Editing PsionNet (section)

=== The proxy ===

Requests arrive as ordinary HTTP. The proxy fetches the page over modern TLS and then:<ref name="pn-readme" />
* removes hidden markup (<code>hidden</code>, <code>aria-hidden</code> and inline <code>display:none</code>), which on BBC News is 52% of the page;
* rewrites the page to HTML 3.2 for EPOC, against element and attribute lists read out of the Series 7 ROM's own parser tables. The netBook Pro gets HTML 4.01 and keeps its stylesheets;
* converts images to non-interlaced GIF or baseline JPEG, the only formats the devices decode;
* reduces text for EPOC to the characters CP1252 and ISO-8859-1 share. The netBook Pro reads UTF-8, so its text is left alone;
* filters CSS for the netBook Pro to the subset Internet Explorer 6 implements, then drops every <code>class</code> that no remaining rule uses;
* shortens same-site links, which are a quarter to a half of the bytes in a typical page.

{| class="wikitable styled-table" style="width:80%; text-align:center;"
|+'''Pages before and after the proxy, as sent to a Series 7'''
! Page !! From the site !! To the Psion !! Time on the serial link
|-
| BBC News || 1,041 KB || 31 KB || 3.0 s
|-
| The Guardian || 1,496 KB || 48 KB || 4.7 s
|-
| Wikipedia article || 100 KB || 24 KB || 2.3 s
|-
| Hacker News || 34 KB || 19 KB || 1.8 s
|}

The proxy chooses how to shape a page from the browser's User-Agent, so there is nothing to set on the device. An unknown browser gets the EPOC output, the strictest, because a page simple enough for EPOC still works in a richer browser.<ref name="pn-proxy" />

{| class="wikitable styled-table" style="width:100%; text-align:left;"
|+'''Output for each browser'''
! !! Series 5mx, 7, netBook, Revo !! netBook Pro, Windows CE !! netBook Pro, PsionLX
|-
| Browser || STNC WTL 2.0, in ROM || Pocket Internet Explorer || Firefox 1.0 (Gecko 1.7)
|-
| HTML || 3.2 || 4.01 Transitional || 4.01 Transitional
|-
| Character set || ISO-8859-1 || UTF-8, untouched || UTF-8, untouched
|-
| CSS || Removed || Filtered to the IE6 subset || Filtered to the IE6 subset
|-
| Images || GIF and baseline JPEG || GIF and baseline JPEG; PNG did not render in a test on the device || GIF and baseline JPEG
|-
| Page budget || 64 KB || 192 KB || 400 KB, of which CSS 96 KB
|}

The Windows CE output was worked out with a test page served to a real netBook Pro. Forwarding BBC News's 190 KB of inline CSS would take about a hundred seconds at 19200 bit/s for rules the device throws away; the filter cuts it to 32 KB.<ref name="pn-proxy" />