deblank: find the pages that hold nothing, and drop them

[tannevaled]

BlankPages draws each page and counts the pixels that are not paper; Doc.DropBlank removes them and says which went, numbered as the reader saw them. pdfops deblank exposes both, with -list as the dry run.

Drawing is the only test that agrees with a reader: a page whose content stream is full of operators painting white, or drawing outside the media box, looks blank and is blank, and a page carrying a picture of nothing is blank too.

This closes the last of the three entries go-pdfkit/conformance's parity table had written down as not covered with a reason — and that reason was the specification:

remove-blank-pages. It needs a blank detector over rendered pixels, and a detector is a measurement, not a wiring job.

⛔ The default resolution was measured, and my first one was wrong

This started at 50 DPI, with a comment claiming that was "enough to see a full stop". Measured on A4:

	50 DPI	150 DPI
the page	414 × 585	1240 × 1755
one pixel	4.1e-06 of it	4.6e-07
an 8-point full stop	1 pixel	4 pixels
an 8-point comma	2 pixels	8 pixels

At 50 DPI a full stop is exactly the mark a speck of dust makes, so no threshold at that resolution can tell punctuation from dirt. TestMeasureTheInkLadder prints the whole ladder at 50, 100, 150 and 300, so a renderer change that moved either end fails there instead of silently reclassifying somebody's pages.

⛔ The zero value is the strictest setting

A page with a single inked pixel is kept. This is the input to deleting pages: a page wrongly called blank destroys content, while a blank page wrongly kept merely leaves a blank page. The scan threshold (-max-ink 1e-6, about two pixels — forgives a speck, keeps a full stop) is documented with its measurements rather than made the default, because whether that margin holds depends on the scanner, which this package cannot see.

Three more refusals: it will not call a page of solid colour blank (the test is distance from white, not uniformity — a measure of variance would delete a deliberate black separator sheet); it will not empty the document; and it does not count annotations as ink unless asked.

⛔ Two mutations survived the first suite, and both were real weaknesses

mutation	why nothing caught it
defaultTolerance = 0	every fixture rendered to exactly white, so no test could see the field at all and the doc comment's claim was unproven. Measured: a mark at 247/255 inks 2.2e-03 at tolerance zero and exactly nothing at twelve. That page is now asserted paper by default and ink at a tolerance of one level.
dropping the alpha channel	the only transparent pixel any test used was transparent black, whose colour channels are ink anyway. White-and-transparent is the case the rule is about, and it is now its own test.

Three fixtures were wrong before the code was

• a "solid page" made by stamping U+2588 FULL BLOCK drew nothing — Helvetica's encoding has no such character, so the page really was blank and the test was right to fail;
• fixture(t, 2) in the command's tests carries a content stream that is not drawing operators, so its pages render as blank paper. A fixture whose "content" is invisible is no fixture for a tool that measures ink;
• Picture(data, 1) means one image pixel per inch, so a 64-pixel square became a 64-inch page and the renderer refused it.

One rule removed, one nearly removed by mistake

DropBlank lost its own emptiness check: Bytes already refuses a document with no pages, and a second copy was a branch no test could reach. ⛔ I nearly removed the write's error check as unreachable too — Bytes also refuses a form field with nowhere to go, which is a non-empty document. That is the third time this session I have almost called reachable code dead.

Checks

100 % statement coverage on both packages with zero uncovered statements, go vet and gofmt clean. Six mutations caught.

🤖 Generated with Claude Code