Treat \\<space> as non-breaking space in inline text (issue #162, bd-1aip)

[cscheid]

Fixes #162.

Summary

A literal \ followed by an ASCII space inside inline text was being absorbed into a single pandoc_str token together with the space (and any following text), producing AST shapes that drifted on round trip — a \ b parsed as [Str "a", Space, Str "\ b"] and re-parsed as [Str "a", Space, Str "\", Space, Str "b"] after one cycle. Reporter linked three quarto-web docs that hit this in pipe-table cells (cell padding inserts the space).

This change adopts Pandoc's non-breaking-space shorthand: \<ASCII space> collapses to a single U+00A0 (NBSP). The writer already emits NBSP literally as its UTF-8 bytes, and the existing grammar already parses literal NBSP correctly, so the round trip is now stable end-to-end with no grammar regen and no writer changes — the entire fix is one extra branch in process_backslash_escapes (crates/pampa/src/pandoc/treesitter_utils/text_helpers.rs).

This changes the AST for inputs containing \<space> in inline text: previously a single Str containing literal \<space>X, now a Str containing U+00A0 X (matching Pandoc). Consumers reading raw Str text will see a different character, but the source rendering is unchanged.

End-to-end verification

$ printf -- 'a \\ b\n' | cargo run -q --bin pampa
[ Para [Str "a", Space, Str "\u{a0}b"] ]

$ printf -- 'a \\ b\n' | cargo run -q --bin pampa -- -t qmd | xxd
00000000: 6120 c2a0 620a                           a ..b.

$ printf -- 'a \\ b\n' | cargo run -q --bin pampa -- -t qmd \
    | cargo run -q --bin pampa
[ Para [Str "a", Space, Str "\u{a0}b"] ]

Round-trip stable, byte-for-byte equal to Pandoc's markdown -> markdown output for the same input.

Tests

• Three unit tests in crates/pampa/tests/test_treesitter_refactoring.rs:
  • test_backslash_space_becomes_nbsp — reporter's exact case.
  • test_backslash_space_at_paragraph_start — \<space> at the start of inline content.
  • test_backslash_space_in_table_cell_round_trips — pipe-table cell case (the reporter's wild incidence). Compares native AST text rather than going through the JSON-roundtrip suite — see "Side find" below.
• Two new round-trip fixtures under crates/pampa/tests/roundtrip_tests/qmd-json-qmd/ for stability: inline_backslash_space.qmd and inline_backslash_space_paragraph_start.qmd.
• Existing test_escaped_backslash_then_space_remains_literal locks in that \\<space> (escaped backslash + space) is unchanged and does NOT produce NBSP.

cargo xtask verify --skip-hub-build --skip-hub-tests --skip-trace-viewer-build --skip-trace-viewer-tests is clean.

Scope

Only \<ASCII space> for now. Pandoc treats \<TAB> and \<NBSP> similarly, but those didn't surface in the reporter's bug or the linked docs, so they're left out of this change.

\<EOL> remains a LineBreak (different code path, unchanged).

Side find — bd-j9wp

While building the table-cell test, I tried to use a qmd-json-qmd round-trip fixture and the test failed even though the round-trip is content-stable. The codebase is fully deterministic — verified across 5 repeated runs that the same input always produces the same JSON1 (captionS: 17) and the same JSON3 (captionS: 15). The mismatch is real and reproducible.

What's happening: each parse builds its own astContext.sourceInfoPool (23 entries for the original qmd, 21 for the regenerated qmd — each pool is deterministic but the regenerated qmd has different source positions, so the pool contents and traversal-order IDs differ). The test's remove_location_fields (crates/pampa/tests/test.rs:395-411) scrubs astContext itself plus most foreign-keys-into-the-pool (s, attrS, targetS, citationIdS), but it misses scalar S-foreign-keys not nested inside a scrubbed envelope. captionS is one such miss. After scrubbing, captionS: 17 and captionS: 15 are both dangling references to the now-deleted pools, but the test still compares them numerically and fails.

I worked around this for #165 by switching the table case to a unit test that compares native AST text (no source info). The underlying scrub-list gap is filed as bd-j9wp (priority 3) — not a determinism issue, just a missing key in the comparison filter.

Test plan

• Confirm that the AST change (Str "\<space>X" → Str "<NBSP>X") doesn't break a downstream consumer that was working around the previous behavior. None expected, but worth a glance at quarto-web rendering of one of the three linked docs to be safe.

🤖 Generated with Claude Code

[cscheid]

(We're not currently rendering quarto-web, but nice of you to remind us, Claude)