更新日志#

Changes in version 1.25.4 ()

  • 修复的问题:

    • 修复 4224: MuPDF 错误：格式错误：1d faxd 中的负码

    • 修复 4303: page.get_image_info() 在替换图像后返回过时的缓存结果

    • 修复 4309: 删除第一页时发生 FzErrorFormat 错误

    • 修复 4336: 性能回退：pix.color_count 在版本 1.25.3 中比 1.23.8 慢 150 倍

    • 修复 4341: 当 /Kids 是多个 /Nums 数组时获取无效标签

  • 其他:

    • 修复了合并 PDF 时重复小部件名称的处理（PR #4347）。

    • 改进了 Pyodide 构建。

    • 通过禁用 PY_LIMITED_API 来避免 Python-3.13 中与 SWIG 相关的构建错误。

Changes in version 1.25.3 (2025-02-06)

  • 使用 MuPDF-1.25.4.

  • 修复的问题:

    • 修复 4139: 文本颜色数字在 1.24.14 和 1.25.0 之间发生变化

    • 修复 4141: 一些插入方法在没有 /Resources 对象的页面上失败

    • 修复 4180: 搜索问题

    • 修复 4182: 文本坐标提取错误

    • 修复 4245: 高亮显示问题在最近版本中失真

    • 修复 4254: add_freetext_annot 在注释框外绘制文本

  • 其他:

    • 在注释方面: * 添加对 FreeTextCallout 子类型的支持。 * 添加对富文本的支持。

    • 为 insert_text*() 添加 miter_limit 参数，以便抑制长斜接线造成的尖角。

    • 在 Document.insert_pdf() 中添加小部件支持。

    • 向 span 字典中添加 bibi。

    • 向字符字典中添加 synthetic。

    • 修复了 Pyodide 构建。

Changes in version 1.25.2 (2025-01-17)

  • 修复的问题:

    • 修复 4055: 所有复选框的“是”对于所有 PDF 渲染引擎不起作用

    • 修复 4155: samples_mv 不安全

    • 修复 4162: 尝试添加签名字段时出现 AttributeError

    • 修复 4186: 错误处理带有 CMYK 色彩空间的 JPEG 图像提取

    • 修复 4195: 倒置且具有 alpha 通道的 pixmap 无法正确渲染

    • 修复 4225: pixmap.pil_save() 由于色彩空间定义失败

    • 修复 4232: 错误的字体样式和大小

  • 其他:

    • 使用 Python 内置的 glyphname <> unicode 转换。

    • 提高 pixmap 色彩反转的速度。

    • 向 span 字典添加新的 char_flags 成员，允许检测不可见文本。

    • 在 TextPage 输出中检测图像遮罩。

    • 添加 Pixmap.pil_image()。

Changes in version 1.25.1 (2024-12-11)

  • 使用 MuPDF-1.25.2。

  • 修复的问题:

    • 修复 4125: 转换 Pixmap 的色彩空间时内存泄漏

    • 修复 4034: 保存时可能发生 PDF 清理回归问题

Changes in version 1.25.0 (2024-12-05)

  • 使用 MuPDF-1.25.1。

  • 修复的问题:

    • 修复 4026: page.get_text(‘blocks’) 输出两个非常相似的文本，具有不同的边界框

    • 修复 4004: 更新 PDF 表单字段值时发生段错误

    • 修复 3887: 使用回退字体时的子集字体问题

    • 修复 3886: 插入 HTML 时销毁 PDF 的另一个问题

    • 修复 3751: apply_redactions 导致页面内容部分隐藏 / 透明

Changes in version 1.24.14 (2024-11-19)

  • 使用 MuPDF-1.24.11。

  • 修复的问题:

    • 修复 3448: get_pixmap 函数移除了表格，仅留下内容

    • 修复 3758: 使用 add_redact_annot/apply_redactions 时出现 “malloc(): unaligned tcache chunk detected Aborted (core dumped)” 错误

    • 修复 3813: Stories: 嵌套无序列表时有序列表计数出错

    • 修复 3933: font.valid_codepoints() - 故障

    • 修复 4018: PyMuPDF 在反向遍历零页 PDF 页面时挂起

    • 修复 4043: fullcopypage 错误

    • 修复 4047: add_redact_annot 时的段错误

    • 修复 4050: doc.embfile_info() 返回的字典内容与文档不符

  • 其他:

    • 确保 Page.get_text() 中的单词不包含 RTL/LTR 字符混合。

    • 修复使用系统 MuPDF 时的构建问题。

    • 为点和向量添加点积功能。

Changes in version 1.24.13 (2024-10-29)

  • 修复的问题:

    • 修复 3848: Pixmap 程序崩溃

    • 修复 3950: 无法一致地从 PDF 中提取字段标签

    • 修复 3981: PyMuPDF 1.24.12 与 pyinstaller 一起使用时抛出错误

    • 修复 3994: pix.color_topusage 引发段错误 (core dumped)

Changes in version 1.24.12 (2024-10-21)

  • 修复的问题:

    • 修复 3914: 能够将 MuPDF 错误打印到日志中，而不是标准输出

    • 修复 3916: insert_htmlbox 错误：整数过大无法转换为浮动

    • 修复 3950: 无法一致地从 PDF 中提取字段标签

  • 支持的 Python 版本现在是 3.9 到 3.13。

Changes in version 1.24.11 (2024-10-03)

  • 使用 MuPDF-1.24.10。

  • 修复的问题:

    • 修复 3624: 将 PDF 文件转换为图像时出现黑色块

    • 修复 3859: doc.need_appearances() 报告 “AttributeError: module ‘pymupdf.mupdf’ has no attribute ‘PDF_TRUE’”

    • 修复 3863: apply_redactions() 无法按预期工作

    • 修复 3905: 打开流时，可能会抛出 FzErrorFormat 错误，而不是 FileDataError

  • 轮子现在使用 Python 稳定 ABI：

    • 每个平台都有一个 PyMuPDF 轮子。

    • 每个轮子适用于所有受支持的 Python 版本。

    • 每个轮子使用最旧的支持 Python 版本（目前为 3.8）构建。

    • 不再提供 PyMuPDFb 轮子。

  • 其他:

    • 对 get_text_words() 进行了改进，sort=True。

    • 测试现在始终获取最新版本的所需 Python 包。

    • 移除了对 setuptools 的依赖。

    • 在 PyMuPDF-1.24.10 的变更中添加了修复 #3630。

Changes in version 1.24.10 (2024-09-02)

  • 使用 MuPDF-1.24.9。

  • 修复的问题:

    • 修复 3450: get_pixmap 函数处理时间过长

    • 修复 3569: SVG 图像创建时未忽略无效的 OCGs

    • 修复 3603: ObjStm 压缩和 PDF 线性化不能一起工作

    • 修复 3650: 每个字母之间插入了换行符

    • 修复 3661: 更新 Document 以检查 /XYZ 长度

    • 修复 3698: 文档问题 - 注释文档中的旧代码

    • 修复 3705: 在某些特定类型的 PDF 文件中，Document.select() 表现异常

    • 修复 3706: 扩展 Document.__getitem__ 类型注解，以反映该方法也接受切片

    • 修复 3727: get_pixmap() 方法导致程序在没有任何异常或消息的情况下退出

    • 修复 3767: 无法获取 Tesseract-OCR 5 的 Tessdata

    • 修复 3773: Link.set_border 报告 TypeError: '< not supported between instances of 'NoneType' and 'int'

    • 修复 3774: fitz.__version__ 不再工作

    • 修复 3789: 调用 insert_pdf 时抛出 ValueError: not enough values to unpack (expected 3, got 2)

    • 修复 3820: 改进了 namedDest 处理

    • 修复 3630: page.apply_redactions 给出了不需要的黑色矩形

  • 其他:

    • 无法同时使用对象流和线性化；如果尝试这么做，将引发异常。 (#3603)

    • 修复了对不存在的 /Contents 对象的处理。

Changes in version 1.24.9 (2024-07-24)

  • 使用 MuPDF-1.24.8.

Changes in version 1.24.8 (2024-07-22)

  • 修复的问题:

    • 修复 3636: open 函数的 API 文档不容易找到。

    • 修复 3654: docx 解析在 1.24.7 中已损坏。

    • 修复 3677: 无法在 PyMuPDF 1.24.6 和 1.24.7 中提取子集字体名称。

    • 修复 3687: Page.get_text 在处理 EPUB 文件时导致 AssertionError。

  • 其他:

    • 修复了代码中由 codespell 检测到的各种拼写错误。

    • 改进了我们在 Windows 上修改 MuPDF 默认配置的方式。

    • 使文本搜索支持联字。

Changes in version 1.24.7 (2024-06-26)

Changes in version 1.24.6 (2024-06-25)

基于 MuPDF-1.24.4 的版本

  • 修复 3599：Story.fit_width() 产生异常换行。

  • 修复 3594：Amazon 可持续发展报告的文本提取乱码问题。

  • 修复 3591：Page.get_drawings() 返回的 width 始终为 0。

  • 修复 3561：page.apply_redactions() 可能导致 ZeroDivisionError: float division by zero。

  • 修复 3559：insert_htmlbox 方法在 H1、H2、H3、H4 等标签为空时引发 SegFault 11。

  • 修复 3539：表格识别中新增对虚线网格的检测支持。

  • 修复 3519：get_toc(simple=False) 抛出 AttributeError: 'Outline' object has no attribute 'rect'。

  • 修复 3510：page.get_label() 可能在文档第一页返回错误标签。

  • 修复 3494：在 subset_fonts 与 insert_pdf 结合使用时，1.24.2/1.24.3 版本可能会引入错误字符。

  • 修复 3470：subset_fonts 可能在发生错误时无异常或警告直接退出。

  • 修复 3400：在 PyMuPDF 1.24.2 版本中，set_toc 可能导致某些旋转页面的链接坐标被修改。

  • 修复 3347：当 PDF 页面的高度不同时，某些指向页面内部的链接可能不正确。

  • 修复 3237：set_metadata() 方法无法正常工作。

  • 修复 3493：隔离 PyMuPDF 与其他库的相互影响，解决 PyMuPDF 与 GdkPixbuf 等其他库同时加载时的兼容性问题。

其他改动

  • 解决了 PyMuPDF 在并发使用时由于临时文件名固定而导致

Changes in version 1.24.5 (2024-05-30)

  • Fixed issues:

    • Fixed 3479: regression: fill_textbox: IndexError: pop from empty list

    • Fixed 3488: set_toc method error

  • Other:

    • Some more fixes to use MuPDF floating formatting.

    • Removed/disabled some unnecessary diagnostics.

    • Fixed utils.do_links() crash.

    • Experimental new functions pymupdf.apply_pages() and pymupdf.get_text().

    • Addresses wrong label generation for label styles “a” and “A”.

Changes in version 1.24.4 (2024-05-16)

  • Fixed 3418: Re-introduced bug, text align add_redact_annot

  • Fixed 3472: insert_pdf gives SystemError

  • Other:

    • Fixed sysinstall test failing to remove all of prior installation before new install.

    • Fixed utils.do_links() crash.

    • Correct TextPage creation Code.

    • Unified various diagnostics.

    • Fix bug in page_merge().

Changes in version 1.24.3 (2024-05-09)

  • The Python module is now called pymupdf. fitz is still supported for backwards compatibility.

  • Use MuPDF-1.24.2.

  • Fixed issues:

    • Fixed 3357: PyMuPDF==1.24.0 will hanging when using page.get_text(“text”)

    • Fixed 3376: Redacting results are not as expected in 1.24.x.

    • Fixed 3379: Documentation mismatch for get_text_blocks return value order.

    • Fixed 3381: Contents stream contains floats in scientific notation

    • Fixed 3402: Cannot add Widgets containing inter-field-calculation JavaScript

    • Fixed 3414: missing attribute set_dpi()

    • Fixed 3430: page.get_text() cause process freeze with certain pdf on v1.24.2

  • Other:

    • New/modified methods:

    • Fixed some problems when checking for PDF properties.

    • Fixed pip builds from sdist (see discussion 3360: Alpine linux docker build failing “No matching distribution found for pymupdfb==1.24.1”).

Changes in version 1.24.2 (2024-04-17)

  • Removed obsolete classic implementation from releases (previously available as module fitz_old).

  • Fixed issues:

    • Fixed 3331: Document.pages() is incorrectly type-hinted

    • Fixed 3354: PyMuPDF==1.24.1: AttributeError: property ‘metadata’ of ‘Document’ object has no setter

  • Other:

Changes in version 1.24.1 (2024-04-02)

  • Fixed issues:

    • Fixed 3278: apply_redactions moves some unredacted text

    • Fixed 3301: Be more permissive when classifying links as kind LINK_URI

    • Fixed 3306: Text containing capital ‘ET’ not appearing as annotation

  • Other:

    • Use MuPDF-1.24.1.

    • Support ObjStm Compression. Methods Document.save(), Document.ez_save() and Document.write() now support new parameters use_objstm, compression_effort` and preserve_metadata.

Changes in version 1.24.0 (2024-03-21)

  • Fixed issues:

    • Fixed 3281: Preparing metadata (pyproject.toml) did not run successfully

    • Fixed 3279: PyMuPDF no longer builds in Alpine Linux

    • Fixed 3257: apply_redactions() deleting text outside of annoted box

    • Fixed 3216: AttributeError: ‘Annot’ object has no attribute ‘__del__’

    • Fixed 3207: get_drawings’s items is missing line from h path operator

    • Fixed 3201: Memory leaks when merging PDFs

    • Fixed 3197: page.get_text() returns hexadecimal text for some characters

    • Fixed 3196: Remove text not working in 1.23.25 version vs 1.20.2

    • Fixed 3172: PDF’s 45º lines dissapearing in png conversion

    • Fixed 3135: Do not log warnings to stdout

    • Fixed 3125: get_pixmap method stuck on one page and runs forever

    • Fixed 2964: There is an issue with the image generated by the page.get_pixmap() function

  • Other:

    • Use MuPDF-1.24.0.

    • Add support for redacting vector graphics.

    • Several fixes for table module

      • Add new method for outputting the table as a markdown string.

      • Address errors in computing the table header object:

        We now allow None as the cell value, because this will be resolved where needed (e.g. in the pandas DataFrame).

        We previously tried to enforce rect-like tuples in all header cell bboxes, however this fails for tables with all-None columns. This fix enables this and constructs an empty string in the corresponding cell string.

        We now correctly include start / stop points of lines in the bbox of the clustered graphic. We previously joined the line’s rectangle - which had no effect because this is always empty.

    • Improved exception text if we fail to open document.

    • Fixed build with new libclang 18.

Changes in version 1.23.26 (2024-02-29)

  • Fixed issues:

    • Fixed 3199: Add entry_points to setuptools configuration to provide command-line console scripts

    • Fixed 3209: Empty vertices in ink annotation

  • Other:

    • Improvements to table detection:

      • Improved check for empty tables, fixes bugs when determining table headers.

      • Improved computation of enveloping vector graphic rectangles.

      • Ignore more meaningless “pseudo” tables

    • Install command-line ‘pymupdf’ command that runs fitz/__main__.py.

    • Don’t overwrite MuPDF’s config.h when building on non-Windows.

    • Fix Story constructor’s Archive arg to match docs - now accepts a single Archive constructor arg.

    • Do not include MuPDF source in sdist; will be downloaded automatically when building.

Changes in version 1.23.25 (2024-02-20)

  • Fixed issues:

    • Fixed 3182: Pixmap.invert_irect argument type error

    • Fixed 3186: extractText() extracts broken text from pdf

    • Fixed 3191: Error on .find_tables()

  • Other:

    • When building, be able to specify python-config directly, with environment variable PIPCL_PYTHON_CONFIG.

Changes in version 1.23.24 (2024-02-19)

  • Fixed issues:

    • Fixed 3148: Table extraction - vertical text not handled correctly

    • Fixed 3179: Table Detection: Incorrect Separation of Vector Graphics Clusters

    • Fixed 3180: Cannot show optional content group: AttributeError: module ‘fitz.mupdf’ has no attribute ‘pdf_array_push_drop’

  • Other:

    • Be able to test system install using sudo pip install instead of a venv.

Changes in version 1.23.23 (2024-02-18)

  • Fixed issues:

    • Fixed 3126: Initialising Archive with a pathlib.Path fails.

    • Fixed 3131: Calling the next attribute of an Annot raises a “No attribute .parent” warning

    • Fixed 3134: Using an IRect as clip parameter in Page.get_pixmap no longer works since 1.23.9

    • Fixed 3140: PDF document stays in use after closing

    • Fixed 3150: doc.select() hangs on this doc.

    • Fixed 3163: AssertionError on using fitz.IRect

    • Fixed 3177: fitz.Pixmap(None, pix) Unrecognised args for constructing Pixmap

  • Other:

    • Improved Document.select() by using new MuPDF function `pdf_rearrange_pages(). This is a more complete (and faster) implementation of what needs to be done here in that not only pages will be rearranged, but also consequential changes will be made to the table of contents, links to removed pages and affected entries in the Optional Content definitions.

    • TextWriter.appendv(): added small_caps arg.

    • Fixed some valgrind errors with MuPDF master.

    • Fixed Document.insert_image() when build with MuPDF master.

Changes in version 1.23.22 (2024-02-12)

  • Fixed issues:

    • Fixed 3143: Difference in decoding of OCGs names between doc.get_ocgs() and page.get_drawings()

    • Fixed 3139: Pixmap resizing needs positional arg “clip” - even if None.

  • Other:

    • Removed the use of MuPDF function fz_image_size() from PyMuPDF.

Changes in version 1.23.21 (2024-02-01)

Changes in version 1.23.20 (2024-01-29)

  • Bug fixes:

    • Fixed 3100: Wrong internal property accessed in get_xml_metadata

  • Other:

Changes in version 1.23.19 (2024-01-25)

  • Bug fixes:

    • Fixed 3087: Exception in insert_image with mask specified

    • Fixed 3094: TypeError: ‘<’ not supported between instances of ‘FzLocation’ and ‘int’ in doc.delete_pages

  • Other:

    • When finding tables:

      • Allow addition of user-defined “virtual” vector graphics when finding tables.

      • Confirm that the enveloping bboxes of vector graphics are inside the clip rectangle.

      • Avoid slow finding of rectangle intersections.

    • Added Font.bbox property.

Changes in version 1.23.18 (2024-01-23)

  • Bug fixes:

    • Fixed 3081: doc.close() not closing the document

  • Other:

    • Reduced size of sdist to fit on pypi.org (by reducing size of two test files).

    • Fix Annot.file_info() if no Desc item.

Changes in version 1.23.17 (2024-01-22)

  • Bug fixes:

    • Fixed 3062: page_rotation_reset does not return page to original rotation

    • Fixed 3070: update_link(): AttributeError: ‘Page’ object has no attribute ‘super’

  • Other:

Changes in version 1.23.16 (2024-01-18)

  • Bug fixes:

    • Fixed 3058: Pixmap created from CMYK JPEG delivers RGB format

  • Other:

    • In table detection strategy “lines_strict”, exclude fill-only vector graphics.

    • Fixed sysinstall test failure.

    • In documentation, update feature matrix with item about text writing.

Changes in version 1.23.15 (2024-01-16)

  • Bug fixes:

    • Fixed 3050: python3.9 pix.set_pixel has something wrong in c.append( ord(i))

  • Other:

    • Improved docs for Page.find_tables().

Changes in version 1.23.14 (2024-01-15)

  • Bug fixes:

    • Fixed 3038: JM_pixmap_from_display_list > Assertion Error : Checking for wrong type

    • Fixed 3039: Issue with doc.close() not closing the document in PyMuPDF

  • Other:

Changes in version 1.23.13 (2024-01-15)

  • Bug fixes:

    • Fixed 2979: list index out of range in to_pandas()

    • Fixed 3001: Calling find_tables() on one document alters the bounding boxes of a subsequent document

  • Other:

    • Fixed Rect.height and Rect.width to never return negative values.

    • Fixed TextPage.extractIMGINFO()’s returned dictkey_yres value.

Changes in version 1.23.12 (2024-01-12)

    • Fixed 3027: Page.get_text throws Attribute Error for ‘parent’

Changes in version 1.23.11 (2024-01-12)

  • Fixed some Pixmap construction bugs.

  • Fixed Pixmap.yres().

Changes in version 1.23.10 (2024-01-12)

  • Bug fixes:

    • Fixed 3020: Can’t resize a PixMap

  • Other:

    • Fixed Page.delete_image().

Changes in version 1.23.9 (2024-01-11)

  • Default to new “rebased” implementation.

  • Use MuPDF-1.23.9.

  • Bug fixes (rebased implementation only):

    • Fixed 2911: Page.derotation_matrix returns a tuple instead of a Matrix with rebased implementation

    • Fixed 2919: Rebased version: KeyError in resolve_names when merging pdfs

    • Fixed 2922: New feature that allows inserting named-destination links doesn’t work

    • Fixed 2943: ZeroDivisionError: float division by zero when use apply_redactions()

    • Fixed 2950: Shelling out to pip during tests is problematic

    • Fixed 2954: Replacement unicode character in text extraction

    • Fixed 2957: apply_redactions() moving text

    • Fixed 2961: Passing a string as a page number raises IndexError instead of TypeError.

    • Fixed 2969: annot.next throws AttributeError

    • Fixed 2978: 1.23.9rc1: module ‘fitz.mupdf’ has no attribute ‘fz_copy_pixmap_rect’

    • Fixed 2907: segfault trying to call clean_contents on certain pdfs with python 3.12

    • Fixed 2905: SystemError: <built-in function TextPage_extractIMGINFO> returned a result with an exception set

    • Fixed 2742: Segmentation Fault when inserting three (but not two) copies of the same source page into one destination page

  • Other:

    • Add optional setting of opacity to Page.insert_htmlbox().

    • Fixed issue with add_redact_annot() mentioned in #2934.

    • Fixed Page.rotation() to return 0 for non-PDF documents instead of raising an exception.

    • Fixed internal quad detection to cope with any Python sequence.

    • Fixed rebased fitz.pymupdf_version_tuple - was previously set to mupdf version.

    • Improved support for Linux system installs, including adding regular testing on Github.

    • Add missing flake8 to scripts/gh_release.py:test_packages.

    • Use newly public functions in MuPDF-1.23.8.

    • Improved scripts/test.py to help investigation of MuPDF issues.

Changes in version 1.23.8 (2023-12-19)

  • Bug fixes (rebased implementation only):

    • Fixed 2634: get_toc and set_toc do not behave consistently for rotated pages

    • Fixed 2861: AttributeError in getLinkDict during PDF Merge

    • Fixed 2871: KeyError in getLinkDict during PDF merge

    • Fixed 2886: Error in Skeleton for Named Link Destinations

  • Bug fixes (rebased and classic implementations):

    • Fixed 2885: pymupdf find tables too slow

  • Other:

    • Rebased implementation:

      • Page.insert_htmlbox(): new, much more powerful alternative to Page.insert_textbox() or TextWriter.fill_textbox(), using Story.

      • Story.fit*(): new methods for fitting a Story into an expanded rect.

      • Story.write_with_links(): add support for external links.

      • Document.language(): fixed to use MuPDF’s new mupdf.fz_string_from_text_language2().

      • Document.subset_fonts() - fixed.

      • Fixed internal Archive._add_treeitem() method.

      • Fixed fitz_new.__doc__ to contain PyMuPDF and Python version information, and OS name.

      • Removed use of (*args, **kwargs) in API, we now specify keyword args explicitly.

      • Work with new MuPDF Python exception classes.

    • Fixed bug where button_states() returns None when /AP points to an indirect object.

    • Fixed pillow test to not ignore all errors, and install pillow when testing.

    • Added test for fitz.css_for_pymupdf_font() (uses package pymupdf-fonts).

    • Simplified Github Actions test specifications.

    • Updated tests/README.md.

Changes in version 1.23.7 (2023-11-30)

  • Bug fixes in rebased implementation, not fixed in classic implementation:

    • Fixed 2232: Geometry helper classes should support keyword arguments

    • Fixed 2788: Problem with get_toc in pymupdf 1.23.6

    • Fixed 2791: Experiencing small memory leak in save()

  • Bug fixes (rebased and classic implementations):

    • Fixed 2736: Failure when set cropbox with mediabox negative value

    • Fixed 2749: RuntimeError: cycle in structure tree

    • Fixed 2753: Story.write_with_links will ignore everything after the first “page break” in the HTML.

    • Fixed 2812: find_tables on landscape page generates reversed text

    • Fixed 2829: [cannot create /Annot for kind] is still printed despite #2345 is closed.

    • Fixed 2841: Unexpected KeyError when using scrub with fitz_new

  • Use MuPDF-1.23.7.

  • Other:

    • Rebased implementation:

      • Added flake8 code checking to test suite, and made various fixes.

      • Disable diagnostics during Document constructor to match classic implementation.

    • Additional fix to 2553: Invalid characters in versions >= 1.22

    • Fixed MuPDF Bug 707324: Story: HTML table row background color repeated incorrectly

    • Added scripts/test.py, for simple build+test of PyMuPDF git checkout.

    • Added fitz.pymupdf_version_tuple, e.g. (1, 23, 6).

    • Restored mistakenly-reverted fix for 2345: Turn off print statements in utils.py

    • Include any trailing ... repeated <N> times... text in warnings returned by mupdf_warnings() (rebased only).

Changes in version 1.23.6 (2023-11-06)

  • Bug fixes:

    • Fixed 2553: Invalid characters in versions >= 1.22

    • Fixed 2608: Incorrect utf32 text extraction (high & low surrogates are split)

    • Fixed 2710: page.rect and text location wrong / differing from older version

    • Fixed 2774: wrong encoding for “?” character when sort=True

    • Fixed 2775: fitz_new does not work with python3.10 or earlier

    • Fixed 2777: With fitz_new, wrong type for Page.mediabox

  • Other:

    • Use MuPDF-1.23.5.

    • Added Document.resolve_names() (rebased implementation only).

Changes in version 1.23.5 (2023-10-11)

  • Bug fixes:

    • Fixed 2341: Handling negative values in the zoom section for LINK_GOTO in linkDest

    • Fixed 2522: Typo in set_layer() - NameError: name ‘f’ is not defined

    • Fixed 2548: Fitz freezes on some PDFs when calling the fitz.Page.get_text_blocks method.

    • Fixed 2596: save(garbage=3) breaks get_pixmap() with side effect

    • Fixed 2635: “clean=True” makes objects invisible in the pdf

    • Fixed 2637: Page.insert_textbox incorrectly handles the last word if it starts a new line

    • Fixed 2699: extract paragraph with below table

    • Fixed 2703: Wrong fontsize calculation in corner cases (“page.get_texttrace()”)

    • Fixed 2710: page.rect and text location wrong / differing from older version

    • Fixed 2723: When will a Python 3.12 wheel be available?

    • Fixed 2730: persistent get_text() formatting

  • Other:

    • Use MuPDF-1.23.4.

    • Fix optimisation flags with system installs.

    • Fixed the problem that the clip parameter does not take effect during table recognition

    • Support Pillow mode “RGBa”

    • Support extra word delimiters

    • Support checking valid PDF name objects

Changes in version 1.23.4 (2023-09-26)

  • Improved build instructions.

  • Fixed Tesseract in rebased implementation.

  • Improvements to build/install with system MuPDF.

  • Fixed Pyodide builds.

  • Fixed rebased bug in _insert_image().

  • Bug fixes:

    • Fixed 2556: Segmentation fault at caling get_cdrawings(extended=True)

    • Fixed 2637: Page.insert_textbox incorrectly handles the last word if it starts a new line

    • Fixed 2683: Windows sdist build failure - non-quoting of path and using UNIX which command

    • Fixed 2691: Page.get_textpage_ocr() bug in rebased fitz_new version

    • Fixed 2692: Page.get_pixmap(clip=Rect()) bug in rebased fitz_new version

Changes in version 1.23.3 (2023-08-31)

  • Fixed use of Tesseract for OCR.

Changes in version 1.23.2 (2023-08-28)

  • Fixed #2613: release 1.23.0 not MacOS-arm64 compatible

Changes in version 1.23.1 (2023-08-24)

  • Updated README and package summary description.

  • Fixed a problem on some Linux installations with Python-3.10 (and possibly earlier versions) where import fitz failed with ImportError: libcrypt.so.2: cannot open shared object file: No such file or directory.

  • Fixed incompatible architecture error on MacOS arm64.

  • Fixed installation warning from Poetry about missing entry in wheels’ RECORD files.

Changes in version 1.23.0 (2023-08-22)

  • Add method find_tables() to the Page object.

    This allows locating tables on any supported document page, and extracting table content by cell.

  • New “rebased” implementation of PyMuPDF.

    The rebased implementation is available as Python module fitz_new. It can be used as a drop-in replacement with import fitz_new as fitz.

  • Python-independent MuPDF libraries are now in a second wheel called PyMuPDFb that will be automatically installed by pip.

    This is to save space on pypi.org - a full release only needs one PyMuPDFb wheel for each OS.

  • Bug fixes:

    • Fixed #2542: fitz.utils.scrub AttributeError Annot object has no attribute fileUpd inside

    • Fixed #2533: get_texttrace returned a incorrect character bbox

    • Fixed #2537: Validation when setting a grouped RadioButton throws a RuntimeError: path to ‘V’ has indirects

  • Other changes:

    • Dropped support for Python-3.7.

    • Fix for wrong page / annot /Contents cleaning.

      We need to set pdf_filter_options::no_update to zero.

    • Added new function get_tessdata().

    • Cope with problem /Annot arrays.

      When copying page annotations in method Document.insert_pdf we previously did not check the validity of members of the /Annots array. For faulty members (like null or non-dictionary items) this could cause unnecessary exceptions. This fix implements more checks and skips such array items.

    • Additional annotation type checks.

      We did not previously check for annotation type when getting / setting annotation border properties. This is now checked in accordance with MuPDF.

    • Increase fault tolerance.

      Avoid exceptions in method insert_pdf() when source pages contains invalid items in the /Annots array.

    • Return empty border dict for applicable annots.

      We previously were returning a non-empty border dictionary even for non-applicable annotation types. We now return the empty dictionary {} in these cases. This requires some corresponding changes in the annotation .update() method, namely for dashes and border width.

    • Restrict set_rect to applicable annot types.

      We were insufficiently excluding non-applicable annotation types from set_rect() method. We now let MuPDF catch unsupported annotations and return False in these cases.

    • Wrong fontsize computation in page.get_texttrace().

      When computing the font size we were using the final text transformation matrix, where we should have taken span->trm instead. This is corrected here.

    • Updates to cope with changes to latest MuPDF.

      pdf_lookup_anchor() has been removed.

    • Update fill_textbox to better respect rect.width

      The function norm_words in fill_textbox had a bug in its last loop, appending n+1 characters when actually measuring width of n characters. It led to a bug in fill_texbox when you tried to write a single word mostly composed of “wide” letters (M,m, W, w…), causing the written text to exceed the given rect.

      The fix was just to replace n+1 by n.

    • Add script_focus and script_blur options to widget.

Changes in version 1.22.5 (2023-06-21)

  • This release uses MuPDF-1.22.2.

  • Bug fixes:

    • Fixed #2365: Incorrect dictionary values for type “fs” drawings.

    • Fixed #2391: Check box automatically uncheck when we update same checkbox more than 1 times.

    • Fixed #2400: Gaps within text of same line not filled with spaces.

    • Fixed #2404: Blacklining an image in PDF won’t remove underlying content in version 1.22.X.

    • Fixed #2430: Incorrectly reducing ref count of Py_None.

    • Fixed #2450: Empty fill color and fill opacity for paths with fill and stroke operations with 1.22.*

    • Fixed #2462: Error at “get_drawing(extended=True )”

    • Fixed #2468: Decode error when trying to get drawings

    • Fixed #2710: page.rect and text location wrong / differing from older version

    • Fixed #2723: When will a Python 3.12 wheel be available?

  • New features:

    • Changed Annotations now support “cloudy” borders. The Annot.border property has the new item clouds, and method Annot.set_border() supports the corresponding clouds argument.

    • Changed Radio button widgets in the same RB group are now consistently updated if the group is defined in the standard way.

    • Added Support for the /Locked key in PDF Optional Content. This array inside the catalog entry /OCProperties can now be extracted and set.

    • Added Support for new parameter tessdata in OCR functions. New function get_tessdata() locates the language support folder if Tesseract is installed.

Changes in version 1.22.3 (2023-05-10)

  • This release uses MuPDF-1.22.0.

  • Bug fixes:

    • Fixed #2333: Unable to set any of button radio group in form

Changes in version 1.22.2 (2023-04-26)

  • This release uses MuPDF-1.22.0.

  • Bug fixes:

    • Fixed #2369: Image extraction bugs with newer versions

Changes in version 1.22.1 (2023-04-18)

  • This release uses MuPDF-1.22.0.

  • Bug fixes:

    • Fixed #2345: Turn off print statements in utils.py

    • Fixed #2348: extract_image returns an extension “flate” instead of “png”

    • Fixed #2350: Can not make widget (checkbox) to read-only by adding flags PDF_FIELD_IS_READ_ONLY

    • Fixed #2355: 1.22.0 error when using get_toc (AttributeError: ‘SwigPyObject’ object has no attribute)

Changes in version 1.22.0 (2023-04-14)

  • This release uses MuPDF-1.22.0.

  • Behavioural changes:

    • Text extraction now includes glyphs that overlap with clip rect; previously they were included only if they were entirely contained within the clip rect.

  • Bug fixes:

    • Fixed #1763: Interactive(smartform) form PDF calculation not working in pymupdf

    • Fixed #1995: RuntimeError: image is too high for a long paged pdf file when trying

    • Fixed #2093: Image in pdf changes color after applying redactions

    • Fixed #2108: Redaction removing more text than expected

    • Fixed #2141: Failed to read JPX header when trying to get blocks

    • Fixed #2144: Replace image throws an error

    • Fixed #2146: Wrong Handling of Reference Count of “None” Object

    • Fixed #2161: Support adding images as pages directly

    • Fixed #2168: page.add_highlight_annot(start=pointa, stop=pointb) not working

    • Fixed #2173: Double free of Colorspace used in Pixmap

    • Fixed #2179: Incorrect documentation for pixmap.tint_with()

    • Fixed #2208: Pushbutton widget appears as check box

    • Fixed #2210: apply_redactions() move pdf text to right after redaction

    • Fixed #2220: Page.delete_image() | object has no attribute is_image

    • Fixed #2228: open some pdf cost too much time

    • Fixed #2238: Bug - can not extract data from file in the newest version 1.21.1

    • Fixed #2242: Python quits silently in Story.element_positions() if callback function prototype is wrong

    • Fixed #2246: TextWriter write text in a wrong position

    • Fixed #2248: After redacting the content, the position of the remaining text changes

    • Fixed #2250: docs: unclear or broken link in page.rst

    • Fixed #2251: mupdf_display_errors does not apply to Pixmap when loading broken image

    • Fixed #2270: Annot.get_text("words") - doesn’t return the first line of words

    • Fixed #2275: insert_image: document that rotations are counterclockwise

    • Fixed #2278: Can not make widget (checkbox) to read-only by adding flags PDF_FIELD_IS_READ_ONLY

    • Fixed #2290: Different image format/data from Page.get_text(“dict”) and Fitz.get_page_images()

    • Fixed #2293: 68 failed tests when installing from sdist on my box

    • Fixed #2300: Too much recursion in tree (parents), makes program terminate

    • Fixed #2322: add_highlight_annot using clip generates “A Number is Out of Range” error in PDF

  • Other:

    • Add key “/AS (Yes)” to the underlying annot object of a selected button form field.

    • Remove unused Document methods has_xref_streams() and has_old_style_xrefs() as MuPDF equivalents have been removed.

    • Add new Document methods and properties for getting/setting /PageMode, /PageLayout and /MarkInfo.

    • New Document property version_count, which contains the number of incremental saves plus one.

    • New Document property is_fast_webaccess which tells whether the document is linearized.

    • DocumentWriter is now a context manager.

    • Add support for Pixmap JPEG output.

    • Add support for drawing rectangles with rounded corners.

    • get_drawings(): added optional extended arg.

    • Fixed issue where trace devices’ state was not being initialised correctly; data returned from things like fitz.Page.get_texttrace() might be slightly altered, e.g. linewidth values.

    • Output warning to stderr if it looks like we are being used with current directory containing an invalid fitz/ directory, because this can break import of fitz module. For example this happens if one attempts to use fitz when current directory is a PyMuPDF checkout.

  • Documentation:

    • General rework:

      • Introduces a new home page and new table of contents.

      • Structural update to include new About section.

      • Comparison & performance graphing.

      • Includes performance methodology in appendix.

      • Updates conf.py to understand single back-ticks as code.

      • Converts double back-ticks to single back-ticks.

      • Removes redundant files.

    • Improve insert_file() documentation.

    • get_bboxlog(): aded optional layers to get_bboxlog().

    • Page.get_texttrace(): add new dictionary key layer, name of Optional Content Group.

    • Mention use of Python venv in installation documentation.

    • Added missing fix for #2057 to release 1.21.1’s changelog.

    • Fixes many links to the PyMuPDF-Utilities repo scripts.

    • Avoid duplication of changes.txt and docs/changes.rst.

  • Build

    • Added pyproject.toml file to improve builds using pip etc.

Changes in Version 1.21.1 (2022-12-13)

  • This release uses MuPDF-1.21.1.

  • Bug fixes:

    • Fixed #2110: Fully embedded font is extracted only partially if it occupies more than one object

    • Fixed #2094: Rectangle Detection Logic

    • Fixed #2088: Destination point not set for named links in toc

    • Fixed #2087: Image with Filter “[/FlateDecode/JPXDecode]” not extracted

    • Fixed #2086: Document.save() owner_pw & user_pw has buffer overflow bug

    • Fixed #2076: Segfault in fitz.py

    • Fixed #2057: Document.save garbage parameter not working in PyMuPDF 1.21.0

    • Fixed #2051: Missing DPI Parameter

    • Fixed #2048: Invalid size of TextPage and bbox with newest version 1.21.0

    • Fixed #2045: SystemError: <built-in function Page_get_texttrace> returned a result with an error set

    • Fixed #2039: 1.21.0 fails to build against system libmupdf

    • Fixed #2036: Archive::Archive defined twice

  • Other

    • Swallow “&zoom=nan” in link uri strings.

    • Add new Page utility methods Page.replace_image() and Page.delete_image().

  • Documentation:

    • #2040: Added note about test failure with non-default build of MuPDF, to tests/README.md.

    • #2037: In docs/installation.rst, mention incompatibility with chocolatey.org on Windows.

    • #2061: Fixed description of Annot.file_info.

    • #2065: Show how to insert internal PDF link.

    • Improved description of building from source without an sdist.

    • Added information about running tests.

    • #2084: Fixed broken link to PyMuPDF-Utilities.

Changes in Version 1.21.0 (2022-11-8)

  • This release uses MuPDF-1.21.0.

  • New feature: Stories.

  • Added wheels for Python-3.11.

  • Bug fixes:

    • Fixed #1701: Broken custom image insertion.

    • Fixed #1854: Document.delete_pages() declines keyword arguments.

    • Fixed #1868: Access Violation Error at page.apply_redactions().

    • Fixed #1909: Adding text with fontname="Helvetica" can silently fail.

    • Fixed #1913: draw_rect(): does not respect width if color is not specified.

    • Fixed #1917: subset_fonts(): make it possible to silence the stdout.

    • Fixed #1936: Rectangle detection can be incorrect producing wrong output.

    • Fixed #1945: Segmentation fault when saving with clean=True.

    • Fixed #1965: pdfocr_save() Hard Crash.

    • Fixed #1971: Segmentation fault when using get_drawings().

    • Fixed #1946: block_no and block_type switched in get_text() docs.

    • Fixed #2013: AttributeError: ‘Widget’ object has no attribute ‘_annot’ in delete widget.

  • Misc changes to core code:

    • Fixed various compiler warnings and a sequence-point bug.

    • Added support for Memento builds.

    • Fixed leaks detected by Memento in test suite.

    • Fixed handling of exceptions in set_name() and set_rect().

    • Allow build with latest MuPDF, for regular testing of PyMuPDF master.

    • Cope with new MuPDF exceptions when setting rect for some Annot types.

    • Reduced cosmetic differences between MuPDF’s config.h and PyMuPDF’s _config.h.

    • Cope with various changes to MuPDF API.

  • Other:

    • Fixed various broken links and typos in docs.

    • Mention install of swig-python on MacOS for #875.

    • Added (untested) wheels for macos-arm64.

Changes in Version 1.20.2

  • This release uses MuPDF-1.20.3.

  • Fixed #1787. Fix linking issues on Unix systems.

  • Fixed #1824. SegFault when applying redactions overlapping a transparent image. (Fixed in MuPDF-1.20.3.)

  • Improvements to documentation:

    • Improved information about building from source in docs/installation.rst.

    • Clarified memory allocation setting JM_MEMORY` in ``docs/tools.rst.

    • Fixed link to PDF Reference manual in docs/app3.rst.

    • Fixed building of html documentation on OpenBSD.

    • Moved old docs/faq.rst into separate docs/recipes-* files.

  • Removed some unused files and directories:

    • installation/

    • docs/wheelnames.txt

Changes in Version 1.20.1

  • Fixed #1724. Fix for building on FreeBSD.

  • Fixed #1771. linkDest() had a broken call to re.match(), introduced in 1.20.0.

  • Fixed #1751. get_drawings() and get_cdrawings() previously always returned with closePath=False.

  • Fixed #1645. Default FreeText annotation text color is now black.

  • Improvements to sphinx-generated documentation:

    • Use readthedocs theme with enhancements.

    • Renamed the .txt files to have .rst suffixes.

Changes in Version 1.20.0

This release uses MuPDF-1.20.0, released 2022-06-15.

  • Cope with new MuPDF link uri format, changed from #<int>,<int>,<int> to #page=<int>&zoom=<float>,<float>,<float>.

  • In tests/test_insertpdf.py, use new reference output joined-1.20.pdf. We also check that new output values are approximately the same as the old ones.

  • Fixed #1738. Leak of pdf_graft_map. Also fixed a SEGV issue that this seemed to expose, caused by incorrect freeing of underlying fz_document.

  • Fixed #1733. Fixed ownership of Annotation.get_pixmap().

Changes to build/release process:

  • If pip builds from source because an appropriate wheel is not available, we no longer require MuPDF to be pre-installed. Instead the required MuPDF source is embedded in the sdist and automatically built into PyMuPDF.

  • Various changes to setup.py to download the required MuPDF release as required. See comments at start of setup.py for details.

  • Added .github/workflows/build_wheels.yml to control building of wheels on Github.

Changes in Version 1.19.6

  • Fixed #1620. The TextPage created by Page.get_textpage() will now be freed correctly (removed memory leak).

  • Fixed #1601. Document open errors should now be more concise and easier to interpret. In the course of this, two PyMuPDF-specific Python exceptions have been added:

    • EmptyFileError – raised when trying to create a Document (fitz.open()) from an empty file or zero-length memory.

    • FileDataError – raised when MuPDF encounters irrecoverable document structure issues.

  • Added Page.load_widget() given a PDF field’s xref.

  • Added Dictionary pdfcolor which provide the about 500 colors defined as PDF color values with the lower case color name as key.

  • Added algebra functionality to the Quad class. These objects can now also be added and subtracted among themselves, and be multiplied by numbers and matrices.

  • Added new constants defining the default text extraction flags for more comfortable handling. Their naming convention is like TEXTFLAGS_WORDS for page.get_text("words"). See 文本提取标志默认值.

  • Changed Page.annots() and Page.widgets() to detect and prevent reloading the page (illegally) inside the iterator loops via Document.reload_page(). Doing this brings down the interpretor. Documented clean ways to do annotation and widget mass updates within properly designed loops.

  • Changed several internal utility functions to become standalone (“SWIG inline”) as opposed to be part of the Tools class. This, among other things, increases the performance of geometry object creation.

  • Changed Document.update_stream() to always accept stream updates - whether or not the dictionary object behind the xref already is a stream. Thus the former new parameter is now ignored and will be removed in v1.20.0.

Changes in Version 1.19.5

  • Fixed #1518. A limited “fix”: in some cases, rectangles and quadrupels were not correctly encoded to support re-drawing by Shape.

  • Fixed #1521. This had the same ultimate reason behind issue #1510.

  • Fixed #1513. Some Optional Content functions did not support non-ASCII characters.

  • Fixed #1510. Support more soft-mask image subtypes.

  • Fixed #1507. Immunize against items in the outlines chain, that are "null" objects.

  • Fixed re-opened #1417. (“too many open files”). This was due to insufficient calls to MuPDF’s fz_drop_document(). This also fixes #1550.

  • Fixed several undocumented issues in relation to incorrectly setting the text span origin point_like.

  • Fixed undocumented error computing the character bbox in method Page.get_texttrace() when text is flipped (as opposed to just rotated).

  • Added items to the dictionary returned by image_properties(): orientation and transform report the natural image orientation (EXIF data).

  • Added method Document.xref_copy(). It will make a given target PDF object an exact copy of a source object.

Changes in Version 1.19.4

  • Fixed #1505. Immunize against circular outline items.

  • Fixed #1484. Correct CropBox coordinates are now returned in all situations.

  • Fixed #1479.

  • Fixed #1474. TextPage objects are now properly deleted again.

  • Added Page methods and attributes for PDF /ArtBox, /BleedBox, /TrimBox.

  • Added global attribute TESSDATA_PREFIX for easy checking of OCR support.

  • Changed Document.xref_set_key() such that dictionary keys will physically be removed if set to value "null".

  • Changed Document.extract_font() to optionally return a dictionary (instead of a tuple).

Changes in Version 1.19.3

This patch version implements minor improvements for Pixmap and also some important fixes.

  • Fixed #1351. Reverted code that introduced the memory growth in v1.18.15.

  • Fixed #1417. Developped circumvention for growth of open file handles using Document.insert_pdf().

  • Fixed #1418. Developped circumvention for memory growth using Document.insert_pdf().

  • Fixed #1430. Developped circumvention for mass pixmap generations of document pages.

  • Fixed #1433. Solves a bbox error for some Type 3 font in PyMuPDF text processing.

  • Added Pixmap.color_topusage() to determine the share of the most frequently used color. Solves #1397.

  • Added Pixmap.warp() which makes a new pixmap from a given arbitrary convex quad inside the pixmap.

  • Added Annot.irt_xref and Annot.set_irt_xref() to inquire or set the /IRT (“In Responde To”) property of an annotation. Implements #1450.

  • Added Rect.torect() and IRect.torect() which compute a matrix that transforms to a given other rectangle.

  • Changed Pixmap.color_count() to also return the count of each color.

  • Changed Page.get_texttrace() to also return correct span and character bboxes if span["dir"] != (1, 0).

Changes in Version 1.19.2

This patch version implements minor improvements for Page.get_drawings() and also some important fixes.

  • Fixed #1388. Fixed intermittent memory corruption when insert or updating annotations.

  • Fixed #1375. Inconsistencies between line numbers as returned by the “words” and the “dict” options of Page.get_text() have been corrected.

  • Fixed #1364. The check for being a "rawdict" span in recover_span_quad() now works correctly.

  • Fixed #1342. Corrected the check for rectangle infiniteness in Page.show_pdf_page().

  • Changed Page.get_drawings(), Page.get_cdrawings() to return an indicator on the area orientation covered by a rectangle. This implements #1355. Also, the recognition rate for rectangles and quads has been significantly improved.

  • Changed all text search and extraction methods to set the new flags option TEXT_MEDIABOX_CLIP to ON by default. That bit causes the automatic suppression of all characters that are completely outside a page’s mediabox (in as far as that notion is supported for a document type). This eliminates the need for using clip=page.rect or similar for omitting text outside the visible area.

  • Added parameter "dpi" to Page.get_pixmap() and Annot.get_pixmap(). When given, parameter "matrix" is ignored, and a Pixmap with the desired dots per inch is created.

  • Added attributes Pixmap.is_monochrome and Pixmap.is_unicolor allowing fast checks of pixmap properties. Addresses #1397.

  • Added method Pixmap.color_count() to determine the unique colors in the pixmap.

  • Added boolean parameter "compress" to PDF document method Document.update_stream(). Addresses / enables solution for #1408.

Changes in Version 1.19.1

This is the first patch version to support MuPDF v1.19.0. Apart from one bug fix, it includes important improvements for OCR support and the option to sort extracted text to the standard reading order “from top-left to bottom-right”.

  • Fixed #1328. “words” text extraction again returns correct (x0, y0) coordinates.

  • Changed Page.get_textpage_ocr(): it now supports parameter dpi to control OCR quality. It is also possible to choose whether the full page should be OCRed or only the images displayed by the page.

  • Changed Page.get_drawings() and Page.get_cdrawings() to automatically convert colors to RGB color tuples. Implements #1332. Similar change was applied to Page.get_texttrace().

  • Changed Page.get_text() to support a parameter sort. If set to True the output is conveniently sorted.

Changes in Version 1.19.0

This is the first version supporting MuPDF 1.19.*, published 2021-10-05. It introduces many new features compared to the previous version 1.18.*.

PyMuPDF has now picked up integrated Tesseract OCR support, which was already present in MuPDF v1.18.0.

  • Supported images can be OCRed via their Pixmap which results in a 1-page PDF with a text layer.

  • All supported document pages (i.e. not only PDFs), can be OCRed using specialized text extraction methods. The result is a mixture of standard and OCR text (depending on which part of the page was deemed to require OCRing) that can be searched and extracted without restrictions.

  • All this requires an independent installation of Tesseract. MuPDF actually (only) needs the location of Tesseract’s "tessdata" folder, where its language support data are stored. This location must be available as environment variable TESSDATA_PREFIX.

A new MuPDF feature is journalling PDF updates, which is also supported by this PyMuPDF version. Changes may be logged, rolled back or replayed, allowing to implement a whole new level of control over PDF document integrity – similar to functions present in modern database systems.

A third feature (unrelated to the new MuPDF version) includes the ability to detect when page objects cover or hide each other. It is now e.g. possible to see that text is covered by a drawing or an image.

  • Changed terminology and meaning of important geometry concepts: Rectangles are now characterized as finite, valid or empty, while the definitions of these terms have also changed. Rectangles specifically are now thought of being “open”: not all corners and sides are considered part of the retangle. Please do read the Rect section for details.

  • Added new parameter "no_new_id" to Document.save() / Document.tobytes() methods. Use it to suppress updating the second item of the document /ID which in PDF indicates that the original file has been updated. If the PDF has no /ID at all yet, then no new one will be created either.

  • Added a journalling facility for PDF updates. This allows logging changes, undoing or redoing them, or saving the journal for later use. Refer to Document.journal_enable() and friends.

  • Added new Pixmap methods Pixmap.pdfocr_save() and Pixmap.pdfocr_tobytes(), which generate a 1-page PDF containing the pixmap as PNG image with OCR text layer.

  • Added Page.get_textpage_ocr() which executes optical character recognition for the page, then extracts the results and stores them together with “normal” page content in a TextPage. Use or reuse this object in subsequent text extractions and text searches to avoid multiple efforts. The existing text search and text extraction methods have been extended to support a separately created textpage – see next item.

  • Added a new parameter textpage to text extraction and text search methods. This allows reuse of a previously created TextPage and thus achieves significant runtime benefits – which is especially important for the new OCR features. But “normal” text extractions can definitely also benefit.

  • Added Page.get_texttrace(), a technical method delivering low-level text character properties. It was present before as a private method, but the author felt it now is mature enough to be officially available. It specifically includes a “sequence number” which indicates the page appearance build operation that painted the text.

  • Added Page.get_bboxlog() which delivers the list of rectangles of page objects like text, images or drawings. Its significance lies in its sequence: rectangles intersecting areas with a lower index are covering or hiding them.

  • Changed methods Page.get_drawings() and Page.get_cdrawings() to include a “sequence number” indicating the page appearance build operation that created the drawing.

  • Fixed #1311. Field values in comboboxes should now be handled correctly.

  • Fixed #1290. Error was caused by incorrect rectangle emptiness check, which is fixed due to new geometry logic of this version.

  • Fixed #1286. Text alignment for redact annotations is working again.

  • Fixed #1287. Infinite loop issue for non-Windows systems when applying some redactions has been resolved.

  • Fixed #1284. Text layout destruction after applying redactions in some cases has been resolved.

Changes in Version 1.18.18 / 1.18.19

  • Fixed issue #1266. Failure to set Pixmap.samples in important cases, was hotfixed in a new version 1.18.19.

  • Fixed issue #1257. Removing the read-only flag from PDF fields is now possible.

  • Fixed issue #1252. Now correctly specifying the zoom value for PDF link annotations.

  • Fixed issue #1244. Now correctly computing the transform matrix in Page.get_image__bbox().

  • Fixed issue #1241. Prevent returning artifact characters in Page.get_textbox(), which happened in certain constellations.

  • Fixed issue #1234. Avoid creating infinite rectangles in corner cases – Page.get_drawings(), Page.get_cdrawings().

  • Added test data and test scripts to the source PyPI source distribution.

Changes in Version 1.18.17

Focus of this version are major performance improvements of selected functions.

  • Fixed issue #1199. Using a non-existing page number in Document.get_page_images() and friends will no longer lead to segfaults.

  • Changed Page.get_drawings() to now differentiate between “stroke”, “fill” and combined paths. Paths containing more than one rectangle (i.e. “re” items) are now supported. Extracting “clipped” paths is now available as an option.

  • Added Page.get_cdrawings(), performance-optimized version of Page.get_drawings().

  • Added Pixmap.samples_mv, memoryview of a pixmap’s pixel area. Does not copy and thus always accesses the current state of that area.

  • Added Pixmap.samples_ptr, Python “pointer” to a pixmap’s pixel area. Allows much faster creation (factor 800+) of Qt images.

Changes in Version 1.18.16

  • Fixed issue #1184. Existing PDF widget fonts in a PDF are now accepted (i.e. not forcedly changed to a Base-14 font).

  • Fixed issue #1154. Text search hits should now be correct when clip is specified.

  • Fixed issue #1152.

  • Fixed issue #1146.

  • Added Link.flags and Link.set_flags() to the Link class. Implements enhancement requests #1187.

  • Added option to simulate TextWriter.fill_textbox() output for predicting the number of lines, that a given text would occupy in the textbox.

  • Added text output support as subcommand gettext to the fitz CLI module. Most importantly, original physical text layout reproduction is now supported.

Changes in Version 1.18.15

  • Fixed issue #1088. Removing an annotation’s fill color should now work again both ways, using the fill_color=[] argument in Annot.update() as well as fill=[] in Annot.set_colors().

  • Fixed issue #1081. Document.subset_fonts(): fixed an error which created wrong character widths for some fonts.

  • Fixed issue #1078. Page.get_text() and other methods related to text extraction: changed the default value of the TextPage flags parameter. All whitespace and ligatures are now preserved.

  • Fixed issue #1085. The old snake_cased alias of fitz.detTextlength is now defined correctly.

  • Changed Document.subset_fonts() will now correctly prefix font subsets with an appropriate six letter uppercase tag, complying with the PDF specification.

  • Added new method Widget.button_states() which returns the possible values that a button-type field can have when being set to “on” or “off”.

  • Added support of text with Small Capital letters to the Font and TextWriter classes. This is reflected by an additional bool parameter small_caps in various of their methods.

Changes in Version 1.18.14

  • Finished implementing new, “snake_cased” names for methods and properties, that were “camelCased” and awkward in many aspects. At the end of this documentation, there is section 已废弃名称 with more background and a mapping of old to new names.

  • Fixed issue #1053. Page.insert_image(): when given, include image mask in the hash computation.

  • Fixed issue #1043. Added Pixmap.getPNGdata to the aliases of Pixmap.tobytes().

  • Fixed an internal error when computing the enveloping rectangle of drawn paths as returned by Page.get_drawings().

  • Fixed an internal error occasionally causing loops when outputting text via TextWriter.fill_textbox().

  • Added Font.char_lengths(), which returns a tuple of character widths of a string.

  • Added more ways to specify pages in Document.delete_pages(). Now a sequence (list, tuple or range) can be specified, and the Python del statement can be used. In the latter case, Python slices are also accepted.

  • Changed Document.del_toc_item(), which disables a single item of the TOC: previously, the title text was removed. Instead, now the complete item will be shown grayed-out by supporting viewers.

Changes in Version 1.18.13

Changes in Version 1.18.11 / 1.18.12

Changes in Version 1.18.10

Changes in Version 1.18.9

  • Fixed issue #888. Removed ambiguous statements concerning PyMuPDF’s license, which is now clearly stated to be GNU AGPL V3.

  • Fixed issue #895.

  • Fixed issue #896. Since v1.17.6 PyMuPDF suppresses the font subset tags and only reports the base fontname in text extraction outputs “dict” / “json” / “rawdict” / “rawjson”. Now a new global parameter can request the old behaviour, Tools.set_subset_fontnames().

  • Fixed issue #885. Pixmap creation now also works with filenames given as pathlib.Paths.

  • Changed Document.subset_fonts(): Text is not rewritten any more and should therefore retain all its origial properties – like being hidden or being controlled by Optional Content mechanisms.

  • Changed TextWriter output to also accept text in right to left mode (Arabian, Hebrew): TextWriter.fill_textbox(), TextWriter.append(). These methods now accept a new boolean parameter right_to_left, which is False by default. Implements #897.

  • Changed TextWriter.fill_textbox() to return all lines of text, that did not fit in the given rectangle. Also changed the default of the warn parameter to no longer print a warning message in overflow situations.

  • Added a utility function recover_quad(), which computes the quadrilateral of a span. This function can be used for correctly marking text extracted with the “dict” or “rawdict” options of Page.get_text().

Changes in Version 1.18.8

This is a bug fix version only. We are publishing early because of the potentially widely used functions.

  • Fixed issue #881. Fixed a memory leak in Page.insert_image() when inserting images from files or memory.

  • Fixed issue #878. pathlib.Path objects should now correctly handle file path hierarchies.

Changes in Version 1.18.7

We continued the process of renaming methods and properties from “mixedCase” to “snake_case”. Documentation usually mentions the new names only, but old, deprecated names remain available for some time.

Changes in Version 1.18.6

  • Fixed issue #812.

  • Fixed issue #793. Invalid document metadata previously prevented opening some documents at all. This error has been removed.

  • Fixed issue #792. Text search and text extraction will make no rectangle containment checks at all if the default clip=None is used.

  • Fixed issue #785.

  • Fixed issue #780. Corrected a parameter check error.

  • Fixed issue #779. Fixed typo

  • Added an option to set the desired line height for text boxes. Implements #804.

  • Changed text position retrieval to better cope with Tesseract’s glyphless font. Implements #803.

  • Added an option to choose the prefix of new annotations, fields and links for providing unique annotation ids. Implements request #807.

  • Added getting and setting color and text properties for Table of Contents items for PDFs. Implements #779.

  • Added PDF page label handling: Page.get_label() returns the page label, Document.get_page_numbers() return all page numbers having a specified label, and Document.set_page_labels() adds or updates a PDF’s page label definition.

备注

This version introduces Python type hinting. The goal is to provide each parameter and the return value of all functions and methods with type information. This still is work in progress although the majority of functions has already been handled.

Changes in Version 1.18.5

Apart from several fixes, this version also focusses on several minor, but important feature improvements. Among the latter is a more precise computation of proper line heights and insertion points for writing / inserting text. As opposed to using font-agnostic constants, these values are now taken from the font’s properties.

Also note that this is the first version which does no longer provide pregenerated wheels for Python versions older than 3.6. PIP also discontinues support for these by end of this year 2020.

  • Fixed issue #771. By using “small glyph heights” option, the full page text can be extracted.

  • Fixed issue #768.

  • Fixed issue #750.

  • Fixed issue #739. The “dict”, “rawdict” and corresponding JSON output variants now have two new span keys: "ascender" and "descender". These floats represent special font properties which can be used to compute bboxes of spans or characters of exactly fontsize height (as opposed to the default line height). An example algorithm is shown in section “Span Dictionary” here. Also improved the detection and correction of ill-specified ascender / descender values encountered in some fonts.

  • Added a new, experimental Tools.set_small_glyph_heights() – also in response to issue #739. This method sets or unsets a global parameter to always compute bboxes with fontsize height. If “on”, text searching and all text extractions will returned rectangles, bboxes and quads with a smaller height.

  • Fixed issue #728.

  • Changed fill color logic of ‘Polyline’ annotations: this parameter now only pertains to line end symbols – the annotation itself can no longer have a fill color. Also addresses issue #727.

  • Changed Page.getImageBbox() to also compute the bbox if the image is contained in an XObject.

  • Changed Shape.insertTextbox(), resp. Page.insertTextbox(), resp. TextWriter.fillTextbox() to respect font’s properties “ascender” / “descender” when computing line height and insertion point. This should no longer lead to line overlaps for multi-line output. These methods used to ignore font specifics and used constant values instead.

Changes in Version 1.18.4

This version adds several features to support PDF Optional Content. Among other things, this includes OCMDs (Optional Content Membership Dictionaries) with the full scope of “visibility expressions” (PDF key /VE), text insertions (including the TextWriter class) and drawings.

  • Fixed issue #727. Freetext annotations now support an uncolored rectangle when fill_color=None.

  • Fixed issue #726. UTF-8 encoding errors are now handled for HTML / XML Page.getText() output.

  • Fixed issue #724. Empty values are no longer stored in the PDF /Info metadata dictionary.

  • Added new methods Document.set_oc() and Document.get_oc() to set or get optional content references for existing image and form XObjects. These methods are similar to the same-named methods of Annot.

  • Added Document.set_ocmd(), Document.get_ocmd() for handling OCMDs.

  • Added Optional Content support for text insertion and drawing.

  • Added new method Page.deleteWidget(), which deletes a form field from a page. This is analogous to deleting annotations.

  • Added support for Popup annotations. This includes defining the Popup rectangle and setting the Popup to open or closed. Methods / attributes Annot.set_popup(), Annot.set_open(), Annot.has_popup, Annot.is_open, Annot.popup_rect, Annot.popup_xref.

Other changes:

  • The naming of methods and attributes in PyMuPDF is far from being satisfactory: we have CamelCases, mixedCases and lower_case_with_underscores all over the place. With the Annot as the first candidate, we have started an activity to clean this up step by step, converting to lower case with underscores for methods and attributes while keeping UPPERCASE for the constants.

    • Old names will remain available to prevent code breaks, but they will no longer be mentioned in the documentation.

    • New methods and attributes of all classes will be named according to the new standard.

Changes in Version 1.18.3

As a major new feature, this version introduces support for PDF’s Optional Content concept.

  • Fixed issue #714.

  • Fixed issue #711.

  • Fixed issue #707: if a PDF user password, but no owner password is supplied nor present, then the user password is also used as the owner password.

  • Fixed expand and deflate parameters of methods Document.save() and Document.write(). Individual image and font compression should now finally work. Addresses issue #713.

  • Added a support of PDF optional content. This includes several new Document methods for inquiring and setting optional content status and adding optional content configurations and groups. In addition, images, form XObjects and annotations now can be bound to optional content specifications. Resolved issue #709.

Changes in Version 1.18.2

This version contains some interesting improvements for text searching: any number of search hits is now returned and the hit_max parameter was removed. The new clip parameter in addition allows to restrict the search area. Searching now detects hyphenations at line breaks and accordingly finds hyphenated words.

  • Fixed issue #575: if using quads=False in text searching, then overlapping rectangles on the same line are joined. Previously, parts of the search string, which belonged to different “marked content” items, each generated their own rectangle – just as if occurring on separate lines.

  • Added Document.isRepaired, which is true if the PDF was repaired on open.

  • Added Document.setXmlMetadata() which either updates or creates PDF XML metadata. Implements issue #691.

  • Added Document.getXmlMetadata() returns PDF XML metadata.

  • Changed creation of PDF documents: they will now always carry a PDF identification (/ID field) in the document trailer. Implements issue #691.

  • Changed Page.searchFor(): a new parameter clip is accepted to restrict the search to this rectangle. Correspondingly, the attribute TextPage.rect is now respected by TextPage.search().

  • Changed parameter hit_max in Page.searchFor() and TextPage.search() is now obsolete: methods will return all hits.

  • Changed character selection criteria in Page.getText(): a character is now considered to be part of a clip if its bbox is fully contained. Before this, a non-empty intersection was sufficient.

  • Changed Document.scrub() to support a new option redact_images. This addresses issue #697.

Changes in Version 1.18.1

  • Fixed issue #692. PyMuPDF now detects and recovers from more cyclic resource dependencies in PDF pages and for the first time reports them in the MuPDF warnings store.

  • Fixed issue #686.

  • Added opacity options for the Shape class: Stroke and fill colors can now be set to some transparency value. This means that all Page draw methods, methods Page.insertText(), Page.insertTextbox(), Shape.finish(), Shape.insertText(), and Shape.insertTextbox() support two new parameters: stroke_opacity and fill_opacity.

  • Added new parameter mask to Page.insertImage() for optionally providing an external image mask. Resolves issue #685.

  • Added Annot.soundGet() for extracting the sound of an audio annotation.

Changes in Version 1.18.0

This is the first PyMuPDF version supporting MuPDF v1.18. The focus here is on extending PyMuPDF’s own functionality – apart from bug fixing. Subsequent PyMuPDF patches may address features new in MuPDF.

  • Fixed issue #519. This upstream bug occurred occasionally for some pages only and seems to be fixed now: page layout should no longer be ruined in these cases.

  • Fixed issue #675.

    • Unsuccessful storage allocations should now always lead to exceptions (circumvention of an upstream bug intermittently crashing the interpreter).

    • Pixmap size is now based on size_t instead of int in C and should be correct even for extremely large pixmaps.

  • Fixed issue #668. Specification of dashes for PDF drawing insertion should now correctly reflect the PDF spec.

  • Fixed issue #669. A major source of memory leakage in Page.insert_pdf() has been removed.

  • Added keyword “images” to Page.apply_redactions() for fine-controlling the handling of images.

  • Added Annot.getText() and Annot.getTextbox(), which offer the same functionality as the Page versions.

  • Added key “number” to the block dictionaries of Page.getText() / Annot.getText() for options “dict” and “rawdict”.

  • Added glyph_name_to_unicode() and unicode_to_glyph_name(). Both functions do not really connect to a specific font and are now independently available, too. The data are now based on the Adobe Glyph List.

  • Added convenience functions adobe_glyph_names() and adobe_glyph_unicodes() which return the respective available data.

  • Added Page.getDrawings() which returns details of drawing operations on a document page. Works for all document types.

  • Improved performance of Document.insert_pdf(). Multiple object copies are now also suppressed across multiple separate insertions from the same source. This saves time, memory and target file size. Previously this mechanism was only active within each single method execution. The feature can also be suppressed with the new method bool parameter final=1, which is the default.

  • For PNG images created from pixmaps, the resolution (dpi) is now automatically set from the respective Pixmap.xres and Pixmap.yres values.

Changes in Version 1.17.7

  • Fixed issue #651. An upstream bug causing interpreter crashes in corner case redaction processings was fixed by backporting MuPDF changes from their development repo.

  • Fixed issue #645. Pixmap top-left coordinates can be set (again) by their own method, Pixmap.set_origin().

  • Fixed issue #622. Page.insertImage() again accepts a rect_like parameter.

  • Added severeal new methods to improve and speed-up table of contents (TOC) handling. Among other things, TOC items can now changed or deleted individually – without always replacing the complete TOC. Furthermore, access to some PDF page attributes is now possible without first loading the page. This has a very significant impact on the performance of TOC manipulation.

  • Added an option to Document.insert_pdf() which allows displaying progress messages. Adresses #640.

  • Added Page.getTextbox() which extracts text contained in a rectangle. In many cases, this should obsolete writing your own script for this type of thing.

  • Added new clip parameter to Page.getText() to simplify and speed up text extraction of page sub areas.

  • Added TextWriter.appendv() to add text in vertical write mode. Addresses issue #653

Changes in Version 1.17.6

  • Fixed issue #605

  • Fixed issue #600 – text should now be correctly positioned also for pages with a CropBox smaller than MediaBox.

  • Added text span dictionary key origin which contains the lower left coordinate of the first character in that span.

  • Added attribute Font.buffer, a bytes copy of the font file.

  • Added parameter sanitize to Page.cleanContents(). Allows switching of sanitization, so only syntax cleaning will be done.

Changes in Version 1.17.5

  • Fixed issue #561 – second go: certain TextWriter usages with many alternating fonts did not work correctly.

  • Fixed issue #566.

  • Fixed issue #568.

  • Fixed – opacity is now correctly taken from the TextWriter object, if not given in TextWriter.writeText().

  • Added a new global attribute fitz_fontdescriptors. Contains information about usable fonts from repository pymupdf-fonts.

  • Added Font.valid_codepoints() which returns an array of unicode codepoints for which the font has a glyph.

  • Added option text_as_path to Page.getSVGimage(). this implements #580. Generates much smaller SVG files with parseable text if set to False.

Changes in Version 1.17.4

  • Fixed issue #561. Handling of more than 10 Font objects on one page should now work correctly.

  • Fixed issue #562. Annotation pixmaps are no longer derived from the page pixmap, thus avoiding unintended inclusion of page content.

  • Fixed issue #559. This MuPDF bug is being temporarily fixed with a pre-version of MuPDF’s next release.

  • Added utility function repair_mono_font() for correcting displayed character spacing for some mono-spaced fonts.

  • Added utility method Document.need_appearances() for fine-controlling Form PDF behavior. Addresses issue #563.

  • Added utility function sRGB_to_pdf() to recover the PDF color triple for a given color integer in sRGB format.

  • Added utility function sRGB_to_rgb() to recover the (R, G, B) color triple for a given color integer in sRGB format.

  • Added utility function make_table() which delivers table cells for a given rectangle and desired numbers of columns and rows.

  • Added support for optional fonts in repository pymupdf-fonts.

Changes in Version 1.17.3

  • Fixed an undocumented issue, which prevented fully cleaning a PDF page when using Page.cleanContents().

  • Fixed issue #540. Text extraction for EPUB should again work correctly.

  • Fixed issue #548. Documentation now includes LINK_NAMED.

  • Added new parameter to control start of text in TextWriter.fillTextbox(). Implements #549.

  • Changed documentation of Page.add_redact_annot() to explain the usage of non-builtin fonts.

Changes in Version 1.17.2

  • Fixed issue #533.

  • Added options to modify ‘Redact’ annotation appearance. Implements #535.

Changes in Version 1.17.1

  • Fixed issue #520.

  • Fixed issue #525. Vertices for ‘Ink’ annots should now be correct.

  • Fixed issue #524. It is now possible to query and set rotation for applicable annotation types.

Also significantly improved inline documentation for better support of interactive help.

Changes in Version 1.17.0

This version is based on MuPDF v1.17. Following are highlights of new and changed features:

  • Added extended language support for annotations and widgets: a mixture of Latin, Greece, Russian, Chinese, Japanese and Korean characters can now be used in ‘FreeText’ annotations and text widgets. No special arrangement is required to use it.

  • Faster page access is implemented for documents supporting a “chapter” structure. This applies to EPUB documents currently. This comes with several new Document methods and changes for Document.loadPage() and the “indexed” page access doc[n]: In addition to specifying a page number as before, a tuple (chaper, pno) can be specified to identify the desired page.

  • Changed: Improved support of redaction annotations: images overlapped by redactions are permanantly modified by erasing the overlap areas. Also links are removed if overlapped by redactions. This is now fully in sync with PDF specifications.

Other changes:

  • Changed TextWriter.writeText() to support the “morph” parameter.

  • Added methods Rect.morph(), IRect.morph(), and Quad.morph(), which return a new Quad.

  • Changed Page.add_freetext_annot() to support text alignment via a new “align” parameter.

  • Fixed issue #508. Improved image rectangle calculation to hopefully deliver correct values in most if not all cases.

  • Fixed issue #502.

  • Fixed issue #500. Document.convertToPDF() should no longer cause memory leaks.

  • Fixed issue #496. Annotations and widgets / fields are now added or modified using the coordinates of the unrotated page. This behavior is now in sync with other methods modifying PDF pages.

  • Added Page.rotationMatrix and Page.derotationMatrix to support coordinate transformations between the rotated and the original versions of a PDF page.

Potential code breaking changes:

  • The private method Page._getTransformation() has been removed. Use the public Page.transformationMattrix instead.

Changes in Version 1.16.18

This version introduces several new features around PDF text output. The motivation is to simplify this task, while at the same time offering extending features.

One major achievement is using MuPDF’s capabilities to dynamically choosing fallback fonts whenever a character cannot be found in the current one. This seemlessly works for Base-14 fonts in combination with CJK fonts (China, Japan, Korea). So a text may contain any combination of characters from the Latin, Greek, Russian, Chinese, Japanese and Korean languages.

  • Fixed issue #493. Pixmap(doc, xref) should now again correctly resemble the loaded image object.

  • Fixed issue #488. Widget names are now modifiable.

  • Added new class Font which represents a font.

  • Added new class TextWriter which serves as a container for text to be written on a page.

  • Added Page.writeText() to write one or more TextWriter objects to the page.

Changes in Version 1.16.17

  • Fixed issue #479. PyMuPDF should now more correctly report image resolutions. This applies to both, images (either from images files or extracted from PDF documents) and pixmaps created from images.

  • Added Pixmap.set_dpi() which sets the image resolution in x and y directions.

Changes in Version 1.16.16

  • Fixed issue #477.

  • Fixed issue #476.

  • Changed annotation line end symbol coloring and fixed an error coloring the interior of ‘Polyline’ /’Polygon’ annotations.

Changes in Version 1.16.14

  • Changed text marker annotations to accept parameters beyond just quadrilaterals such that now text lines between two given points can be marked.

  • Added Document.scrub() which removes potentially sensitive data from a PDF. Implements #453.

  • Added Annot.blendMode() which returns the blend mode of annotations.

  • Added Annot.setBlendMode() to set the annotation’s blend mode. This resolves issue #416.

  • Changed Annot.update() to accept additional parameters for setting blend mode and opacity.

  • Added advanced graphics features to control the anti-aliasing values, Tools.set_aa_level(). Resolves #467

  • Fixed issue #474.

  • Fixed issue #466.

Changes in Version 1.16.13

  • Added Document.getPageXObjectList() which returns a list of Form XObjects of the page.

  • Added Page.setMediaBox() for changing the physical PDF page size.

  • Added Page methods which have been internal before: Page.cleanContents() (= Page._cleanContents()), Page.getContents() (= Page._getContents()), Page.getTransformation() (= Page._getTransformation()).

Changes in Version 1.16.12

  • Fixed issue #447

  • Fixed issue #461.

  • Fixed issue #397.

  • Fixed issue #463.

  • Added JavaScript support to PDF form fields, thereby fixing #454.

  • Added a new annotation method Annot.delete_responses(), which removes ‘Popup’ and response annotations referring to the current one. Mainly serves data protection purposes.

  • Added a new form field method Widget.reset(), which resets the field value to its default.

  • Changed and extended handling of redactions: images and XObjects are removed if contained in a redaction rectangle. Any partial only overlaps will just be covered by the redaction background color. Now an overlay text can be specified to be inserted in the rectangle area to take the place the deleted original text. This resolves #434.

Changes in Version 1.16.11

Changes in Version 1.16.10

  • Fixed issue #421 (“annot.set_rect(rect) has no effect on text Annotation”)

  • Fixed issue #417 (“Strange behavior for page.deleteAnnot on 1.16.9 compare to 1.13.20”)

  • Fixed issue #415 (“Annot.setOpacity throws mupdf warnings”)

  • Changed all “add annotation / widget” methods to store a unique name in the /NM PDF key.

  • Changed Annot.setInfo() to also accept direct parameters in addition to a dictionary.

  • Changed Annot.info to now also show the annotation’s unique id (/NM PDF key) if present.

  • Added Page.annot_names() which returns a list of all annotation names (/NM keys).

  • Added Page.load_annot() which loads an annotation given its unique id (/NM key).

  • Added Document.reload_page() which provides a new copy of a page after finishing any pending updates to it.

Changes in Version 1.16.9

  • Fixed #412 (“Feature Request: Allow controlling whether TOC entries should be collapsed”)

  • Fixed #411 (“Seg Fault with page.firstWidget”)

  • Fixed #407 (“Annot.setOpacity trouble”)

  • Changed methods Annot.setBorder(), Annot.setColors(), Link.setBorder(), and Link.setColors() to also accept direct parameters, and not just cumbersome dictionaries.

Changes in Version 1.16.8

  • Added several new methods to the Document class, which make dealing with PDF low-level structures easier. I also decided to provide them as “normal” methods (as opposed to private ones starting with an underscore “_”). These are Document.xrefObject(), Document.xrefStream(), Document.xrefStreamRaw(), Document.PDFTrailer(), Document.PDFCatalog(), Document.metadataXML(), Document.updateObject(), Document.updateStream().

  • Added Tools.mupdf_disply_errors() which sets the display of mupdf errors on sys.stderr.

  • Added a commandline facility. This a major new feature: you can now invoke several utility functions via “python -m fitz …”. It should obsolete the need for many of the most trivial scripts. Please refer to 命令行界面.

Changes in Version 1.16.7

Minor changes to better synchronize the binary image streams of TextPage image blocks and Document.extractImage() images.

  • Fixed issue #394 (“PyMuPDF Segfaults when using TOOLS.mupdf_warnings()”).

  • Changed redirection of MuPDF error messages: apart from writing them to Python sys.stderr, they are now also stored with the MuPDF warnings.

  • Changed Tools.mupdf_warnings() to automatically empty the store (if not deactivated via a parameter).

  • Changed Page.getImageBbox() to return an infinite rectangle if the image could not be located on the page – instead of raising an exception.

Changes in Version 1.16.6

  • Fixed issue #390 (“Incomplete deletion of annotations”).

  • Changed Page.searchFor() / Document.searchPageFor() to also support the flags parameter, which controls the data included in a TextPage.

  • Changed Document.getPageImageList(), Document.getPageFontList() and their Page counterparts to support a new parameter full. If true, the returned items will contain the xref of the Form XObject where the font or image is referenced.

Changes in Version 1.16.5

More performance improvements for text extraction.

  • Fixed second part of issue #381 (see item in v1.16.4).

  • Added Page.getTextPage(), so it is no longer required to create an intermediate display list for text extractions. Page level wrappers for text extraction and text searching are now based on this, which should improve performance by ca. 5%.

Changes in Version 1.16.4

  • Fixed issue #381 (“TextPage.extractDICT … failed … after upgrading … to 1.16.3”)

  • Added method Document.pages() which delivers a generator iterator over a page range.

  • Added method Page.links() which delivers a generator iterator over the links of a page.

  • Added method Page.annots() which delivers a generator iterator over the annotations of a page.

  • Added method Page.widgets() which delivers a generator iterator over the form fields of a page.

  • Changed Document.is_form_pdf to now contain the number of widgets, and False if not a PDF or this number is zero.

Changes in Version 1.16.3

Minor changes compared to version 1.16.2. The code of the “dict” and “rawdict” variants of Page.getText() has been ported to C which has greatly improved their performance. This improvement is mostly noticeable with text-oriented documents, where they now should execute almost two times faster.

  • Fixed issue #369 (“mupdf: cmsCreateTransform failed”) by removing ICC colorspace support.

  • Changed Page.getText() to accept additional keywords “blocks” and “words”. These will deliver the results of Page.getTextBlocks() and Page.getTextWords(), respectively. So all text extraction methods are now available via a uniform API. Correspondingly, there are now new methods TextPage.extractBLOCKS() and TextPage.extractWords().

  • Changed Page.getText() to default bit indicator TEXT_INHIBIT_SPACES to off. Insertion of additional spaces is not suppressed by default.

Changes in Version 1.16.2

  • Changed text extraction methods of Page to allow detail control of the amount of extracted data.

  • Added planish_line() which maps a given line (defined as a pair of points) to the x-axis.

  • Fixed an issue (w/o Github number) which brought down the interpreter when encountering certain non-UTF-8 encodable characters while using Page.getText() with te “dict” option.

  • Fixed issue #362 (“Memory Leak with getText(‘rawDICT’)”).

Changes in Version 1.16.1

  • Added property Quad.is_convex which checks whether a line is contained in the quad if it connects two points of it.

  • Changed Document.insert_pdf() to now allow dropping or including links and annotations independently during the copy. Fixes issue #352 (“Corrupt PDF data and …”), which seemed to intermittently occur when using the method for some problematic PDF files.

  • Fixed a bug which, in matrix division using the syntax “m1/m2”, caused matrix “m1” to be replaced by the result instead of delivering a new matrix.

  • Fixed issue #354 (“SyntaxWarning with Python 3.8”). We now always use “==” for literals (instead of the “is” Python keyword).

  • Fixed issue #353 (“mupdf version check”), to no longer refuse the import when there are only patch level deviations from MuPDF.

Changes in Version 1.16.0

This major new version of MuPDF comes with several nice new or changed features. Some of them imply programming API changes, however. This is a synopsis of what has changed:

  • PDF document encryption and decryption is now fully supported. This includes setting permissions, passwords (user and owner passwords) and the desired encryption method.

  • In response to the new encryption features, PyMuPDF returns an integer (ie. a combination of bits) for document permissions, and no longer a dictionary.

  • Redirection of MuPDF errors and warnings is now natively supported. PyMuPDF redirects error messages from MuPDF to sys.stderr and no longer buffers them. Warnings continue to be buffered and will not be displayed. Functions exist to access and reset the warnings buffer.

  • Annotations are now only supported for PDF.

  • Annotations and widgets (form fields) are now separate object chains on a page (although widgets technically still are PDF annotations). This means, that you will never encounter widgets when using Page.firstAnnot or Annot.next(). You must use Page.firstWidget and Widget.next() to access form fields.

  • As part of MuPDF’s changes regarding widgets, only the following four fonts are supported, when adding or changing form fields: Courier, Helvetica, Times-Roman and ZapfDingBats.

List of change details:

  • Added Document.can_save_incrementally() which checks conditions that are preventing use of option incremental=True of Document.save().

  • Added Page.firstWidget which points to the first field on a page.

  • Added Page.getImageBbox() which returns the rectangle occupied by an image shown on the page.

  • Added Annot.setName() which lets you change the (icon) name field.

  • Added outputting the text color in Page.getText(): the “dict”, “rawdict” and “xml” options now also show the color in sRGB format.

  • Changed Document.permissions to now contain an integer of bool indicators – was a dictionary before.

  • Changed Document.save(), Document.write(), which now fully support password-based decryption and encryption of PDF files.

  • Changed the names of all Python constants related to annotations and widgets. Please make sure to consult the Constants and Enumerations chapter if your script is dealing with these two classes. This decision goes back to the dropped support for non-PDF annotations. The old names (starting with “ANNOT_*” or “WIDGET_*”) will be available as deprecated synonyms.

  • Changed font support for widgets: only Cour (Courier), Helv (Helvetica, default), TiRo (Times-Roman) and ZaDb (ZapfDingBats) are accepted when adding or changing form fields. Only the plain versions are possible – not their italic or bold variations. Reading widgets, however will show its original font.

  • Changed the name of the warnings buffer to Tools.mupdf_warnings() and the function to empty this buffer is now called Tools.reset_mupdf_warnings().

  • Changed Page.getPixmap(), Document.get_page_pixmap(): a new bool argument annots can now be used to suppress the rendering of annotations on the page.

  • Changed Page.add_file_annot() and Page.add_text_annot() to enable setting an icon.

  • Removed widget-related methods and attributes from the Annot object.

  • Removed Document attributes openErrCode, openErrMsg, and Tools attributes / methods stderr, reset_stderr, stdout, and reset_stdout.

  • Removed thirdparty zlib dependency in PyMuPDF: there are now compression functions available in MuPDF. Source installers of PyMuPDF may now omit this extra installation step.

No version published for MuPDF v1.15.0

Changes in Version 1.14.20 / 1.14.21

  • Changed text marker annotations to support multiple rectangles / quadrilaterals. This fixes issue #341 (“Question : How to addhighlight so that a string spread across more than a line is covered by one highlight?”) and similar (#285).

  • Fixed issue #331 (“Importing PyMuPDF changes warning filtering behaviour globally”).

Changes in Version 1.14.19

  • Fixed issue #319 (“InsertText function error when use custom font”).

  • Added new method Document.get_sigflags() which returns information on whether a PDF is signed. Resolves issue #326 (“How to detect signature in a form pdf?”).

Changes in Version 1.14.17

  • Added Document.fullcopyPage() to make full page copies within a PDF (not just copied references as Document.copyPage() does).

  • Changed Page.getPixmap(), Document.get_page_pixmap() now use alpha=False as default.

  • Changed text extraction: the span dictionary now (again) contains its rectangle under the bbox key.

  • Changed Document.movePage() and Document.copyPage() to use direct functions instead of wrapping Document.select() – similar to Document.delete_page() in v1.14.16.

Changes in Version 1.14.16

  • Changed Document methods around PDF /EmbeddedFiles to no longer use MuPDF’s “portfolio” functions. That support will be dropped in MuPDF v1.15 – therefore another solution was required.

  • Changed Document.embfile_Count() to be a function (was an attribute).

  • Added new method Document.embfile_Names() which returns a list of names of embedded files.

  • Changed Document.delete_page() and Document.delete_pages() to internally no longer use Document.select(), but instead use functions to perform the deletion directly. As it has turned out, the Document.select() method yields invalid outline trees (tables of content) for very complex PDFs and sophisticated use of annotations.

Changes in Version 1.14.15

  • Fixed issues #301 (“Line cap and Line join”), #300 (“How to draw a shape without outlines”) and #298 (“utils.updateRect exception”). These bugs pertain to drawing shapes with PyMuPDF. Drawing shapes without any border is fully supported. Line cap styles and line line join style are now differentiated and support all possible PDF values (0, 1, 2) instead of just being a bool. The previous parameter roundCap is deprecated in favor of lineCap and lineJoin and will be deleted in the next release.

  • Fixed issue #290 (“Memory Leak with getText(‘rawDICT’)”). This bug caused memory not being (completely) freed after invoking the “dict”, “rawdict” and “json” versions of Page.getText().

Changes in Version 1.14.14

  • Added new low-level function ImageProperties() to determine a number of characteristics for an image.

  • Added new low-level function Document.is_stream(), which checks whether an object is of stream type.

  • Changed low-level functions Document._getXrefString() and Document._getTrailerString() now by default return object definitions in a formatted form which makes parsing easy.

Changes in Version 1.14.13

  • Changed methods working with binary input: while ever supporting bytes and bytearray objects, they now also accept io.BytesIO input, using their getvalue() method. This pertains to document creation, embedded files, FileAttachment annotations, pixmap creation and others. Fixes issue #274 (“Segfault when using BytesIO as a stream for insertImage”).

  • Fixed issue #278 (“Is insertImage(keep_proportion=True) broken?”). Images are now correctly presented when keeping aspect ratio.

Changes in Version 1.14.12

  • Changed the draw methods of Page and Shape to support not only RGB, but also GRAY and CMYK colorspaces. This solves issue #270 (“Is there a way to use CMYK color to draw shapes?”). This change also applies to text insertion methods of Shape, resp. Page.

  • Fixed issue #269 (“AttributeError in Document.insert_page()”), which occurred when using Document.insert_page() with text insertion.

Changes in Version 1.14.11

  • Changed Page.show_pdf_page() to always position the source rectangle centered in the target. This method now also supports rotation by arbitrary angles. The argument reuse_xref has been deprecated: prevention of duplicates is now handled internally.

  • Changed Page.insertImage() to support rotated display of the image and keeping the aspect ratio. Only rotations by multiples of 90 degrees are supported here.

  • Fixed issue #265 (“TypeError: insertText() got an unexpected keyword argument ‘idx’”). This issue only occurred when using Document.insert_page() with also inserting text.

Changes in Version 1.14.10

  • Changed Page.show_pdf_page() to support rotation of the source rectangle. Fixes #261 (“Cannot rotate insterted pages”).

  • Fixed a bug in Page.insertImage() which prevented insertion of multiple images provided as streams.

Changes in Version 1.14.9

  • Added new low-level method Document._getTrailerString(), which returns the trailer object of a PDF. This is much like Document._getXrefString() except that the PDF trailer has no / needs no xref to identify it.

  • Added new parameters for text insertion methods. You can now set stroke and fill colors of glyphs (text characters) independently, as well as the thickness of the glyph border. A new parameter render_mode controls the use of these colors, and whether the text should be visible at all.

  • Fixed issue #258 (“Copying image streams to new PDF without size increase”): For JPX images embedded in a PDF, Document.extractImage() will now return them in their original format. Previously, the MuPDF base library was used, which returns them in PNG format (entailing a massive size increase).

  • Fixed issue #259 (“Morphing text to fit inside rect”). Clarified use of get_text_length() and removed extra line breaks for long words.

Changes in Version 1.14.8

  • Added Pixmap.set_rect() to change the pixel values in a rectangle. This is also an alternative to setting the color of a complete pixmap (Pixmap.clear_with()).

  • Fixed an image extraction issue with JBIG2 (monochrome) encoded PDF images. The issue occurred in Page.getText() (parameters “dict” and “rawdict”) and in Document.extractImage() methods.

  • Fixed an issue with not correctly clearing a non-alpha Pixmap (Pixmap.clear_with()).

  • Fixed an issue with not correctly inverting colors of a non-alpha Pixmap (Pixmap.invert_irect()).

Changes in Version 1.14.7

Changes in Version 1.14.5

  • Changed: Shape methods now strictly use the transformation matrix of the Page – instead of “manually” calculating locations.

  • Added method Pixmap.pixel() which returns the pixel value (a list) for given pixel coordinates.

  • Added method Pixmap.tobytes() which returns a bytes object representing the pixmap in a variety of formats. Previously, this could be done for PNG outputs only (Pixmap.tobytes()).

  • Changed: output of methods Pixmap.save() and (the new) Pixmap.tobytes() may now also be PSD (Adobe Photoshop Document).

  • Added method Shape.drawQuad() which draws a Quad. This actually is a shorthand for a Shape.drawPolyline() with the edges of the quad.

  • Changed method Shape.drawOval(): the argument can now be either a rectangle (rect_like) or a quadrilateral (quad_like).

Changes in Version 1.14.4

  • Fixes issue #239 “Annotation coordinate consistency”.

Changes in Version 1.14.3

This patch version contains minor bug fixes and CJK font output support.

  • Added support for the four CJK fonts as PyMuPDF generated text output. This pertains to methods Page.insertFont(), Shape.insertText(), Shape.insertTextbox(), and corresponding Page methods. The new fonts are available under “reserved” fontnames “china-t” (traditional Chinese), “china-s” (simplified Chinese), “japan” (Japanese), and “korea” (Korean).

  • Added full support for the built-in fonts ‘Symbol’ and ‘Zapfdingbats’.

  • Changed: The 14 standard fonts can now each be referenced by a 4-letter abbreviation.

Changes in Version 1.14.1

This patch version contains minor performance improvements.

  • Added support for Document filenames given as pathlib object by using the Python str() function.

Changes in Version 1.14.0

To support MuPDF v1.14.0, massive changes were required in PyMuPDF – most of them purely technical, with little visibility to developers. But there are also quite a lot of interesting new and improved features. Following are the details:

  • Added “ink” annotation.

  • Added “rubber stamp” annotation.

  • Added “squiggly” text marker annotation.

  • Added new class Quad (quadrilateral or tetragon) – which represents a general four-sided shape in the plane. The special subtype of rectangular, non-empty tetragons is used in text marker annotations and as returned objects in text search methods.

  • Added a new option “decrypt” to Document.save() and Document.write(). Now you can keep encryption when saving a password protected PDF.

  • Added suppression and redirection of unsolicited messages issued by the underlying C-library MuPDF. Consult 诊断 for details.

  • Changed: Changes to annotations now always require Annot.update() to become effective.

  • Changed free text annotations to support the full Latin character set and range of appearance options.

  • Changed text searching, Page.searchFor(), to optionally return Quad instead Rect objects surrounding each search hit.

  • Changed plain text output: we now add a n to each line if it does not itself end with this character.

  • Fixed issue 211 (“Something wrong in the doc”).

  • Fixed issue 213 (“Rewritten outline is displayed only by mupdf-based applications”).

  • Fixed issue 214 (“PDF decryption GONE!”).

  • Fixed issue 215 (“Formatting of links added with pyMuPDF”).

  • Fixed issue 217 (“extraction through json is failing for my pdf”).

Behind the curtain, we have changed the implementation of geometry objects: they now purely exist in Python and no longer have “shadow” twins on the C-level (in MuPDF). This has improved processing speed in that area by more than a factor of two.

Because of the same reason, most methods involving geometry parameters now also accept the corresponding Python sequence. For example, in method “page.show_pdf_page(rect, …)” parameter rect may now be any rect_like sequence.

We also invested considerable effort to further extend and improve the FAQ chapter.

Changes in Version 1.13.19

This version contains some technical / performance improvements and bug fixes.

  • Changed memory management: for Python 3 builds, Python memory management is exclusively used across all C-level code (i.e. no more native malloc() in MuPDF code or PyMuPDF interface code). This leads to improved memory usage profiles and also some runtime improvements: we have seen > 2% shorter runtimes for text extractions and pixmap creations (on Windows machines only to date).

  • Fixed an error occurring in Python 2.7, which crashed the interpreter when using TextPage.extractRAWDICT() (= Page.getText(“rawdict”)).

  • Fixed an error occurring in Python 2.7, when creating link destinations.

  • Extended the FAQ chapter with more examples.

Changes in Version 1.13.18

Changes in Version 1.13.17

  • Fixed an error that intermittently caused an exception in Page.show_pdf_page(), when pages from many different source PDFs were shown.

  • Changed method Document.extractImage() to now return more meta information about the extracted imgage. Also, its performance has been greatly improved. Several demo scripts have been changed to make use of this method.

  • Changed method Document._getXrefStream() to now return None if the object is no stream and no longer raise an exception if otherwise.

  • Added method Document._deleteObject() which deletes a PDF object identified by its xref. Only to be used by the experienced PDF expert.

  • Added a method paper_rect() which returns a Rect for a supplied paper format string. Example: fitz.paper_rect(“letter”) = fitz.Rect(0.0, 0.0, 612.0, 792.0).

  • Added a FAQ chapter to this document.

Changes in Version 1.13.16

  • Added support for correctly setting transparency (opacity) for certain annotation types.

  • Added a tool property (Tools.fitz_config) showing the configuration of this PyMuPDF version.

  • Fixed issue #193 (‘insertText(overlay=False) gives “cannot resize a buffer with shared storage” error’) by avoiding read-only buffers.

Changes in Version 1.13.15

  • Fixed issue #189 (“cannot find builtin CJK font”), so we are supporting builtin CJK fonts now (CJK = China, Japan, Korea). This should lead to correctly generated pixmaps for documents using these languages. This change has consequences for our binary file size: it will now range between 8 and 10 MB, depending on the OS.

  • Fixed issue #191 (“Jupyter notebook kernel dies after ca. 40 pages”), which occurred when modifying the contents of an annotation.

Changes in Version 1.13.14

This patch version contains several improvements, mainly for annotations.

  • Changed Annot.lineEnds is now a list of two integers representing the line end symbols. Previously was a dict of strings.

  • Added support of line end symbols for applicable annotations. PyMuPDF now can generate these annotations including the line end symbols.

  • Added Annot.setLineEnds() adds line end symbols to applicable annotation types (‘Line’, ‘PolyLine’, ‘Polygon’).

  • Changed technical implementation of Page.insertImage() and Page.show_pdf_page(): they now create there own contents objects, thereby avoiding changes of potentially large streams with consequential compression / decompression efforts and high change volumes with incremental updates.

Changes in Version 1.13.13

This patch version contains several improvements for embedded files and file attachment annotations.

  • Added Document.embfile_Upd() which allows changing file content and metadata of an embedded file. It supersedes the old method Document.embfile_SetInfo() (which will be deleted in a future version). Content is automatically compressed and metadata may be unicode.

  • Changed Document.embfile_Add() to now automatically compress file content. Accompanying metadata can now be unicode (had to be ASCII in the past).

  • Changed Document.embfile_Del() to now automatically delete all entries having the supplied identifying name. The return code is now an integer count of the removed entries (was None previously).

  • Changed embedded file methods to now also accept or show the PDF unicode filename as additional parameter ufilename.

  • Added Page.add_file_annot() which adds a new file attachment annotation.

  • Changed Annot.fileUpd() (file attachment annot) to now also accept the PDF unicode ufilename parameter. The description parameter desc correctly works with unicode. Furthermore, all parameters are optional, so metadata may be changed without also replacing the file content.

  • Changed Annot.fileInfo() (file attachment annot) to now also show the PDF unicode filename as parameter ufilename.

  • Fixed issue #180 (“page.getText(output=’dict’) return invalid bbox”) to now also work for vertical text.

  • Fixed issue #185 (“Can’t render the annotations created by PyMuPDF”). The issue’s cause was the minimalistic MuPDF approach when creating annotations. Several annotation types have no /AP (“appearance”) object when created by MuPDF functions. MuPDF, SumatraPDF and hence also PyMuPDF cannot render annotations without such an object. This fix now ensures, that an appearance object is always created together with the annotation itself. We still do not support line end styles.

Changes in Version 1.13.12

  • Fixed issue #180 (“page.getText(output=’dict’) return invalid bbox”). Note that this is a circumvention of an MuPDF error, which generates zero-height character rectangles in some cases. When this happens, this fix ensures a bbox height of at least fontsize.

  • Changed for ListBox and ComboBox widgets, the attribute list of selectable values has been renamed to Widget.choice_values.

  • Changed when adding widgets, any missing of the PDF Base 14 字体 is automatically added to the PDF. Widget text fonts can now also be chosen from existing widget fonts. Any specified field values are now honored and lead to a field with a preset value.

  • Added Annot.updateWidget() which allows changing existing form fields – including the field value.

Changes in Version 1.13.11

While the preceeding patch subversions only contained various fixes, this version again introduces major new features:

  • Added basic support for PDF widget annotations. You can now add PDF form fields of types Text, CheckBox, ListBox and ComboBox. Where necessary, the PDF is tranformed to a Form PDF with the first added widget.

  • Fixed issues #176 (“wrong file embedding”), #177 (“segment fault when invoking page.getText()”)and #179 (“Segmentation fault using page.getLinks() on encrypted PDF”).

Changes in Version 1.13.7

  • Added support of variable page sizes for reflowable documents (e-books, HTML, etc.): new parameters rect and fontsize in Document creation (open), and as a separate method Document.layout().

  • Added Annot creation of many annotations types: sticky notes, free text, circle, rectangle, line, polygon, polyline and text markers.

  • Added support of annotation transparency (Annot.opacity, Annot.setOpacity()).

  • Changed Annot.vertices: point coordinates are now grouped as pairs of floats (no longer as separate floats).

  • Changed annotation colors dictionary: the two keys are now named “stroke” (formerly “common”) and “fill”.

  • Added Document.isDirty which is True if a PDF has been changed in this session. Reset to False on each Document.save() or Document.write().

Changes in Version 1.13.6

  • Fix #173: for memory-resident documents, ensure the stream object will not be garbage-collected by Python before document is closed.

Changes in Version 1.13.5

  • New low-level method Page._setContents() defines an object given by its xref to serve as the contents object.

  • Changed and extended PDF form field support: the attribute widget_text has been renamed to Annot.widget_value. Values of all form field types (except signatures) are now supported. A new attribute Annot.widget_choices contains the selectable values of listboxes and comboboxes. All these attributes now contain None if no value is present.

Changes in Version 1.13.4

  • Document.convertToPDF() now supports page ranges, reverted page sequences and page rotation. If the document already is a PDF, an exception is raised.

  • Fixed a bug (introduced with v1.13.0) that prevented Page.insertImage() for transparent images.

Changes in Version 1.13.3

Introduces a way to convert any MuPDF supported document to a PDF. If you ever wanted PDF versions of your XPS, EPUB, CBZ or FB2 files – here is a way to do this.

  • Document.convertToPDF() returns a Python bytes object in PDF format. Can be opened like normal in PyMuPDF, or be written to disk with the “.pdf” extension.

Changes in Version 1.13.2

The major enhancement is PDF form field support. Form fields are annotations of type (19, ‘Widget’). There is a new document method to check whether a PDF is a form. The Annot class has new properties describing field details.

  • Document.is_form_pdf is true if object type /AcroForm and at least one form field exists.

  • Annot.widget_type, Annot.widget_text and Annot.widget_name contain the details of a form field (i.e. a “Widget” annotation).

Changes in Version 1.13.1

  • TextPage.extractDICT() is a new method to extract the contents of a document page (text and images). All document types are supported as with the other TextPage extract*() methods. The returned object is a dictionary of nested lists and other dictionaries, and exactly equal to the JSON-deserialization of the old TextPage.extractJSON(). The difference is that the result is created directly – no JSON module is used. Because the user needs no JSON module to interpet the information, it should be easier to use, and also have a better performance, because it contains images in their original binary format – they need not be base64-decoded.

  • Page.getText() correspondingly supports the new parameter value “dict” to invoke the above method.

  • TextPage.extractJSON() (resp. Page.getText(“json”)) is still supported for convenience, but its use is expected to decline.

Changes in Version 1.13.0

This version is based on MuPDF v1.13.0. This release is “primarily a bug fix release”.

In PyMuPDF, we are also doing some bug fixes while introducing minor enhancements. There only very minimal changes to the user’s API.

  • Document construction is more flexible: the new filetype parameter allows setting the document type. If specified, any extension in the filename will be ignored. More completely addresses issue #156. As part of this, the documentation has been reworked.

  • Changes to Pixmap constructors:

    • Colorspace conversion no longer allows dropping the alpha channel: source and target alpha will now always be the same. We have seen exceptions and even interpreter crashes when using alpha = 0.

    • As a replacement, the simple pixmap copy lets you choose the target alpha.

  • Document.save() again offers the full garbage collection range 0 thru 4. Because of a bug in xref maintenance, we had to temporarily enforce garbage > 1. Finally resolves issue #148.

  • Document.save() now offers to “prettify” PDF source via an additional argument.

  • Page.insertImage() has the additional stream -parameter, specifying a memory area holding an image.

  • Issue with garbled PNGs on Linux systems has been resolved (“Problem writing PNG” #133).

Changes in Version 1.12.4

This is an extension of 1.12.3.

  • Fix of issue #147: methods Document.getPageFontlist() and Document.getPageImagelist() now also show fonts and images contained in resources nested via “Form XObjects”.

  • Temporary fix of issue #148: Saving to new PDF files will now automatically use garbage = 2 if a lower value is given. Final fix is to be expected with MuPDF’s next version. At that point we will remove this circumvention.

  • Preventive fix of illegally using stencil / image mask pixmaps in some methods.

  • Method Document.getPageFontlist() now includes the encoding name for each font in the list.

  • Method Document.getPageImagelist() now includes the decode method name for each image in the list.

Changes in Version 1.12.3

This is an extension of 1.12.2.

Changes in Version 1.12.2

This is an extension of 1.12.1.

  • Method Page.show_pdf_page() now accepts the new clip argument. This specifies an area of the source page to which the display should be restricted.

  • New Page.CropBox and Page.MediaBox have been included for convenience.

Changes in Version 1.12.1

This is an extension of version 1.12.0.

  • New method Page.show_pdf_page() displays another’s PDF page. This is a vector image and therefore remains precise across zooming. Both involved documents must be PDF.

  • New method Page.getSVGimage() creates an SVG image from the page. In contrast to the raster image of a pixmap, this is a vector image format. The return is a unicode text string, which can be saved in a .svg file.

  • Method Page.getTextBlocks() now accepts an additional bool parameter “images”. If set to true (default is false), image blocks (metadata only) are included in the produced list and thus allow detecting areas with rendered images.

  • Minor bug fixes.

  • “text” result of Page.getText() concatenates all lines within a block using a single space character. MuPDF’s original uses “\n” instead, producing a rather ragged output.

  • New properties of Page objects Page.MediaBoxSize and Page.CropBoxPosition provide more information about a page’s dimensions. For non-PDF files (and for most PDF files, too) these will be equal to Page.rect.bottom_right, resp. Page.rect.top_left. For example, class Shape makes use of them to correctly position its items.

Changes in Version 1.12.0

This version is based on and requires MuPDF v1.12.0. The new MuPDF version contains quite a number of changes – most of them around text extraction. Some of the changes impact the programmer’s API.

  • Outline.saveText() and Outline.saveXML() have been deleted without replacement. You probably haven’t used them much anyway. But if you are looking for a replacement: the output of Document.get_toc() can easily be used to produce something equivalent.

  • Class TextSheet does no longer exist.

  • Text “spans” (one of the hierarchy levels of TextPage) no longer contain positioning information (i.e. no “bbox” key). Instead, spans now provide the font information for its text. This impacts our JSON output variant.

  • HTML output has improved very much: it now creates valid documents which can be displayed by browsers to produce a similar view as the original document.

  • There is a new output format XHTML, which provides text and images in a browser-readable format. The difference to HTML output is, that no effort is made to reproduce the original layout.

  • All output formats of Page.getText() now support creating complete, valid documents, by wrapping them with appropriate header and trailer information. If you are interested in using the HTML output, please make sure to read 控制 HTML 输出的质量.

  • To support finding text positions, we have added special methods that don’t need detours like TextPage.extractJSON() or TextPage.extractXML(): use Page.getTextBlocks() or resp. Page.getTextWords() to create lists of text blocks or resp. words, which are accompanied by their rectangles. This should be much faster than the standard text extraction methods and also avoids using additional packages for interpreting their output.

Changes in Version 1.11.2

This is an extension of v1.11.1.

  • New Page.insertFont() creates a PDF /Font object and returns its object number.

  • New Document.extractFont() extracts the content of an embedded font given its object number.

  • Methods FontList(…) items no longer contain the PDF generation number. This value never had any significance. Instead, the font file extension is included (e.g. “pfa” for a “PostScript Font for ASCII”), which is more valuable information.

  • Fonts other than “simple fonts” (Type1) are now also supported.

  • New options to change Pixmap size:

    • Method Pixmap.shrink() reduces the pixmap proportionally in place.

    • A new Pixmap copy constructor allows scaling via setting target width and height.

Changes in Version 1.11.1

This is an extension of v1.11.0.

  • New class Shape. It facilitates and extends the creation of image shapes on PDF pages. It contains multiple methods for creating elementary shapes like lines, rectangles or circles, which can be combined into more complex ones and be given common properties like line width or colors. Combined shapes are handled as a unit and e.g. be “morphed” together. The class can accumulate multiple complex shapes and put them all in the page’s foreground or background – thus also reducing the number of updates to the page’s contents object.

  • All Page draw methods now use the new Shape class.

  • Text insertion methods insertText() and insertTextBox() now support morphing in addition to text rotation. They have become part of the Shape class and thus allow text to be freely combined with graphics.

  • A new Pixmap constructor allows creating pixmap copies with an added alpha channel. A new method also allows directly manipulating alpha values.

  • Binary algebraic operations with geometry objects (matrices, rectangles and points) now generally also support lists or tuples as the second operand. You can add a tuple (x, y) of numbers to a Point. In this context, such sequences are called “point_like” (resp. matrix_like, rect_like).

  • Geometry objects now fully support in-place operators. For example, p /= m replaces point p with p * 1/m for a number, or p * ~m for a matrix_like object m. Similarly, if r is a rectangle, then r |= (3, 4) is the new rectangle that also includes fitz.Point(3, 4), and r &= (1, 2, 3, 4) is its intersection with fitz.Rect(1, 2, 3, 4).

Changes in Version 1.11.0

基于 MuPDF v1.11 的版本

尽管 MuPDF 将该版本声明为主要是修复错误的版本，但它确实包含了一个重要的新特性：**支持嵌入文件**（亦称为“作品集”或“集合”）。我们扩展了 PyMuPDF 的功能，使其在这一方面的能力略超 mutool 实用程序，具体如下：

  • Document 类新增多个方法和一个新属性，用于处理嵌入文件：

    • embfile_Info()：返回嵌入文件列表中某个条目的元数据信息。相比 mutool，它提供了更详细的信息，包括用于嵌入文件的所有相关数据（不仅仅是文件名）。

    • embfile_Get()：将嵌入文件的内容解压到 bytes 缓冲区中。

    • embfile_Add(…)：向 PDF 作品集中插入新文件。与 mutool 不同，我们 限制 该方法只能添加 新名称 的文件（不允许重复文件名）。

    • embfile_Del(…)：从作品集中删除某个嵌入文件（此功能未在 MuPDF 中提供）。

    • embfile_SetInfo()：修改嵌入文件的名称或描述信息。

    • embfile_Count：包含嵌入文件的数量。

  • 我们进行了多项优化，以简化几何对象的处理。这些更改与新 MuPDF 版本无关，其中大多数在 PyMuPDF v1.10.0 中也已实现。这些优化包括：

    • 通过新的属性以名称识别矩形的角，例如：Rect.bottom_right。

    • 新增多个方法来处理集合论相关的问题，例如： - Rect.contains(x)：判断矩形是否包含点 x。 - IRect.intersects(x)：判断矩形是否与 x 相交。

    • 加强了对 Pythonic 语言风格的支持。例如:

      if x in rect: # 现在等价于 rect.contains(x)

  • Rect 章节新增了关于**空矩形**和**无限矩形**的背景介绍，并改进了相关的处理方式，以提高一致性。

  • 我们开始支持 生成 PDF 内容，包括以下方法：

    • Document.insert_page()：向 PDF 中添加一个新页面，可选地包含一些文本。

    • Page.insertImage()：在 PDF 页面上插入一张新图片。

    • Page.insertText()：在现有页面上添加新文本。

  • 现在可以提取和修改**文件附件**注释的内容及其名称。

Changes in Version 1.10.0

MuPDF v1.10 影响

MuPDF 版本 1.10 对我们的绑定产生了重大影响。其中一些变更也影响了 API —— 换句话说，作为 PyMuPDF 用户的您 可能需要注意这些更改。

  • 链接目标信息已减少。linkDest 类的多个属性不再包含有价值的信息。事实上，该类已从 MuPDF 库中删除，PyMuPDF 仅保留它以确保与现有代码兼容。

  • 为了最小化内存占用，MuPDF v1.10 进行了多项优化：

    • 新增的 config.h 文件允许在 C 代码中禁用不需要的功能。利用此功能，我们将二进制文件 _fitz.o / _fitz.pyd 的大小从 9 MB 缩小到 4.5 MB。使用 UPX 压缩后，文件大小甚至可降至 2.3 MB。

    • Pixmap 现在的 alpha（透明度）通道是可选的。默认情况下，alpha=False，这大大减少了 Pixmap 大小（CMYK 约 20%，RGB 约 25%，GRAY 约 50%）。因此，许多 Pixmap 构造函数现在接受一个 alpha 布尔参数，以控制是否包含透明通道。而其他构造函数（如文件和图像输入）创建的 Pixmap 默认不包含 alpha 通道。需要注意的是，Pixmap 的保存方法不再支持 savealpha 选项：如果 alpha 通道存在，则始终会保存它。为尽量减少代码破坏，我们保留了该参数，但它将被忽略。

  • **DisplayList 和 TextPage 现在需要提供 mediabox**（即 page.bound() 矩形）。这一信息无法从其他来源推导，因此需要修改源代码。我们预计，大多数用户不会直接使用这些较低级的类，因此该更改的影响应较小。

—

与 1.9.3 版本相比的其他更改

  • 新增 Document 方法 write()，可将打开的 PDF 写入内存（而 save() 方法则是写入文件）。

  • 现在可以缩放和移动注释（annotations），方法是修改其矩形属性。

  • 现在可以删除注释。Page 类新增方法 deleteAnnot()。

  • 现在可以修改各种注释属性，例如内容、日期、标题（作者）、边框、颜色等。

  • 方法 Document.insert_pdf() 现在也会复制源页面的注释。

  • Pages 类已删除。由于文档现在可以通过索引访问页面（如 doc[n] = doc.loadPage(n)），并且文档对象本身可用作迭代器，因此该类的维护价值过低，故被移除。

  • loadPage(n) / doc[n] 现在支持任意整数作为页面编号，只要满足 n < pageCount。例如，doc[-500] 仍然是有效的，并会加载页面 (-500) % pageCount。

  • 现在文档对象可直接作为迭代器，例如:

    for page in doc:
    ...  # 对 "page" 进行处理

    这将遍历文档中的所有页面。

  • Pixmap 方法 getSize() 已被属性 size 取代。与以前一样，Pixmap.size == len(Pixmap) 仍然成立。

  • 由于透明度（alpha）变为可选，Pixmap 和 Colorspace 类新增多个参数和属性，以支持查询相关特性。

  • Page 类新增属性 firstAnnot 和 firstLink，分别提供注释链和链接链的起始点。其中，firstLink 只是方法 loadLinks() 的一个助记符，后者仍然可用。同样，rect 是方法 bound() 的同义词，后者仍然可用。

  • 由于 Pixmap 现在可以不包含透明通道，Pixmap 的方法 samplesRGB() 和 samplesAlpha() 已被删除。

  • Rect 现在有一个属性 irect，它是方法 round() 的同义词。同样，IRect 现在有一个属性 rect，用于返回具有相同坐标但浮点数值的 Rect。

  • 文档新增方法 searchPageFor()，用于搜索文本字符串。其工作方式与 Page.searchFor() 相同，但额外接受页面编号作为参数。

Changes in Version 1.9.3

  • 此版本同样基于 MuPDF v1.9a。与 1.9.2 版本相比，主要变更如下：

    • 主要增强：现在支持注释（annotations），方式类似于链接（links）。可以显示注释（作为 pixmap），并访问其属性。

    • 除了文档的 select() 方法外，现在可以使用一些更简单的方法来操作 PDF：

      • copyPage() 复制文档中的某个页面。

      • movePage() 功能类似，但会删除原始页面。

      • delete_page() 删除指定页面。

      • delete_pages() 删除一个页面范围。

    • rotation 或 setRotation() 分别用于访问或更改 PDF 页面旋转角度。

    • 之前已存在但未记录的功能：IRect、Rect、Point 和 Matrix 现在支持 len() 方法，并且其坐标属性可通过索引访问，例如 IRect.x1 == IRect[2]。

    • 为方便起见，现在文档支持简单的索引访问：doc.loadPage(n) == doc[n]。索引范围可以是 -pageCount < n < pageCount，例如 doc[-1] 代表文档的最后一页。

Changes in Version 1.9.2

  • 此版本同样基于 MuPDF v1.9a。与 1.9.1 版本相比，主要变更如下：

    • fitz.open()*（无参数）现在会创建一个新的空 **PDF* 文档，即如果随后保存，则必须指定 .pdf 扩展名。

    • Document 现在接受以下所有格式（Document 和 open 是同义词）：

      • open()，

      • open(filename)*（等效于 *open(filename, None)），

      • open(filetype, area)*（等效于 *open(filetype, stream=area)）。

      其中，stream 的类型可以是 bytes 或 bytearray。因此，例如可以直接使用 `area = open(“file.pdf”, “rb”).read()`（无需先转换为 bytearray）。

    • 新方法 *Document.insert_pdf()*（仅适用于 PDF）可以从另一个 PDF 中插入一系列页面。

    • 现在 Document 对象支持 len() 函数：len(doc) == doc.pageCount。

    • 新方法 Document.getPageImageList() 创建一个页面所使用的图像列表。

    • 新方法 Document.getPageFontList() 创建一个页面所引用的字体列表。

    • 新的 Pixmap 构造函数 fitz.Pixmap(doc, xref) 允许基于已打开的 PDF 文档和图像的 xref 号创建 Pixmap。

    • 新的 Pixmap 构造函数 fitz.Pixmap(cspace, spix) 允许基于已有 Pixmap spix，将其颜色空间转换为 cspace，支持所有颜色空间转换组合。

    • Pixmap 构造函数 fitz.Pixmap(colorspace, width, height, samples) 现在允许 samples 既可以是 bytes，也可以是 bytearray。

Changes in Version 1.9.1

  • 此版本的 PyMuPDF 基于 MuPDF 库源代码版本 1.9a，该版本发布于 2016 年 4 月 21 日。

    请访问 MuPDF 的官方网站查看其中包含的更改和增强功能。

  • 版本 1.9.1 相较于版本 1.8.0 的变化:

    • 新方法 get_area() 适用于 fitz.Rect 和 fitz.IRect。

    • 现在可以直接从文件创建 Pixmap，使用新的构造函数 fitz.Pixmap(filename)。

    • Pixmap 构造函数 fitz.Pixmap(image) 已相应扩展。

    • 现在可以使用所有可能的点和坐标组合来创建 fitz.Rect。

    • PyMuPDF 的所有类和方法现在都包含 __doc__ 字符串，大多数由 SWIG 自动创建。虽然 PyMuPDF 的文档更加详细，但此功能应该有助于在支持 Python 的 IDE 中进行编程。

    • 新的文档方法 getPermits() 返回与当前访问文档相关的权限（打印、编辑、注释、复制），作为 Python 字典。

    • 身份矩阵 fitz.Identity 现在是 不可变 的。

    • 新的文档方法 select(list) 会从文档中删除不在列表中的所有页面。页面也可以被复制和重新排列。

    • 演示和示例集合中的各种改进和新成员。最显著的是：PDF_display 现在支持使用鼠标滚轮滚动，此外还有一个新的示例程序 wxTableExtract，可以图形化识别和提取文档中的表格数据。

    • fitz.open() 现在是 fitz.Document() 的别名。

    • 新的 Pixmap 方法 tobytes() 将返回一个以 PNG 图像格式表示的 pixmap 的 bytearray。

    • 新的 Pixmap 方法 samplesRGB() 提供一个 samples 版本，去除了 alpha 字节（仅限 RGB 颜色空间）。

    • 新的 Pixmap 方法 samplesAlpha() 提供 samples 区域的 alpha 字节。

    • 新的迭代器 fitz.Pages(doc) 用于遍历文档的页面集合。

    • 新的矩阵方法 *invert()*（计算反转矩阵）、*concat()*（计算矩阵乘积）、*pretranslate()*（执行平移操作）。

    • 新的 IRect 方法 *intersect()*（与另一个矩形的交集）、*translate()*（执行平移操作）。

    • 新的 Rect 方法 *intersect()*（与另一个矩形的交集）、*transform()*（使用矩阵进行转换）、*include_point()*（扩大矩形以包含一个点）、*include_rect()*（扩大矩形以包含另一个矩形）。

    • 文档了 *Point.transform()*（使用矩阵转换点）。

    • Matrix、IRect、Rect 和 Point 类现在支持紧凑的代数表达式来操作这些对象。

    • 现在可以使用调用模式 doc.save(doc.name, incremental=True) 实现增量保存更改。

    • 现在可以使用文档方法 set_metadata() 删除、设置或更改 PDF 的元数据。支持增量保存。

    • 现在可以使用文档方法 set_toc(list) 删除、设置或更改 PDF 的书签（或目录）。支持增量保存。

本软件按原样提供，不作任何明示或暗示担保。本软件根据许可分发，除非根据该许可条款明确授权，否则不得复制、修改或分发。请参阅 artifex.com 上的许可信息，或联系 Artifex Software Inc., 39 Mesa Street, Suite 108A, San Francisco CA 94129, United States 了解更多信息。

This documentation covers all versions up to 1.25.4.