rpdfium 0.4.2 → 0.4.4

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 28e6412dd11dd39727aebd3d62c084d3fdbd7a7a3b15ebd7ceb61b978de08770
-  data.tar.gz: 6a7dbd79f442d703020f286003a183e946f3b35eb8d206e339804e493d1d7c04
+  metadata.gz: b20ddc3e261b5d9a8a8fa0d84c8ff65e26a6a952a7e3d57dcfe64680944c652c
+  data.tar.gz: cf5c332e8231777062ecd593458b2192794dec3429a2875cf679279ee4ac6a5f
 SHA512:
-  metadata.gz: 6781340e17cdc757fde6334e6266bb1cf5db109b5935ea76eadc7fa6af1b750c91b2bdb0e579af09eb6504bac87fc1d8fd906b2e02dea0229147294380f44c97
-  data.tar.gz: e6aee42be9d8f6b80ce248cf90b7b44e1fa39feae71d0d8d56d32f32c19298e55d1fc5f6acf67189c9d55fc6b9825a293cecd170aa7e410236ff80ab88c1ff92
+  metadata.gz: d59b9039332abdf90ee46032ac7b6501b99b753e9045da3526c410e2806a53b2b5dd416ec83c254d2692a5926f64ff0aea8c1c5859633a26a26e84a43fd2cdb5
+  data.tar.gz: 7e218d1c1f6632ae871fa6b0453de3e97a256a9e9871788f4bed34a93c3ee71d91bee3042712ab083be963fec38fa5aa9bb20a608970bf45e7358115ddf554ed

data/CHANGELOG.md

@@ -8,6 +8,34 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.4.4] - 2026-06-18
+
+### Changed
+
+- **`Page#words` allocates ~77% fewer objects.** The row/word pipeline was
+  rebuilt to avoid transient allocations: the by-row sort now uses a
+  comparator block instead of `sort_by { [top, x0] }` (no per-char 2-element
+  key array nor Schwartzian pair array), the per-row sort runs in place with
+  `sort!`, and `word_from_chars` folds `text`/`top`/`bottom` in a single pass
+  instead of three separate `map`+`join`/`min`/`max` passes (each of which
+  allocated an intermediate array). Measured on a real 3567-char page (438
+  words): 6552 → 1534 allocated objects per call (1.84 → 0.43 objects/char).
+  Behavior is unchanged.
+
+## [0.4.3] - 2026-06-16
+
+### Fixed
+
+- **`Page#words` now returns numeric `top`/`bottom` coordinates.** Each word's
+  `:top` and `:bottom` fields were computed with `chars.min { |c| c[:top] }` /
+  `chars.max { |c| c[:bottom] }`. The block form of `Enumerable#min`/`max` must
+  return a comparator (-1/0/1) and receives two arguments, so the numeric value
+  returned by the single-argument block was used as a broken comparator and the
+  method returned the whole char hash instead of the position. As a result
+  `word[:top]` and `word[:bottom]` were no longer positional numbers. Fixed by
+  using `chars.map { |c| c[:top] }.min` / `.max`, which return the expected
+  scalar.
+
 ## [0.4.2] - 2026-06-15
 
 ### Added

data/lib/rpdfium/page.rb

@@ -876,13 +876,20 @@ module Rpdfium
       cs = chars(**char_opts)
       return [] if cs.empty?
 
-      # Group into rows by y
-      rows = group_consecutive(cs.sort_by { |c| [c[:top], c[:x0]] }) do |a, b|
+      # Group into rows by y. Comparator sort instead of sort_by avoids one
+      # 2-element key array per char (plus the Schwartzian pair array) — the
+      # dominant transient allocation on pages with thousands of chars.
+      sorted_cs = cs.sort do |a, b|
+        c = a[:top] <=> b[:top]
+        c.zero? ? a[:x0] <=> b[:x0] : c
+      end
+      rows = group_consecutive(sorted_cs) do |a, b|
         (a[:top] - b[:top]).abs <= y_tolerance
       end
 
       rows.flat_map do |row|
-        sorted = row.sort_by { |c| c[:x0] }
+        # In-place: `row` is a fresh group array, no need for a new copy.
+        sorted = row.sort! { |a, b| a[:x0] <=> b[:x0] }
         # Split on gap > x_tolerance or explicit space
         word_groups = []
         buf = []
@@ -1595,14 +1602,27 @@ module Rpdfium
     end
 
     def word_from_chars(chars)
+      # Single pass: fold text/top/bottom together instead of 3 separate
+      # map+join/min/max passes (each allocating an intermediate array).
+      first = chars.first
+      text = +''
+      top = first[:top]
+      bottom = first[:bottom]
+      chars.each do |c|
+        text << c[:char].to_s
+        t = c[:top]
+        b = c[:bottom]
+        top = t if t < top
+        bottom = b if b > bottom
+      end
       {
-        text: chars.map { |c| c[:char] }.join,
-        x0: chars.first[:x0],
+        text: text,
+        x0: first[:x0],
         x1: chars.last[:x1],
-        top: chars.min { |c| c[:top] },
-        bottom: chars.max { |c| c[:bottom] },
-        fontsize: chars.first[:fontsize],
-        font: chars.first[:font],
+        top: top,
+        bottom: bottom,
+        fontsize: first[:fontsize],
+        font: first[:font],
         chars: chars
       }
     end

data/lib/rpdfium/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Rpdfium
-  VERSION = "0.4.2"
+  VERSION = "0.4.4"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: rpdfium
 version: !ruby/object:Gem::Version
-  version: 0.4.2
+  version: 0.4.4
 platform: ruby
 authors:
 - Roberto Scinocca