html_export.py

"""Self-contained review HTML. OCR text is escaped, never executed."""
import base64,html,json
from pathlib import Path
from display_language import english_message

def render(job,cfg):
    esc=lambda s:html.escape(str(s),quote=True)
    from context_export import page_context, adoption_notice
    sections=[]
    for idx in cfg['selected']:
        folder=Path(job)/f'page-{idx+1:05d}'
        if not (folder/'source.png').exists():continue
        image=base64.b64encode((folder/'source.png').read_bytes()).decode()
        error=json.loads((folder/'review-error.json').read_text()).get('error','') if (folder/'review-error.json').exists() else ''
        status='Suggestions available; approve or reject in the app' if (folder/'review.json').exists() else 'Not reviewed by Gemma'
        blocks=[]
        if (folder/'regions.json').exists():
            regions=json.loads((folder/'regions.json').read_text())['regions']
            for r in regions:
                blocks.append(f'<article><small>{esc(r["role"])} · Region {r["id"]+1} · Coordinates {esc(r["box"])}</small><pre>{esc(r["text"])}</pre></article>')
        else:
            text=(folder/'original.md').read_text() if (folder/'original.md').exists() else 'OCR incomplete'
            blocks=[f'<p>No region data; using the saved OCR reading order.</p><pre>{esc(text)}</pre>']
        context=page_context(folder)
        blocks.insert(0,'<article><h3>Text for model input</h3><p>'+esc(context['status'])+'</p><pre>'+esc(context['text'])+'</pre><h4>Applied changes</h4><pre>'+esc(json.dumps(context['edits'],ensure_ascii=False,indent=2))+'</pre><h4>Suggestions requiring review; not applied to the text</h4><pre>'+esc(json.dumps(context['pending'],ensure_ascii=False,indent=2))+'</pre><p>Unverified model notes: '+esc(' / '.join(context['notes']))+'</p></article>')
        sections.append(f'<section><h2>Page {idx+1}</h2><p>{esc(english_message(error or status))}</p><div class="pair"><img alt="Source page {idx+1}" src="data:image/png;base64,{image}"><div>{"".join(blocks)}</div></div></section>')
    return '<!doctype html><html lang="en"><meta charset="utf-8"><meta name="viewport" content="width=device-width"><title>OCR review document</title><style>body{font:16px/1.7 sans-serif;margin:2rem;background:#f5f5f5;color:#222}.pair{display:grid;grid-template-columns:1fr 1fr;gap:2rem}.pair img{width:100%;position:sticky;top:1rem;align-self:start}article{background:white;padding:1rem;margin-bottom:1rem}pre{white-space:pre-wrap;overflow-wrap:anywhere;font:inherit}small{color:#555}@media(max-width:800px){.pair{display:block}}</style><h1>Model input text, suggestions and source images</h1><p>Region order is detected automatically. Check it against the images. Original OCR is preserved. Model input text is original OCR or explicitly adopted content, including optional Quick mode candidates without manual verification. Unapproved suggestions are listed separately for review.</p><p>'+esc(adoption_notice(job,cfg))+'</p>'+''.join(sections)+'</html>'