fix: track sc/scn color operators to avoid skipping visible text

PDFs that set fill color via `sc`/`scn` (named color space operators)
instead of `g`/`rg`/`k` were incorrectly keeping fill_is_white=true
from earlier page background fills, causing all text to be skipped.

Infer color model from operand count: 3=RGB, 4=CMYK, other=non-white.

Fixes extraction of tension_chart PDF (D'Addario string specifications)
which previously output only glyph names.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Abimael Martell
2026-03-04 15:04:18 -08:00
co-authored by Claude Opus 4.6
parent ccd833dd59
commit d5e03b8915
2 changed files with 36 additions and 0 deletions
+23
View File
@@ -180,6 +180,29 @@ pub(crate) fn extract_page_text_items(
fill_is_white = c < 0.05 && m < 0.05 && y < 0.05 && k < 0.05;
}
}
"sc" | "scn" => {
// Set fill color in current color space.
// Infer color model from operand count (ignoring trailing
// pattern-name operands that are not numbers).
// Note: 1-operand form is ambiguous (could be Separation,
// ICCBased, etc.) so we only detect white for 3/4 operands
// where the color model is almost certainly RGB/CMYK.
let nums: Vec<f32> = op.operands.iter().filter_map(get_number).collect();
match nums.len() {
3 => {
fill_is_white = nums[0] > 0.95 && nums[1] > 0.95 && nums[2] > 0.95;
}
4 => {
fill_is_white =
nums[0] < 0.05 && nums[1] < 0.05 && nums[2] < 0.05 && nums[3] < 0.05;
}
_ => {
// 1-operand or unknown — could be any color space,
// assume non-white to avoid hiding visible text.
fill_is_white = false;
}
}
}
"BT" => {
// Begin text block
in_text_block = true;
+13
View File
@@ -218,6 +218,19 @@ pub(crate) fn extract_form_xobject_text(
fill_is_white = c < 0.05 && m < 0.05 && y < 0.05 && k < 0.05;
}
}
"sc" | "scn" => {
let nums: Vec<f32> = op.operands.iter().filter_map(get_number).collect();
match nums.len() {
3 => {
fill_is_white = nums[0] > 0.95 && nums[1] > 0.95 && nums[2] > 0.95;
}
4 => {
fill_is_white =
nums[0] < 0.05 && nums[1] < 0.05 && nums[2] < 0.05 && nums[3] < 0.05;
}
_ => fill_is_white = false,
}
}
"Tj" => {
if in_text_block && !op.operands.is_empty() {
if fill_is_white {