pardes.git/src/pdf_view.zig, branch release-0.24

A PDF place `:P:H` whose search hit H is not on page P, or with no search, is a miss with an err, as a page past the last is

2026-10-01T11:53:11Z

The second number of a PDF place is a hit of the pane's search (its search rows are `file.pdf:P:H`), and one that named no hit turned to page P silently. pdf State.hitExists asks the pane's search for page P: a hit it does not have, or any hit with no search, is now ` has no search hit H on page P`, on the pane itself (`:P:H`) and from anywhere (`file.pdf:P:H`); a search row's hit is always there. Co-Authored-By: Claude Opus 5.5

A +PdfSections row's first number is the page: a section that is not there, or is on another page, is a miss said, not a silent success

2026-10-01T04:35:12Z

A look at `file.pdf:P:S` in +PdfSections went to section S whatever P said, and one past the last section did nothing and answered ok. The row's page is now held to: a section not in the outline, or on another page than the row names, fails with which, as every other miss does. Co-Authored-By: Claude Opus 5.5

A PDF page taller than one raster is drawn at the size it is shown, in bands around the screen, not squeezed into 4096 rows and blown up

2026-10-01T03:14:45Z

The render geometry (pdf_bridge.c) clamped a page's LONGEST side to max_dimension: 4096 in the SDL policy, 1200 in Kitty's. A page 841.89 x 4818.9 pt, fit to a 1570 px wide pane, came out 715 x 4096 and was drawn 2.2 times too large at 1x, 4.4 times at 2x; Kitty's raster was 210 px wide. Every request now caps a raster ROW at max_dimension, and caps the height too only while that keeps the page at least as large as it is shown (RenderRequest.display_width/height, set from the fit). A tall page fit to width is rasterized at the pane's own width, in SDL, and at Kitty's 96 dpi as any other page is there. Such a page is not one buffer. A raster over 8192 rows (a texture's and Kitty's 10000-pixel limit) or over 64 MiB is banded (pdf_view.zig): its slot holds the rows on screen and a screenful either side, in 512-row chunks, and is rendered again, keeping the chunks it has, once the screen comes within half a screenful of its edge. Each chunk comes from its own render reaching 32 rows past both edges, so a row of the page is the same however it was scrolled to, and no chunk shows its clip edge: band rows are within 3 levels of a whole-page render (108 pixels of 45 million on the reported PDF). Highlights paint over the band's clean rows as over a page's, and the band uses the page's CTM, so search, selection and pointer geometry are unchanged. A display-list render now culls to the band (a scissor in page space). No pixel changes, but a frame that renders new rows of the reported page at 2x fell from 33-43 ms to 13-16 ms (pdf-scroll-bench, --cell 16x32). design.pdf's scroll-bench pixels are identical. Co-Authored-By: Claude Opus 5.5

PdfSections on a PDF with no outline says so, "this PDF has no outline", rather than open an empty +PdfSections

2026-10-01T03:12:17Z

An empty pane with no word of why read as a failure to load, and was kept to be re-shown on the next press. With no outline there is nothing to list: the press is refused with the reason, and a read or slot failure is said too, where it returned silently. Co-Authored-By: Claude Opus 5.5

A look at a PDF's page past its last is a miss, "has no page N", never the last page in its place

2026-10-01T03:12:17Z

manual.pdf:999 on a twelve-page PDF showed page 12 and said nothing, so a stale or mistyped page number looked like a place found. It is now the miss a line past a file's end is, whether the PDF is open (it stays on its page) or not (nothing opens). A restore or a start-up page past the end still lands on the last. Co-Authored-By: Claude Opus 5.5

Dumps and init files from before 0.0.4 load no more of their old tag formats, and three orphan test scripts go

2026-10-01T03:12:17Z

A2: a restored pane's tag tail comes from the dump's tag_tail, which every dump since 0.0.4 writes; the 27 earlier default tails that were recognised to upgrade, restoreTail, tailClass, savedPrefix and restoreTailAt, pdf_view's and image's legacySavedPrefix, a terminal's TTY tag prefix and its raw output tail replay, and the ThemeSel, FontSel and ColumnTags words (init lines and dumped output names) go, with their tests; the CHANGELOG says so. A4: test/fs_soak.py, referenced nowhere (monkey9p.py does its work). A6: test/column_grip.py and test/look_placement.py, referenced nowhere; the six review scripts docs/ui-review.md names stay. Co-Authored-By: Claude Opus 5.5

Every catch unreachable, orelse unreachable and syscall assert outside tests is a real refusal or says why it cannot fire

2026-10-01T03:12:16Z

A sweep for round 23's crash: a run's answer (pty/run) was bufPrint'd into 48 bytes with catch unreachable, so a foreground program's long name (macOS gives up to 32 bytes) panicked; it now cuts at the room, keeping its newline, in 96 bytes. The rest were numbers into buffers sized for them, a braille codepoint, pthread calls on the queue's own mutex, and pdf_view's resolved outline entries: each now carries a one-line comment saying why it cannot fire. Co-Authored-By: Claude Opus 5.5

A column can be empty, as acme's can; new panes go where acme's makenewwindow puts them

2026-10-01T03:12:15Z

Newcol makes an empty column; closing a column's last pane leaves it empty with the keyboard on its tag; Delcol and Joincol alone take a column away; the session's last pane closing quits. The +New stand-in and replaceStillborn / stillborn_joiner are gone. Every new pane goes through exec.placeNew, which follows acme's makenewwindow (active column, empty column whole, blank space, halving the biggest) or, with Placement pardes, the old rules. ColumnTags is gone: column tags are always shown. A grip drag shows acme's box cursor. Co-Authored-By: Claude Opus 5.5

Place every part of the frame once, before painting it

2026-10-01T03:12:14Z

draw.place builds the Surface's region list at the head of render: the page; each pane's grip, tag rows, body, rail with its thumb, and notice chips; each column's grip and tag; the workspace tag. The painters read their geometry from it instead of working it out again: the grid's notice chips and the notice layers share one chip rectangle, the rails (file, terminal, image, PDF page) paint the placed thumb (a native PDF measures its own into the region as it places its pages), and the body, tag, notice and column layers take their viewports from their regions. A collapsed pane places only its grip and tag. The column-move rail's one-cell clip of the layers under it is now a region field set in place, not a mutation of the built layers. No behaviour change; a test checks every region against the old geometry, the painted rail and the layer viewports across tag bottom, column tags, notices, a collapsed pane and a multi-line tag.

Paint into a target Surface instead of swapping the frame's out

2026-10-01T03:12:14Z

The painters of a pane's body and tag (renderBody and its selection, gutter, syntax, ANSI and wrap-marker passes; paintPaneTag; paintHeaders; renderPane; the image and PDF grid paths) take the Surface they paint as a parameter. The body layer and the tag layers paint straight into their own temporary surfaces, and the std.mem.swap of p.surface around those calls is gone. No behaviour change.