Keep the space a whitespace-only span puts between two words

[kevin9327]

Wikipedia's HTML writes every   as <span typeof="mw:Entity">&nbsp;</span>, and oc glued the words on either side of it. oc wiki article "Python_(programming_language)":

main:    Firstappeared
         20February 1991 ; 35 years ago
         ... Python3.0, released in 2008, was a major revision ...
         Python implementation began in December1989.
branch:  First appeared
         20 February 1991 ; 35 years ago
         ... Python 3.0, released in 2008, was a major revision ...
         Python implementation began in December 1989.

That is 43 of the article's 2,368 blocks. On the Barack Obama article it is 143 of 7,793 blocks, among them $1.6million and January 20, 2009–January 20, 2017, which the page shows with spaces.

Cause. A text fragment remembers whether space touched its edges (pre/post), so mergeText can glue the pieces a parser split out of one word and put a space between the rest. In distill's walk (src/distill.js), a text node that is only whitespace has no text once collapsed and was skipped without a trace. The fragments around the span share a parent and neither touches a space itself, so they looked like one word.

Change. A whitespace-only text node now sets post on the text block just before it, which is what it was on the page. If the last block is a break or not text, nothing changes. Nothing is added to the output except the missing spaces.

Test. tests/distill.test.js gains one test with three shapes: the infobox label and date above, and ten<b> </b>items. On main it fails with Firstappeared, 20February 1991, tenitems. No fixture page in tests/pages/ changes a single block. Full suite on Windows (Node 24): 316 tests, 313 pass, 3 skipped, 0 fail.

Left alone. A fragment in another element with no space between (1991<span>;</span>) is still joined with a space, as it was; that rule is not about whitespace nodes. No CHANGELOG line, to keep this round's PRs from conflicting there.

🤖 Generated with Claude Code