
PDFiumPage::GetTextRunInfo() already has good heuristics to calculate the set of characters that form a text run. Use it in PDFiumRange::GetScreenRects() to create a rectangle for every text run. Then add more heuristics to merge the rectangles when they have sufficient overlap in the horizontal direction. As a result, the text selection, which is based on GetScreenRects() calculations, will appear more continuous and less jagged. Update test expectations to match the new code's behavior. Keep the original FPDFText_CountRects()-based implementation around, behind an emergency kill switch, in case the new code has serious problems. Bug: 40448046 Change-Id: I17b9dd744671c6721faaf9060622cff788371f1f Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/6237116 Reviewed-by: Andy Phan <andyphan@chromium.org> Reviewed-by: Alan Screen <awscreen@chromium.org> Commit-Queue: Lei Zhang <thestig@chromium.org> Cr-Commit-Position: refs/heads/main@{#1417508}
332 lines
11 KiB
C++
332 lines
11 KiB
C++
// Copyright 2010 The Chromium Authors
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file.
|
|
|
|
#include "pdf/pdfium/pdfium_range.h"
|
|
|
|
#include <string>
|
|
|
|
#include "base/check_op.h"
|
|
#include "base/containers/span.h"
|
|
#include "base/feature_list.h"
|
|
#include "base/strings/string_util.h"
|
|
#include "pdf/accessibility_structs.h"
|
|
#include "pdf/pdfium/pdfium_api_string_buffer_adapter.h"
|
|
#include "third_party/pdfium/public/fpdf_searchex.h"
|
|
#include "ui/gfx/geometry/point.h"
|
|
#include "ui/gfx/geometry/rect.h"
|
|
#include "ui/gfx/geometry/rect_f.h"
|
|
|
|
namespace chrome_pdf {
|
|
|
|
namespace {
|
|
|
|
// Enables AccessibilityTextRunInfo-based screen rects.
|
|
// TODO(crbug.com/40448046): Remove this kill switch after a safe rollout.
|
|
BASE_FEATURE(kPdfAccessibilityTextRunInfoScreenRects,
|
|
"PdfAccessibilityTextRunInfoScreenRects",
|
|
base::FEATURE_ENABLED_BY_DEFAULT);
|
|
|
|
void AdjustForBackwardsRange(int& index, int& count) {
|
|
if (count < 0) {
|
|
count *= -1;
|
|
index -= count - 1;
|
|
}
|
|
}
|
|
|
|
// Struct with only the text run info needed for PDFiumRange::GetScreenRects().
|
|
struct ScreenRectTextRunInfo {
|
|
gfx::Rect screen_rect;
|
|
size_t char_count;
|
|
};
|
|
|
|
// Returns a ratio between [0, 1].
|
|
float GetVerticalOverlap(const gfx::Rect& rect1, const gfx::Rect& rect2) {
|
|
CHECK(!rect1.IsEmpty());
|
|
CHECK(!rect2.IsEmpty());
|
|
|
|
gfx::Rect union_rect = rect1;
|
|
union_rect.Union(rect2);
|
|
|
|
if (union_rect.height() == rect1.height() ||
|
|
union_rect.height() == rect2.height()) {
|
|
return 1.0f;
|
|
}
|
|
|
|
gfx::Rect intersect_rect = rect1;
|
|
intersect_rect.Intersect(rect2);
|
|
return static_cast<float>(intersect_rect.height()) / union_rect.height();
|
|
}
|
|
|
|
// Returns true if there is sufficient horizontal and vertical overlap.
|
|
bool ShouldMergeHorizontalRects(const ScreenRectTextRunInfo& text_run1,
|
|
const ScreenRectTextRunInfo& text_run2) {
|
|
static constexpr float kVerticalOverlapThreshold = 0.8f;
|
|
const gfx::Rect& rect1 = text_run1.screen_rect;
|
|
const gfx::Rect& rect2 = text_run2.screen_rect;
|
|
if (GetVerticalOverlap(rect1, rect2) < kVerticalOverlapThreshold) {
|
|
return false;
|
|
}
|
|
|
|
static constexpr float kHorizontalWidthFactor = 1.0f;
|
|
const float average_width1 =
|
|
kHorizontalWidthFactor * rect1.width() / text_run1.char_count;
|
|
const float average_width2 =
|
|
kHorizontalWidthFactor * rect2.width() / text_run2.char_count;
|
|
const float rect1_left = rect1.x() - average_width1;
|
|
const float rect1_right = rect1.right() + average_width1;
|
|
const float rect2_left = rect2.x() - average_width2;
|
|
const float rect2_right = rect2.right() + average_width2;
|
|
return rect1_left < rect2_right && rect1_right > rect2_left;
|
|
}
|
|
|
|
// Since PDFiumPage::GetTextRunInfo() can end a text run for a variety of
|
|
// reasons, post-process the collected text run data and merge rectangles.
|
|
std::vector<gfx::Rect> MergeAdjacentRects(
|
|
base::span<ScreenRectTextRunInfo> text_runs) {
|
|
std::vector<gfx::Rect> results;
|
|
const ScreenRectTextRunInfo* previous_text_run = nullptr;
|
|
gfx::Rect current_screen_rect;
|
|
for (const auto& text_run : text_runs) {
|
|
if (previous_text_run) {
|
|
// TODO(crbug.com/40448046): Improve vertical text handling.
|
|
// For now, treat all text as horizontal, as that is the majority of the
|
|
// text. Also, PDFiumPage::GetTextPage() has bugs in its heuristics where
|
|
// it mistakenly reports horizontal text as vertical.
|
|
if (ShouldMergeHorizontalRects(*previous_text_run, text_run)) {
|
|
current_screen_rect.Union(text_run.screen_rect);
|
|
} else {
|
|
results.push_back(current_screen_rect);
|
|
current_screen_rect = text_run.screen_rect;
|
|
}
|
|
} else {
|
|
current_screen_rect = text_run.screen_rect;
|
|
}
|
|
previous_text_run = &text_run;
|
|
}
|
|
|
|
if (!current_screen_rect.IsEmpty()) {
|
|
results.push_back(current_screen_rect);
|
|
}
|
|
return results;
|
|
}
|
|
|
|
} // namespace
|
|
|
|
bool IsIgnorableCharacter(char16_t c) {
|
|
return c == kZeroWidthSpace || c == kPDFSoftHyphenMarker;
|
|
}
|
|
|
|
// static
|
|
PDFiumRange PDFiumRange::AllTextOnPage(PDFiumPage* page) {
|
|
return PDFiumRange(page, 0, page->GetCharCount());
|
|
}
|
|
|
|
PDFiumRange::PDFiumRange(PDFiumPage* page, int char_index, int char_count)
|
|
: page_unload_preventer_(page),
|
|
page_(page),
|
|
char_index_(char_index),
|
|
char_count_(char_count) {
|
|
DCHECK(page_);
|
|
// Ensure page load, while `page_unload_preventer_` prevents page unload.
|
|
// This prevents GetScreenRects() from triggering page loads, which can have
|
|
// surprising side effects, considering GetScreenRects() is const.
|
|
[[maybe_unused]] FPDF_TEXTPAGE text_page = page_->GetTextPage();
|
|
#if DCHECK_IS_ON()
|
|
AdjustForBackwardsRange(char_index, char_count);
|
|
DCHECK_LE(char_count, FPDFText_CountChars(text_page));
|
|
#endif
|
|
}
|
|
|
|
PDFiumRange::PDFiumRange(const PDFiumRange&) = default;
|
|
|
|
PDFiumRange& PDFiumRange::operator=(const PDFiumRange&) = default;
|
|
|
|
PDFiumRange::PDFiumRange(PDFiumRange&&) noexcept = default;
|
|
|
|
PDFiumRange& PDFiumRange::operator=(PDFiumRange&&) noexcept = default;
|
|
|
|
PDFiumRange::~PDFiumRange() = default;
|
|
|
|
void PDFiumRange::SetCharCount(int char_count) {
|
|
if (char_count == char_count_) {
|
|
return;
|
|
}
|
|
|
|
char_count_ = char_count;
|
|
#if DCHECK_IS_ON()
|
|
int dummy_index = 0;
|
|
AdjustForBackwardsRange(dummy_index, char_count);
|
|
DCHECK_LE(char_count, FPDFText_CountChars(page_->GetTextPage()));
|
|
#endif
|
|
|
|
cached_screen_rects_point_ = gfx::Point();
|
|
cached_screen_rects_zoom_ = 0;
|
|
}
|
|
|
|
const std::vector<gfx::Rect>& PDFiumRange::GetScreenRects(
|
|
const gfx::Point& point,
|
|
double zoom,
|
|
PageOrientation orientation) const {
|
|
if (point == cached_screen_rects_point_ &&
|
|
zoom == cached_screen_rects_zoom_) {
|
|
return cached_screen_rects_;
|
|
}
|
|
|
|
cached_screen_rects_.clear();
|
|
cached_screen_rects_point_ = point;
|
|
cached_screen_rects_zoom_ = zoom;
|
|
|
|
if (char_count_ == 0) {
|
|
return cached_screen_rects_;
|
|
}
|
|
|
|
FPDF_TEXTPAGE text_page = page_->GetTextPage();
|
|
if (!text_page) {
|
|
return cached_screen_rects_;
|
|
}
|
|
|
|
int char_index = char_index_;
|
|
int char_count = char_count_;
|
|
|
|
AdjustForBackwardsRange(char_index, char_count);
|
|
DCHECK_GE(char_index, 0) << " start: " << char_index_
|
|
<< " count: " << char_count_;
|
|
DCHECK_LT(char_index, FPDFText_CountChars(text_page))
|
|
<< " start: " << char_index_ << " count: " << char_count_;
|
|
|
|
if (!base::FeatureList::IsEnabled(kPdfAccessibilityTextRunInfoScreenRects)) {
|
|
int count = FPDFText_CountRects(text_page, char_index, char_count);
|
|
for (int i = 0; i < count; ++i) {
|
|
double left;
|
|
double top;
|
|
double right;
|
|
double bottom;
|
|
FPDFText_GetRect(text_page, i, &left, &top, &right, &bottom);
|
|
gfx::Rect rect = page_->PageToScreen(point, zoom, left, top, right,
|
|
bottom, orientation);
|
|
if (rect.IsEmpty()) {
|
|
continue;
|
|
}
|
|
cached_screen_rects_.push_back(rect);
|
|
}
|
|
|
|
return cached_screen_rects_;
|
|
}
|
|
|
|
std::vector<ScreenRectTextRunInfo> text_runs;
|
|
const int end_char_index = char_index + char_count;
|
|
bool reached_end = false;
|
|
while (!reached_end) {
|
|
// Should not fail because `text_page` is non-null and `char_index` is
|
|
// always in range.
|
|
std::optional<AccessibilityTextRunInfo> text_run_info =
|
|
page_->GetTextRunInfo(char_index);
|
|
CHECK(text_run_info.has_value());
|
|
|
|
// Figure out how many characters to process in the for-loop below, and
|
|
// determine if this while-loop iteration reached the end of the range.
|
|
int next_char_index = char_index + text_run_info.value().len;
|
|
reached_end = next_char_index >= end_char_index;
|
|
if (reached_end) {
|
|
next_char_index = end_char_index;
|
|
}
|
|
|
|
// Do not use the bounds from `text_run_info`, as those are in the wrong
|
|
// coordinate system. Calculate it here instead.
|
|
gfx::Rect text_run_rect;
|
|
for (int i = char_index; i < next_char_index; ++i) {
|
|
// Use the loose rectangle, which gives the selection a bit more padding.
|
|
// In comparison, the rectangle from FPDFText_GetCharBox() surrounds the
|
|
// glyph too tightly.
|
|
//
|
|
// Should not fail because `text_page` is non-null, `i` is always in
|
|
// range, and the out-parameter is non-null.
|
|
FS_RECTF rect;
|
|
bool got_rect = FPDFText_GetLooseCharBox(text_page, i, &rect);
|
|
CHECK(got_rect);
|
|
|
|
// Check for empty `rect` and skip. PDFiumPage::PageToScreen() may round
|
|
// an empty `rect` to a 1x1 `screen_rect`, which is hard to distinguish
|
|
// from an actual 1x1 `rect`.
|
|
if (rect.left == rect.right || rect.top == rect.bottom) {
|
|
continue;
|
|
}
|
|
|
|
gfx::Rect screen_rect =
|
|
page_->PageToScreen(point, zoom, rect.left, rect.top, rect.right,
|
|
rect.bottom, orientation);
|
|
text_run_rect.Union(screen_rect);
|
|
}
|
|
if (!text_run_rect.IsEmpty()) {
|
|
text_runs.emplace_back(text_run_rect, next_char_index - char_index);
|
|
}
|
|
|
|
char_index = next_char_index;
|
|
}
|
|
|
|
cached_screen_rects_ = MergeAdjacentRects(text_runs);
|
|
return cached_screen_rects_;
|
|
}
|
|
|
|
std::u16string PDFiumRange::GetText() const {
|
|
int index = char_index_;
|
|
int count = char_count_;
|
|
std::u16string result;
|
|
if (count == 0)
|
|
return result;
|
|
|
|
AdjustForBackwardsRange(index, count);
|
|
if (count > 0) {
|
|
// Note that the `expected_size` value includes the NUL terminator.
|
|
//
|
|
// Cannot set `check_expected_size` to true here because the fix to
|
|
// https://crbug.com/pdfium/1139 made it such that FPDFText_GetText() is
|
|
// not always consistent with FPDFText_CountChars() and may trim characters.
|
|
//
|
|
// Instead, treat `count` as the requested count, but use the size of
|
|
// `result` as the source of truth for how many characters
|
|
// FPDFText_GetText() actually wrote out.
|
|
PDFiumAPIStringBufferAdapter<std::u16string> api_string_adapter(
|
|
&result, /*expected_size=*/count + 1, /*check_expected_size=*/false);
|
|
unsigned short* data =
|
|
reinterpret_cast<unsigned short*>(api_string_adapter.GetData());
|
|
int written = FPDFText_GetText(page_->GetTextPage(), index, count, data);
|
|
// FPDFText_GetText() returns 0 on failure. Never negative value.
|
|
DCHECK_GE(written, 0);
|
|
api_string_adapter.Close(written);
|
|
|
|
const gfx::RectF page_bounds = page_->GetCroppedRect();
|
|
std::u16string in_bound_text;
|
|
in_bound_text.reserve(result.size());
|
|
|
|
// If FPDFText_GetText() trimmed off characters, figure out how many were
|
|
// trimmed from the front. Store the result in `index_offset`, so the
|
|
// IsCharInPageBounds() calls below can have the correct index.
|
|
CHECK_GE(static_cast<size_t>(count), result.size());
|
|
size_t trimmed_count = static_cast<size_t>(count) - result.size();
|
|
int index_offset = 0;
|
|
while (trimmed_count) {
|
|
if (FPDFText_GetTextIndexFromCharIndex(page_->GetTextPage(),
|
|
index + index_offset) >= 0) {
|
|
break;
|
|
}
|
|
--trimmed_count;
|
|
++index_offset;
|
|
}
|
|
|
|
for (size_t i = 0; i < result.size(); ++i) {
|
|
// Filter out characters outside the page bounds, which are semantically
|
|
// not part of the page.
|
|
if (page_->IsCharInPageBounds(index + index_offset + i, page_bounds))
|
|
in_bound_text += result[i];
|
|
}
|
|
result = in_bound_text;
|
|
std::erase_if(result, IsIgnorableCharacter);
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
} // namespace chrome_pdf
|