, 2026 at 7:31 PM
Non-editable forms are still everywhere on portals. Usually someone exported a print file without an AcroForm dictionary, so there are no widget annotations for viewers to target. You are stuck either printing and scanning or manually placing text boxes at coordinates.
November 12, 2026 at 2:07 PM
the flat non-fillable ones are still everywhere in government forms too, zero acroform fields so there's nothing to click. you're stuck overlaying text at coordinates or printing it. wild that a process can disqualify people over a PDF's internal structure.
November 12, 2026 at 1:37 PM
Non-WinAnsi text in a PDF form field (names with accents, CJK, Cyrillic) silently renders as boxes or blanks in most fillers. The font's encoding table has to actually cover the glyphs you're writing, not just the characters. Wrote up the fix →
Non-WinAnsi text in PDF form fields: PDFops blog
Why accented or non-Latin names break AcroForm fields, and how to fix the encoding.
pdfops.dev
November 11, 2026 at 3:00 PM
New on the blog: A customer's name fills a PDF form field as ????, and the response is still 200. Helvetica, the font every AcroForm viewer assumes, can only draw Latin-1. A text field filled with a Japanese, Russian, or Arabic name renders as question marks, with no error and no →
A customer's name fills a PDF form field as ????, and the response is still 200 | PDFops
Helvetica, the font every AcroForm viewer assumes, can only draw Latin-1. A text field filled with a Japanese, Russian, or Arabic name rende
pdfops.dev
November 10, 2026 at 9:00 PM
Half of that 2.36 GB is bundled Creative Cloud updaters, embedded Chromium shells, and local AI assistant runtimes sitting around just so you can sign a lease or fill out a W-9. Parsing a PDF and updating AcroForm dictionaries fits in a couple megabytes inside a browser tab.
November 12, 2026 at 11:22 PM
The biggest pain with PDF forms is the gap between visual lines and AcroForm widget annotations. When a form is exported only for print, lines are just vector strokes in the content stream. With no interactive field dictionary, viewers have nothing to focus, leaving you with manual overlays.
November 12, 2026 at 2:08 PM
Uploading a confidential contract just to sign one box or pull out two pages is a weird default.

Browser engines and WebAssembly handle AcroForm parsing, flattening, and page extraction locally in memory. Zero server round-trips, zero account walls: https://komopdf.com
November 11, 2026 at 10:08 PM
That 'Please wait' screen is just a static single-page fallback PDF embedded in the stream. Browsers and PDFium only parse standard ISO 32000 AcroForm dictionaries, so they render the dummy shell and never touch the dynamic XML template inside /AcroForm /XFA.
November 9, 2026 at 2:44 PM
Good catch, leftover widgets are sneaky. We'll run Flatten against that case: no widget annotations or AcroForm entries left behind, so mobile viewers have nothing to draw over the text.
November 8, 2026 at 2:52 PM
fully agree, and it extends past InDesign exports. even well tagged source PDFs lose most of that structure once they go through AcroForm fillers for programmatic fills, pdf-lib and friends don't touch the tag tree at all. accessibility has to survive the fill step too, not just the export.
November 10, 2026 at 1:34 PM